A method for controlling parameters of stacked batteries

Abstract

This application discloses a stacked battery parameter control method, which relates to the field of stacked battery control. The method includes: collecting the battery temperature, voltage, and current parameters of each single cell in the stacked battery; using the battery temperature, voltage, current, and first-order current differential and second-order current differential as inputs, and the corresponding single cell health status as output, to train a single cell health status assessment model; performing a health status assessment on the collected single cell parameters to obtain a health index for each single cell, and calculating the health index of the current stacked battery; obtaining the stacked battery health index, current charge state parameters, and current discharge state parameters, and performing linear combinations to obtain a multidimensional characteristic scalar of the stacked battery; using the multidimensional characteristic scalar as input, and utilizing a model predictive control algorithm to generate a charging control strategy for the stacked battery. This application effectively delays stacked battery aging, addressing the existing problem of stacked battery aging.

Description

Technical Field

[0001] The present application relates to the field of stacked battery control, and in particular to a stacked battery parameter control method. Background Art

[0002] Home stacked batteries still face significant challenges during use, with battery aging being a particular issue. Affected by factors such as material properties, production processes, and the operating environment, the performance parameters of individual cells in a stacked battery, such as capacity and internal resistance, degrade with increasing charge and discharge cycles. This manifests as capacity fade and power reduction, ultimately shortening battery life and reducing energy storage performance. When a stacked battery's capacity fades to 70% to 80% of its initial capacity, it is generally considered to have reached the end of its lifespan and requires replacement, resulting in higher operating costs.

[0003] Because home energy storage systems are often used in conjunction with renewable energy generation systems, fluctuations in wind and solar resources can also exacerbate the aging of stacked batteries. When wind and solar resources are insufficient, the energy storage system needs to be charged and discharged frequently, increasing the number of battery cycles. When wind and solar resources are in excess, batteries may be overcharged, accelerating battery aging. Home stacked batteries are often placed outdoors or underground, where ambient temperatures fluctuate significantly. Both high and low temperatures can adversely affect battery performance. In high-temperature environments, the deterioration of internal battery materials accelerates, exacerbating side reactions and accelerating capacity decay. In low-temperature environments, the reaction rate of active battery materials decreases, internal resistance increases, and charge and discharge efficiency decreases.

[0004] To address the aging problem of stacked household batteries, the industry has conducted extensive research focused on battery material modification, structural design optimization, and balancing management. By optimizing the ratio of positive and negative electrode materials and electrolyte formulations to improve battery material properties, and optimizing the plate and separator structures to enhance ion conductivity, battery aging can be slowed to a certain extent. Active balancing management uses a balancing circuit to achieve energy balance among individual cells in a stacked battery, reducing individual state variations and extending the battery pack's lifespan. However, improvements in battery materials and structures are limited by cost, and balancing management has been unable to fundamentally address the inconsistency issue. Summary of the Invention

[0005] In response to the stacked battery aging problem existing in the prior art, the present application provides a stacked battery parameter control method, which evaluates the health index of the stacked battery by collecting single-cell battery data and uses a model predictive control algorithm to generate an optimal stacked battery control strategy, thereby effectively delaying the aging of the stacked battery and extending the service life of the stacked battery.

[0006] The purpose of this application is achieved through the following technical solutions.

[0007] One aspect of the present application provides a stacked battery parameter control method, comprising: collecting battery temperature, voltage, and current parameters of each single cell in the stacked battery; preprocessing the collected parameters to obtain a preprocessed data set; dividing the preprocessed data set into a training set and a test set; using the battery temperature, voltage, current, and first-order current difference and second-order current difference in the training set as input and the corresponding single cell health status as output, training a support vector machine (SVM) model to obtain a single cell health status assessment model; using the trained single cell health status assessment model to perform health status assessment on the collected single cell parameters to obtain the health status of each single cell. index; obtain the health index of the current stacked battery by weighted averaging the health indexes of all single cells; obtain the total charging current, total charging voltage and cumulative charging time of the stacked battery as current charging state parameters; obtain the total discharge current, total discharge voltage and cumulative discharge time of the stacked battery as current discharge state parameters; linearly combine the obtained stacked battery health index, current charging state parameters and current discharge state parameters to obtain a multidimensional feature scalar of the stacked battery; use the multidimensional feature scalar as input and use the model predictive control algorithm to generate a charging control strategy for the stacked battery; wherein the charging control strategy includes charging current, charging voltage and charging time.

[0008] Furthermore, the collected parameters are preprocessed, including: filtering the battery temperature data using a Kalman filter algorithm; detecting outliers on the voltage data using the 3σ principle; correcting the detected outliers using a linear interpolation algorithm; smoothing the current data using a smoothing filter algorithm based on polynomial fitting, and performing first-order and second-order differences to obtain smoothed current data, as well as first-order differential current data and second-order differential current data; combining the smoothed current data, as well as the first-order differential current data and the second-order differential current data to obtain an expanded current data set; and aligning the filtered battery temperature data, the corrected voltage data, and the expanded current data according to timestamps to obtain a preprocessed data set.

[0009] Furthermore, smoothed current data, as well as first-order differential current data and second-order differential current data, are obtained, including: according to each data point ii in the current data sequence, with data point ii as the center, multiple data points before and after data point ii are selected to form a sliding window; in each sliding window, a least squares method is used to perform polynomial fitting on multiple data points in the corresponding sliding window, and the fitting order is the set filter order; the polynomial fitting value of the center point in the sliding window is used as the filtering result of the corresponding data point ii to obtain a smoothed current data sequence; according to the fitted polynomial, the first-order derivative value of each data point ii is calculated as the first-order differential current, and the first-order differential current reflects the current change rate; according to the fitted polynomial, the second-order derivative value of each data point is calculated as the second-order differential current, and the second-order differential current reflects the current change acceleration.

[0010] Furthermore, the health index of the current stacked battery is obtained, including: extracting the battery temperature, voltage, smoothed current, first-order differential current and second-order differential current data of each single battery to form the input feature vector X of the corresponding single battery i ,i=1,2,.....,N, where N is the number of cells in the stacked battery; the input feature vector X of each cell i Input the single battery health status assessment model, and use the classification hyperplane to classify the input feature vector X i Perform health status classification to obtain the health status H of the corresponding single battery at the current moment i ,i=1,2,......,N; where the health status H i Belong to {"healthy", "sub-healthy", "unhealthy}; according to the preset health status quantification rules, the health status H of each single battery is i Quantified into the corresponding health index value I i ; According to the electrical connection topology of each single cell in the stacked battery, set the weight coefficient w of the single cell i ,i=1,2,.....,N;using weight coefficient w i Health Index Value I i Perform weighted averaging to obtain the health index I of the stacked batteries at the current moment. stack : Among them, I stack is the health index of the stacked battery.

[0011] Preferably, the weight coefficient w of each single cell is set according to the electrical connection topology of each single cell in the stacked battery. i ,i=1,2,....,N, including: analyzing the electrical connection topology of the stacked battery, determining the physical position and electrical connection method of each single battery; setting the position weight coefficient α according to the physical position of the single batteryi ,i=1,2,....,N:For single cells connected in series, the single cell in the middle of the series branch is assigned a higher position weight coefficient, and the single cells at both ends of the series branch are assigned a lower position weight coefficient; for single cells connected in parallel, all single cells are assigned the same position weight coefficient; according to the electrical connection method of the single cell, the connection weight coefficient β is set i ,i=1,2,....,N:For single cells connected in series, all single cells are assigned the same connection weight coefficient; for single cells connected in parallel, the single cells in the parallel branch are assigned a lower connection weight coefficient; the position weight coefficient α of the single cell is comprehensively considered i and the connection weight coefficient β i , calculate the weight coefficient w of the single battery i :

[0012] Among them, w i Satisfy the normalization conditions For stacked batteries with complex nonlinear topology, the circuit analysis method is used to calculate the weight coefficient w of the single battery. i : Establish an equivalent circuit model of the stacked battery and calculate the equivalent internal resistance and equivalent voltage of each single cell; use the node voltage method or loop current method to analyze the current distribution and voltage sharing of the single cell; calculate the weight coefficient w of the single cell based on the current distribution and voltage sharing. i , which satisfies the normalization condition.

[0013] Furthermore, a multi-dimensional characteristic scalar of the stacked battery is obtained, including: the health index I stack , current charging state parameters and current discharging state parameters are arranged in a preset order to form a characteristic parameter set F:

[0014] F={I stack ,I chg ,U chg ,T chg ,I dis ,U dis ,T dis}, where I chg ,U chg ,T chg are the total charging current, total charging voltage and cumulative charging time respectively; I dis ,U dis ,T dis The total discharge current, total discharge voltage and cumulative discharge time are obtained respectively; each parameter in the characteristic parameter set F is normalized to obtain the normalized characteristic parameter set F norm :F norm={f1,f2,......,f7}; where f i is the normalized ith characteristic parameter, and the calculation formula is: Among them, max(F i ) and min(F i ) are the maximum and minimum values of the i-th feature parameter in the historical data set; set each normalized feature parameter f i The weight coefficient μ i , the weight coefficient satisfies the normalization condition: Using the normalized feature parameter f i and its weight coefficient μ i , through weighted linear combination, the multidimensional characteristic scalar X of the stacked battery is obtained stack :

[0015] Furthermore, the charging control strategy of the stacked battery is generated by using the model predictive control algorithm, including: obtaining the multi-dimensional characteristic scalar X of the stacked battery stack (k), where k represents the current control cycle; according to the design parameters of the stacked battery and the user's charging requirements, the objective function f(X target ), where X target is the target state vector of the stacked battery, which contains the target state of charge SOC of the stacked battery target Target health status SOH target and target temperature T target ; Based on the objective function f(X target ), construct an optimization problem, where the optimization variables in the optimization problem are the charging control sequence {U(k|k), U(k+1|k), ..., U(k+M-1|k)} for the next M control cycles; the sequential quadratic programming algorithm is used to solve the optimization problem, and the optimal charging control sequence {U*(k|k), U*(k+1|k), ..., U*(k+M-1|k)} for the next M control cycles is obtained; the first element U*(k|k) of the optimal control sequence is used as the optimal charging control quantity of the stacked battery in the current control cycle k, and it is decoded into the corresponding charging current I chg (k), charging voltage U chg (k) and charging time T chg (k), as the charging control strategy for the stacked batteries in the current control period.

[0016] Furthermore, the objective function f(X target ), the expression is as follows:

[0017] f(X target )=w1×||SOC target-SOC||+w2×||SOH target -SOH||+w3×||T target -T||, where SOC target ,SOH target ,T target They are the target state of charge, target state of health and target temperature of the stacked battery, corresponding to the target state vector X target The three components of the stacked battery are SOC, SOH and T, respectively, which correspond to the current multidimensional characteristic scalar X stack (k); w1, w2, w3 are the weighting coefficients of the three state components.

[0018] Furthermore, we construct an optimization problem, including: taking the target state vector X target To control the target, according to the current state X of the stacked battery stack (k) and the state prediction value of the next M control cycles

[0019] X stack (k+i|k), i=1,2,.....,M, construct the objective function J, which is used to evaluate the degree to which the stacked battery state deviates from the target state in the next M control cycles; the charging control sequence {U(k|k), U(k+1|k),......, U(k+M-1|k)} of the next M control cycles is used as the optimization variable, where the charging control variable U(k) of a single control cycle includes the charging current I chg (k), charging voltage U chg (k) and charging time T chg (k); the value range of the charging control variable U(k) is used as the constraint condition of the optimization problem, and the objective function J is used as the performance indicator of the optimization problem to construct the optimization problem.

[0020] Furthermore, the objective function J is expressed as follows:

[0021] Among them, SOC(k+i|k), SOH(k+i|k) and T(k+i|k) are the predicted values of state of charge, health state and temperature of the kth control cycle, respectively, and the corresponding state prediction value X stack The three components of (k+i|k); w1, w2, w3 are the weighted coefficients of state deviation; I chg (k+i|k),U chg (k+i|k),T chg (k+i|k) are the charging current, charging voltage and charging time of the k+i control cycle, corresponding to the three components of the charging control variable U(k+i|k); λ I ,λ U,λ T are weighted coefficients for the charging current, voltage, and time variations in adjacent control cycles. M represents the length of the prediction time domain, i.e., the number of time steps used by the model predictive control algorithm to predict future states; w1, w2, and w3 are used to balance the relative importance of different state targets; and λ I ,λ U ,λ T , used to smooth the control sequence and suppress the violent fluctuation of the control input; SOC target Indicates the target state of charge of the stacked battery, which is the SOC value expected to be achieved after charging; SOH target Indicates the target health state of the stacked battery, which is the SOH level expected to be maintained; T target represents the target temperature of the stacked battery, which is the temperature range expected to be controlled; SOC(k+i|k) represents the state of charge of the stacked battery predicted in the kth control cycle, which is the k+ith control cycle; SOH(k+i|k) represents the health status of the stacked battery predicted in the kth control cycle, which is the k+ith control cycle; T(k+i|k) represents the temperature of the stacked battery predicted in the kth control cycle, which is the k+ith control cycle; I chg (k+i|k) represents the predicted charging current value of the k+i control cycle in the kth control cycle; U chg (k+i|k) represents the predicted charging voltage value of the k+i control cycle in the kth control cycle; T chg (k+i|k) indicates that in the kth control cycle, the charging duration of the k+ith control cycle is predicted; ||*|| indicates the second norm (Euclidean norm) of the variable, which indicates the degree to which the variable deviates from the target value; for a scalar variable, its second norm is the absolute value of the variable; for a vector variable, its second norm is the square root of the sum of the squares of the absolute values of each element; in summary, the physical meaning of optimizing the objective function J is to measure the degree of balance between the stacked battery in tracking the state target and suppressing control fluctuations in the next M control cycles. The smaller the value of the objective function, the more likely it is that in the future, while achieving the state target, the control decision can adjust the charging current, voltage, and time as smoothly as possible to achieve a better overall effect.

[0022] Furthermore, the optimization problem is expressed as follows: minJ;

[0023] I min ≤I chg (k+i|k)≤I max ,i=0,1,.....,M-1;

[0024] U min ≤U chg (k+i|k)≤U max ,i=0,1,.....,M-1;

[0025] T min ≤T chg (k+i|k)≤T max ,i=0,1,.....,M-1; where,

[0026] I chg (k+i|k),U chg (k+i|k),T chg (k+i|k) are the charging current, charging voltage and charging time of the k+i control cycle respectively, I min ,I max ,U min ,U max ,T min ,T max It is the value range constraint of charging current, voltage and time.

[0027] Compared with the existing technology, the advantages of this application are:

[0028] By performing preprocessing operations such as Kalman filtering, 3σ anomaly detection, interpolation correction, and polynomial smoothing on the collected data from single cells, high-frequency noise and outliers in the measurement process are eliminated, the data curve is smoothed, and the data is made more stable and reliable. Extracting first-order and second-order differential features reflects the dynamic changes in current and enriches the characteristic information.

[0029] The health index and current charge and discharge state parameters are combined through a normalized weighted linear combination to generate a multidimensional characteristic scalar for the stacked battery. This fusion reflects the health and operating status of the stacked battery, reducing the dispersion and one-sidedness of the evaluation indicators. The multidimensional characteristic scalar is a comprehensive representation of the health level and charge and discharge behavior, providing more comprehensive and reliable status information for subsequent charging optimization control. This helps to adaptively adjust the charging strategy and actively balance efficiency, time, and lifespan. This achieves charging optimization oriented towards extending lifespan, avoids biased control decisions caused by a single evaluation indicator, and maximizes the overall collaborative optimization effect.

[0030] The system uses the multidimensional characteristic scalars of stacked batteries as state input, combined with battery design parameters and charging requirements, to quantitatively describe the desired target state through an objective function. An optimization problem is then formulated to determine the deviation between the actual and target states. Charging current, voltage, and time are used as optimization variables, and minimizing the deviation is the optimization objective. A series of optimal charging control sequences are then solved. The current execution sequence is decoded into specific charging current, voltage, and time, directly guiding the charging process. Model predictive control updates the optimization results in real time based on state feedback, enabling adaptive dynamic optimization of the charging process. This minimizes the impact of charging on battery life while meeting charging time requirements, achieving an optimal balance between efficiency and lifespan.

[0031] Model predictive control introduces an objective function and constraints, forming a multi-objective optimization problem. The objective function weights the battery state deviation and charging sequence variations, flexibly adjusting the smoothness and tracking performance of the charging process, achieving a compromise between extending battery life and shortening charging time. Constraints on charging current, voltage, and time prevent damage to the battery caused by excessive charging current or voltage, acting as a safety protection mechanism to prevent premature battery failure and thus extend battery life. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The present application will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same numbers represent the same structures, wherein:

[0033] Figure 1 is an exemplary flow chart of a stacked battery parameter control method according to some embodiments of the present application;

[0034] Figure 2 is an exemplary flow chart of preprocessing a data set according to some embodiments of the present application;

[0035] Figure 3 is an exemplary flow chart of calculating the health index of stacked batteries at the current moment according to some embodiments of the present application;

[0036] Figure 4 This is an exemplary flowchart of generating a charging control strategy according to some embodiments of the present application. DETAILED DESCRIPTION

[0037] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0038] like Figure 1As shown, the battery temperature, voltage and current parameters of each single cell in the stacked battery are collected; the collected parameters are preprocessed to obtain a preprocessed data set; the preprocessed data set is divided into a training set and a test set; the battery temperature, voltage, current, and first-order current difference and second-order current difference in the training set are used as input, and the corresponding single cell health status is used as output to train a support vector machine SVM model to obtain a single cell health status assessment model; the trained single cell health status assessment model is used to assess the health status of the collected single cell parameters to obtain a health index of each single cell; based on the health status of all single cells The health index of the stacked battery is obtained by weighted averaging; the total charging current, total charging voltage and cumulative charging time of the stacked battery are obtained as the current charging state parameters; the total discharge current, total discharge voltage and cumulative discharge time of the stacked battery are obtained as the current discharge state parameters; the obtained stacked battery health index, current charging state parameters and current discharge state parameters are linearly combined to obtain a multidimensional feature scalar of the stacked battery; the multidimensional feature scalar is used as input to generate a charging control strategy for the stacked battery using a model predictive control algorithm; wherein the charging control strategy includes charging current, charging voltage and charging time.

[0039] Specifically, the battery temperature, voltage, and current parameters of each cell in the stack are collected. Specifically, the temperature sensor uses an NTC thermistor temperature sensor with a measurement range of -20°C to 100°C and an accuracy of ±0.5°C. Sensors are numbered T1 to Tn, where n is the number of cells. A temperature sensor is placed on the surface of each cell to collect the battery's operating temperature. The voltage sensor uses a resistor divider voltage sensor with a measurement range of 0 to 5V and an accuracy of ±0.1%. Sensors are numbered U1 to Un. A voltage sensor is connected in parallel between the positive and negative terminals of each cell to collect the battery terminal voltage. The current sensor uses a Hall effect current sensor with a measurement range of -100A to 100A and an accuracy of ±0.5%. Sensors are numbered I1 to In. A current sensor is connected in series with each cell to collect the battery's charge and discharge current. Data frame format definition: Data frame structure: start character (2 bytes) + timestamp (4 bytes) + temperature data (4 × n bytes) + voltage data (4 × n bytes) + current data (4 × n bytes) + checksum (2 bytes).

[0040] like Figure 2 As shown in the figure, the collected parameters are preprocessed to obtain a preprocessed data set; specifically, the battery temperature data is filtered. According to the measurement noise characteristics of the temperature sensor and the dynamic characteristics of the battery temperature change, the process noise covariance matrix Q and the measurement noise covariance matrix R of the Kalman filter are set. Among them, Q reflects the uncertainty of temperature change, and R reflects the size of the sensor measurement noise. For each single battery temperature measurement value sequence {Ti (k)|k=1,2,.....,m}, where i=1,2,.....,n is the battery cell number and m is the number of sampling points. Kalman filter algorithm is used for filtering. Set the initial temperature estimate. and the initial estimation error covariance matrix P i (0|0); Based on the temperature estimate at the previous moment and the state transition matrix, predict the temperature estimate at the current moment and the estimated error covariance matrix P i (k|k-1); Update the current temperature estimate based on the current temperature measurement and Kalman gain matrix and the estimated error covariance matrix P i (k|k); Repeat the prediction step and the update step until all sampling points are processed and the filtered temperature data sequence is obtained

[0041] Specifically, voltage data abnormal value detection and correction, the voltage measurement value sequence of each single cell

[0042] {U i (k)|k=1,2,......,m}, calculate its mean μ i and standard deviation σ i For any sampling point k, if its voltage measurement value U i (k) exceeds the interval [μ i -3σ i ,μ i +3σ i ], it is determined to be an abnormal value and the location of the abnormal point is recorded. i (k)), find the two nearest non-abnormal points (k1, U i (k1)) and (k2,U i (k2)), use the linear interpolation formula to correct the outliers: Repeat the outlier detection and correction steps until all outliers are corrected to obtain the corrected voltage data sequence

[0043]

[0044] Specifically, the current data is smoothed and filtered and differentially processed. Assume that the collected battery current data sequence is I=[I1,I2,......,I n The Savitzky-Golay filter is used to smooth the current sequence. The filter order M is set to 3, the window length K is set to 15; for each current value I in the current sequence i ,i=1,2,......,n, construct sliding window: take Ii The (K-1) / 2 points before and after form the window vector

[0045] In each sliding window F i According to the least squares criterion, a polynomial fitting equation system is constructed: XA=F i ;in, Vandermonde matrix, F is the window vector, A is the fitting polynomial coefficient vector, A=[a0,a1,......,a M ] T ; The function value of the fitting polynomial at the center point is taken as the smoothing value of the point, recorded as I i,smooth ,Right now i is the center point number; repeat until all points in sequence I are processed and the smoothed current sequence I is obtained. smooth ; Using the polynomial fitting coefficient A, calculate the first-order and second-order derivative values of the center point: Reflects the rate of change of current; Reflects the acceleration of current change; repeatedly obtains the first-order differential current sequence I diff1 and the second-order differential current sequence I diff2 . The smoothed current I smooth , first-order differential current I diff1 , second-order differential current I diff2 Combine by columns to get the expanded current data set I augment :I augment =[I smooth ,I diff1 ,I diff2 ].

[0046] The temperature data after Kalman filtering, the voltage data after 3σ correction and the expanded current data are aligned to the same time axis according to their corresponding timestamps to construct the preprocessed multi-parameter data set. This embodiment can effectively remove measurement noise from temperature data, correct outliers in voltage data, smooth current data and extract its differential information, and align multi-parameter data in the time dimension to obtain a high-quality, information-rich preprocessed data set.

[0047] The preprocessed dataset is divided into a training set and a test set. Specifically, the preprocessed multi-parameter dataset is randomized to disrupt the order of the data samples and eliminate the time correlation and batch correlation that may exist in the dataset. The hold-out method is used to divide the randomized dataset into a training set and a test set. Generally, 60% to 80% of the dataset is used as the training set, and the remaining 20% to 40% is used as the test set. The number of training set samples is denoted as N. train , the number of test set samples is Ntest , satisfying N train +N test =m.

[0048] Construct the input feature vector for each sample data in the training set Extract battery temperature Voltage Smoothed current First-order differential current and the second-order differential current Composed input feature vector x i (k). Combine the input feature vectors of all single cells in the stacked battery in the training set at the kth sampling moment into a feature sample X(k)=[x1(k),x2(k),.....,x n (k)], where n is the number of cells in the stack.

[0049] Set sample labels and quantify the health status of single cells based on historical operating data. Set three levels: healthy, subhealthy, and unhealthy, represented by the numbers 2, 1, and 0, respectively. For each feature sample \mathbf{X}(k) in the training set, construct a label vector y(k) = [y1(k), y2(k), ..., y n (k)] T , where y i (k)∈{0,1,2} is the quantified health status value of the i-th battery at the k-th sampling moment.

[0050] To train the SVM model, the modified Gaussian radial basis kernel function is used as the kernel function of the SVM model, and its expression is:

[0051]

[0052] Among them, w T ,w V ,w I ,w dI ,w ddI are the weight coefficients of battery temperature, voltage, smoothed current, first-order differential current and second-order differential current data, respectively, satisfying w T +w V +w I +w dI +w ddI=1; α and β are the shape parameters of the first-order differential current data and the second-order differential current data, respectively, which are used to adjust the local influence range of the first-order differential and second-order differential data. The value range of α and β is (0, 2]; γ is the scale parameter of the kernel function, which controls the overall scale of the kernel function. A method combining random search and K-fold cross validation is used to search for the optimal parameter combination in the parameter space. The search range is set for the regularization parameter C, the kernel function scale parameter γ, the first-order difference shape parameter α, and the second-order difference shape parameter β, for example, C∈[2 -5 ,2 15 ],γ∈[2 -15 ,2 5 ],α,β∈(0,2].

[0053] In the parameter space, several groups of random parameter combinations are generated by sampling with logarithmic uniform distribution, and an extended parameter combination set is formed with the preset parameter combination. Weighted K-fold cross validation is used to evaluate the performance of the SVM model under each parameter combination. The training set is randomly divided into K subsets of similar size, and K-1 subsets are selected each time as training data, and the remaining 1 subset is used as validation data to train K SVM sub-models. The classification accuracy of each sub-model is evaluated on the validation data, and the weighted average of the classification accuracy of the K sub-models is used as the performance metric of the SVM model under this parameter combination. The weighting coefficient is set according to the importance of each single cell in the stacked battery. Each parameter combination in the extended parameter combination set is traversed, and the weighted K-fold cross validation process is repeated. The parameter combination with the highest average classification accuracy is selected as the optimal parameter combination. The SVM model is set using the optimal parameter combination, and the feature samples of the training set and the corresponding labels {X(k), y(k)|k=1,2,......,N train The model is input and the optimal classification hyperplane is obtained by solving a convex quadratic programming problem, completing the training of the single-cell health status assessment model. The generalization performance of the trained single-cell health status assessment model is evaluated using the test set samples. Evaluation metrics such as classification accuracy, precision, recall, and F1 value are calculated on the test set to comprehensively assess the model's classification performance.

[0054] like Figure 3 As shown, the health index of the current stacked battery is calculated, including: for each single battery in the stacked battery, using sensors to collect battery temperature, voltage, and current data over a period of time, with a sampling frequency of 10 Hz and continuous sampling for 2 minutes, to obtain the time series of temperature, voltage, and current: T i =[T i1 ,T i2 ,.......,T i1200 ],i=1,2,......,N;U i =[U i1 ,Ui2 ,.......,U i1200 ],i=1,2,......,N;I i =[I i1 ,I i2 ,.......,I i1200 ],i=1,2,......,N; where T i ,U i ,I i are the temperature, voltage, and current sequences of the ith single cell, and N is the number of single cells in the stacked battery. i Perform Kalman filtering to filter out high-frequency noise and obtain the filtered temperature series T i,filter . For the collected voltage sequence U i Perform abnormal value detection, and set the judgment threshold to ±20% of the rated voltage of the single battery. The abnormal value exceeding the threshold is replaced by the average of the two normal values before and after, and the corrected voltage sequence U is obtained. i,revise . For the collected current sequence I i Smoothing filtering is performed, and a fifth-order polynomial is used to perform sliding fitting on 30 consecutive sampling points. The fitting result is used as the smoothed current value to obtain the smoothed current sequence I i,smooth Calculate the current difference between two adjacent points and obtain the first-order differential current sequence I i,diff1 ; Then calculate the difference between two adjacent points of the first-order differential current sequence to obtain the second-order differential current sequence I i,diff2 .

[0055] For the i-th single cell, extract the filtered temperature sequence mean T i,mean , the corrected voltage series mean U i,mean , smooth current sequence mean I i,mean , the first-order differential current mean I i,diff1,mean , second-order differential current mean I i,diff2,mean , forming the characteristic vector of the i-th single cell: X i =[T i,mean ,U i,mean ,I i,mean ,I i,diff1,mean ,I i,diff2,mean ] T To evaluate the health status of a single battery, for each single battery, its feature vector X i Input the pre-trained support vector machine classification model SVM_model, and use the classification decision function to classify X i To determine health status:

[0056] H i =SVM model ×classify(Xi ), H i ∈{“healthy” “sub-healthy” “unhealthy”}; where the classification decision function is: In the formula, X is the feature vector to be classified, M is the number of support vectors, and a j is the Lagrange multiplier, y j ∈{1,-1} is the category label corresponding to the j-th support vector, X j is the jth support vector, K(X j ,X) is the kernel function, and b is the bias term. The classification decision function divides the feature space into three regions, corresponding to the three states of "healthy", "sub-healthy" and "unhealthy". Calculate the health index of the single battery, set the health state quantification standard, and quantify the three states of "healthy", "sub-healthy" and "unhealthy" as 0.9, 0.6 and 0.3 respectively: I i =0.9, if H i For "health"; I i =0.6, if H i "Sub-health"; I i =0.3, if H i is “unhealthy”; where I i is the health index of the ith single cell, and is the health status H i quantized value of .

[0057] Calculate the stack battery health index and set the weight coefficient w according to the physical position and electrical connection topology of each single cell in the stack battery i The single cells at the end of the battery pack age faster due to poor heat dissipation conditions and large temperature gradients, so their weight is set to 0.6; the single cells in the middle have better heat dissipation conditions, so their weight is set to 1. The single cells on the parallel branches in the battery pack contribute equally to the characteristics of the end of the pack, so their weight is set to the inverse of the number of parallel branches. Calculate the stack battery health index I stack : Among them, w i is the weight coefficient of the i-th single cell, I i is the health index of the i-th single cell, and N is the number of single cells. I is obtained by weighted average stack It not only takes into account the health status of each single cell, but also takes into account its importance in the stacked battery, and can objectively reflect the overall health level of the stacked battery.

[0058] In this embodiment, a stacked battery is taken as an example. The battery is composed of four cells connected in series. According to the above steps, the health status of the four cells is evaluated as "healthy", "healthy", "sub-healthy", and "unhealthy", respectively, with corresponding health indexes of 0.9, 0.9, 0.6, and 0.3. The four cells are numbered 1 to 4 from top to bottom. Batteries 1 and 4 are located at the ends of the group and have a weight of 0.6. Batteries 2 and 3 are located in the middle and have a weight of 1. Substituting: I stack =(0.6×0.9+1×0.9+1×0.6+0.6×0.3) / (0.6+1+1+0.6)=0.7125, which is at a “sub-healthy” level and requires maintenance and management to improve usage conditions and delay battery aging.

[0059] Preferably, the weight coefficient w of each single cell is set according to the electrical connection topology of each single cell in the stacked battery. i , including: Determine the physical arrangement of the internal single cells and the electrical connection topology through disassembly analysis and electrical testing of the stacked battery: For simple series-parallel topology, directly according to the number of series branches N s and the number of parallel branches N p , determine the number of single cells N = N s ×N p , establish the topological connection matrix T connect :

[0060] Among them, T ij Indicates the i-th series branch i=1,2,......,N s , j-th parallel branch j=1,2,......,N p Single battery number on the

[0061] For complex nonlinear topologies, an adjacency matrix or adjacency table is used to describe the connection relationship. For example, for a stacked battery consisting of N single cells, its adjacency matrix A is an N×N square matrix: Among them, element a ij Indicates the connection relationship between single cell i and single cell j: a ij =1 means that the positive terminal of battery i is connected to the negative terminal of battery j; a ij =-1 means that the negative terminal of battery i is connected to the positive terminal of battery j; a ij = 0 means there is no direct connection between batteries i and j. The adjacency matrix A fully describes the topological structure of the stacked batteries.

[0062] After determining the topology, consider the influence of the physical location of the single battery on its importance and set the position weight coefficient a iFor a series branch, the single cell located in the middle has good heat dissipation conditions, low operating temperature, slow aging, and a greater impact on the health of the stacked battery, so a larger position weight coefficient should be assigned; while the single cells located at the two ends have poor heat dissipation conditions, large temperature gradients, and fast aging, so the impact is relatively small and a smaller position weight coefficient should be assigned. For example, the weight coefficient can be set in the form of a quadratic function: i =4h×(1-h), where Where h is the relative position of the single battery in the series branch, and its value is in the range of [0, 1]. i Satisfy the normalization condition ∑a i = 1. For parallel branches, the physical location of each battery has little effect on its importance, and the same position weight coefficient can be assigned:

[0063] Furthermore, considering the influence of the electrical connection mode of the single battery on its importance, the connection weight coefficient β is set i : For the series branch, each single cell carries the same current and has a similar impact on the health level of the stacked battery, so it is given the same connection weight coefficient: For parallel branches, the current carried by each single cell is related to its health status and internal resistance. A single cell with good health and low internal resistance will share more current and have a greater impact on the stacked battery. Therefore, a larger connection weight coefficient should be assigned. For example, based on the capacity C of each single cell, i Set the weight coefficient: In the above formula, C i is the capacity of the single battery in the i-th parallel branch, ∑C j It is the sum of the capacities of all single cells in the parallel branch. i Satisfy the normalization condition ∑β i =1.

[0064] For the series-parallel topology, the position weight coefficient a of the single battery is comprehensively considered. i and the connection weight coefficient β j , calculate the weight w of the single battery on the i-th series branch and the j-th parallel branch ij : In the above formula, the denominator is the sum of the product of the position weights and the connection weights of all single cells, which is used for normalization to ensure that ∑w ij = 1. Fill the weight matrix W into the topological connection matrix T connect The normalized single cell weight matrix W is obtained at the corresponding position in battery :

[0065]

[0066] For stacked batteries with complex nonlinear connections, it is difficult to directly set the position weight and connection weight. Instead, the circuit analysis method is used to calculate the weight of the single battery: based on the adjacency matrix A, an equivalent circuit model of the stacked battery is established. i and equivalent internal resistance r i Construct nodes and connect branches with ideal wires to obtain a circuit model containing N nodes; apply the node voltage method (applicable to the case of fewer branches) or the loop current method (applicable to the case of fewer nodes) to establish the node voltage equation group or the loop current equation group; solve the equation group to obtain the node voltage V at both ends of each single battery i and branch current I i If the equations are difficult to solve, you can use SPICE or other circuit simulation software to solve them numerically; calculate the power P of a single battery. i :P i =U i ×I i Power P i It reflects the influence of the i-th single cell on the performance of the stacked battery; based on the power of the single cell, its weight coefficient w is calculated i : The greater the power of a single cell, the greater its impact on the health level of the stacked battery and the higher its weight coefficient.

[0067] The total charging current, total charging voltage and cumulative charging time of the stacked battery are obtained as the current charging state parameters; the total discharge current, total discharge voltage and cumulative discharge time of the stacked battery are obtained as the current discharge state parameters; specifically, in the charging circuit of the stacked battery, a high-precision current sensor (such as a Hall current sensor, a shunt, etc.) is used to collect the charging current of each single battery respectively to obtain a current sampling value I i , where N is the number of cells in the stacked battery. The charging current sampling values of each cell are accumulated to obtain the total charging current I_charge of the stacked battery: In order to reduce the influence of measurement noise, a sliding average filter can be applied to I_charge: Among them, M is the sliding window length, I 充电，j is the total charging current measurement value at the jth sampling moment.

[0068] To collect the total charging voltage, a high-voltage sampling circuit is connected in parallel across the positive and negative terminals of the stacked battery. A voltage sensor (such as a resistor divider, capacitor divider, or isolated op amp) collects the total charging voltage V_charged in real time. To ensure sampling accuracy, the voltage sensor must meet the following requirements: a range covering the maximum operating voltage of the stacked battery; a sampling frequency exceeding the frequency of voltage fluctuation; strong common-mode interference immunity to suppress high-frequency noise; and high isolation voltage resistance to prevent damage to the control side from the high-voltage side. The collected total charging voltage sequence {V_charged_1, V_charged_2, ..., V_charged_K} is subjected to denoising processing (such as limiting filtering or median filtering) to obtain the filtered total charging voltage measurement value V_charged.

[0069] Get the accumulated charging time, set a dedicated timer, and record the charging start time t start and the current time t current The timer can be a high-precision RTC chip (such as DS3231) or an on-chip timer. start and t current The time difference is used to obtain the cumulative charging time t_charging: t_charging = t current -t start , t_charging can be expressed in the format of "hours: minutes: seconds" or converted to seconds. During the charging process of the stacked battery, interruptions may occur, resulting in discontinuous t_charging. Therefore, the time of each interruption t pause,j and the restart time t resume,j , calculate the cumulative interruption time t pause : t pause Subtract it from t_charge to get the corrected cumulative charging time t_charge_correction: t_charge_correction = t_charge - t pause .

[0070] Generate charging state parameters, and package and upload the total charging current I_charging, total charging voltage V_charging, and corrected cumulative charging time t_charging_correction obtained in the above steps at a fixed period (e.g., 1 second) to form the stacked battery current charging state parameter frame CHG_Param: CHG_Param = [I_charging, V_charging, t_charging_correction].

[0071] The acquisition of discharge state parameters and total discharge current is similar to the acquisition of total charge current. The appropriate current sensor is used to collect the discharge current of each single cell {I1,I1,.....,I N}, and after accumulation, we get I_discharge: Similarly, I_discharge can be filtered and smoothed by methods such as sliding average.

[0072] The total discharge voltage V_discharge is collected and processed in the same way as the total charge voltage V_charge. The voltage across the stacked battery is sampled in real time using a voltage sensor and subjected to denoising filtering. The cumulative discharge time t_discharge is obtained in a similar way to the charge time t_charge. A timer is used to record the start time of discharge and the current time, and the time difference is calculated as t_discharge. The accumulated interruption time during the discharge process is then deducted to obtain the correction value t_discharge_correction: t_discharge_correction = t_discharge - t pause .

[0073] Generate discharge state parameters, package I_discharge, V_discharge, and t_discharge_correction to form the stacked battery current discharge state parameter frame DCH Param :DCH Param =[I_discharge, V_discharge, t_discharge_correction]. Since the actual working process of the stacked battery is to alternately charge and discharge, and CHG Param and DCH Param The upload cycle is short, and the timestamp may be discontinuous, the parameter frame interval may be uneven, and other problems will be unfavorable for subsequent analysis and calculation. Therefore, it is necessary to synchronize the two parameter frames: record CHG Param and DCH Param The receiving time {t chg,1 ,t chg,2 ,......,t chg,N} and {t dch,1 ,t dch,2 ,......,t dch,N}; Determine the synchronization period T sync , T sync Should be greater than CHG Param and DCH Param Upload cycle, and meet T sync =n×t upload , where n is a positive integer; traverse CHG Param and DCH Param For each parameter frame in , check its timestamp t frame Whether t is satisfied frame =k×T sync (k is a non-negative integer): If it is satisfied, the parameter frame is aligned with the synchronization time and remains unchanged; if it is not satisfied, the parameter frame is deleted, or its timestamp is corrected by interpolation; after the traversal is completed, CHG Param and DCH Param Parameter frame timestamp and synchronization period T syncAligned, with data points spaced equally apart. Based on the acquired charge and discharge state parameters, derived parameters such as the stacked battery's state of charge (SOC) and state of health (SOH) can be further calculated. Combined with indicators such as the stacked battery health index, this forms a complete state monitoring and health assessment dataset for subsequent charge control optimization.

[0074] The obtained stacked battery health index, current charge state parameter and current discharge state parameter are linearly combined to obtain a multi-dimensional characteristic scalar of the stacked battery; including: the stacked battery health index I stack , current charging state parameters and current discharging state parameters are arranged in a preset order to form a characteristic parameter set F: F = {I stack ,I chg ,U chg ,T chg ,I dis ,U dis ,T dis}; Among them, I chg ,U chg ,T chg are the total charging current, total charging voltage and cumulative charging time respectively; I dis ,U dis ,T dis The total discharge current, total discharge voltage and cumulative discharge time are obtained respectively; each parameter in the characteristic parameter set F is normalized to obtain the normalized characteristic parameter set F norm :F norm ={f1,f2,.....,f7}, where f i is the normalized ith characteristic parameter, and the calculation formula is: Among them, max(F i ) and min(F i ) are the maximum and minimum values of the i-th feature parameter in the historical data set; set each normalized feature parameter f i The weight coefficient μ i , the weight coefficient satisfies the normalization condition: Using the normalized feature parameter f i and its weight coefficient μ i , through weighted linear combination, the multidimensional characteristic scalar X of the stacked battery is obtained stack :

[0075] like Figure 4As shown in FIG, the multidimensional characteristic scalar is used as input to generate the charging control strategy of the stacked battery using the model predictive control algorithm; wherein the charging control strategy includes charging current, charging voltage and charging time. Specifically, the state information of the stacked battery in the current control cycle k is obtained, and the multidimensional characteristic scalar X of the stacked battery is obtained. stack (k), obtained by fusion of sensor measurement and state estimation algorithm, characterizes the static and dynamic characteristics of the stacked battery at time k in the control cycle:

[0076] X stack (k)=[SOC(k),SOH(k),T(k),I chg (k-1),U chg (k-1),T chg (k-1),.....], where SOC(k) is the state of charge of the stacked battery, SOH(k) is the state of health, T(k) is the temperature,

[0077] I chg (k-1), U chg (k-1) and T chg (k-1) is the charging current, voltage and time of the previous control cycle.

[0078] The target state vector X of the stacked battery target , determined by the stacked battery design parameters and user charging requirements, characterizes the expected charging effect: X target =[SOC target ,SOH target ,T target ], where SOC target is the target state of charge (e.g. 100%), SOH target is the target health state (such as 90%), T target is the target temperature (e.g. 25°C). The target state vector X based on the stacked battery target , construct the objective function f(X) of the charging process. The objective function is used to quantify the pros and cons of the charging control strategy and make the stacked battery state approach the target state. This solution uses the objective function in the form of weighted bi-norm: f(X target )=w1×||SOC target -SOC||+w2×||SOH target -SOH||+w3×||T target -T||, where ||*|| represents the second norm (Euclidean norm), and w1, w2, and w3 are weighting coefficients for the three state quantity deviations, which can be adjusted according to actual needs. For example, if you prioritize rapid improvement of SOC, you can increase the weight of w1 appropriately; if you prioritize balanced management of SOH, you can increase the weight of w2 appropriately.

[0079] The charging control sequence {U(k|k), U(k+1|k), ..., U(k+M-1|k)} of the next M control cycles is used as the optimization variable, where the charging control variable U(k) of a single control cycle includes the charging current I chg (k), charging voltage U chg (k) and charging time T chg (k). At the same time, the physical constraints and change rate constraints of the control quantity in the actual charging process are taken into account and used as constraints of the optimization problem. On this basis, the objective function f(X) is used as the performance indicator to construct the following optimization problem:

[0080]

[0081] Constraints:

[0082] I min ≤I chg (k+i|k)≤I max ,i=0,1,.....,M-1;

[0083] U min ≤U chg (k+i|k)≤U max ,i=0,1,.....,M-1;

[0084] T min ≤T chg (k+i|k)≤T max ,i=0,1,.....,M-1;

[0085] Where J is the optimization target, which consists of two parts: the first part Characterizes the cumulative degree of deviation of the state quantity from the target state in the next M control cycles. SOC(k+i|k), SOH(k+i|k) and T(k+i|k) are the future prediction values based on the current state. is the weighted bi-norm of the control variation in adjacent control cycles, which is used to smooth the control sequence and avoid drastic fluctuations in charging current, voltage, and time. I ,λ U ,λ T is the corresponding weighting coefficient. min / I max 、U min / U max 、T min / T max These are the physical constraints of charging current, voltage, and time, which are determined according to battery parameters and charging equipment performance.

[0086] The above optimization problem can be converted into a sequential quadratic programming problem and solved using the SQP algorithm to obtain the optimal control sequence {U*(k|k),U*(k+1|k),......,U*(k+M-1|k)} for the next M control cycles. Specifically, the original problem is converted into a sequence of quadratic programming subproblems, and the subproblem objective function is expanded by a second-order Taylor at the constrained nonlinear iteration point, and the constraints are linearized; Lagrange multipliers are introduced to convert the subproblems into unconstrained optimization problems, and the Newton method or quasi-Newton method is used to solve them to obtain the search direction of the control variables; the Armijo criterion is used to determine the search step size, update the control variables, and correct the Lagrange multipliers to obtain a new iteration point; repeat until the iteration point meets the termination criterion and output the optimal solution. This solution uses Sequential Least Squares Programming (SLSQP) as the optimizer of the SQP algorithm, which can efficiently solve nonlinear programming problems under multiple constraints and is relatively simple to implement in programming. The charging strategy generation method based on model predictive control in this application can realize closed-loop optimization control of the stacked battery charging process. Compared with traditional open-loop control and PID feedback control, this method can fully exploit the state information of the stacked batteries, balance various performance requirements, predict future state changes, and optimize the charging process as a whole. At the same time, by convexifying and linearizing the multi-constrained, nonlinear optimization control problem through a sequential quadratic programming algorithm, it can significantly reduce the computational load of model predictive control and improve the real-time performance of the control system.

[0087] Extract the charging control strategy of the current control cycle from the optimal control sequence and update the control sequence to achieve rolling optimization: The sequential quadratic programming algorithm is used to solve the optimal control sequence for the next M control cycles.

[0088] {U*(k|k),U*(k+1|k),......,U*(k+M-1|k)}, where each element U*(k+i|k) is a multidimensional vector containing the optimal charging current I in the k+i control cycle chg (k+i|k), optimal charging voltage U chg (k+i|k) and optimal charging time T chg *(k+i|k). Extract the first element U*(k|k) of the optimal control sequence as the optimal charge control quantity for the current control cycle k of the stacked battery. Obtain the three components of U*(k|k) by array indexing or vector slicing:

[0089] I chg (k)=U*(k|k)[0], extract the charging current component; U chg (k)=U*(k|k)[1], extract the charging voltage component; T chg(k) = U*(k|k)[2], extracting the charging time component; where U*(k|k)[i] represents the i-th element of the vector U*(k|k), and the subscript starts at 0. The implementation of component extraction is related to the data type and indexing rules of the programming language.

[0090] According to the extracted optimal control quantity components, the charging control strategy of the stacked battery in the kth control cycle is generated, which mainly includes three variables: charging current I chg (k); Charging voltage U chg (k); Charging time T chg (k) is the target value of the new round of charging control. The generated charging control strategy is packaged into a control instruction and sent to the charging control unit through a communication interface (such as CAN bus, RS485, etc.) to control the power output end to adjust the charging current and charging voltage, and at the same time update the time parameters of the charging timer: set the charging current set_current (I chg (k)); Set the charging voltage set_voltage(U chg (k)); Set charging time set_timer(T chg (k)); here, set_current(*), set_voltage(*), and set_timer(*) represent the current, voltage, and time setting functions corresponding to the charging control unit. The charging state is adjusted internally through a closed-loop control algorithm (such as PI control) so that the actual value tracks the target value.

[0091] After the charging control strategy is issued, the stacked battery starts to perform charging in the new control cycle k, and its status information also changes accordingly, which needs to be measured and updated. The real-time charging current I of the stacked battery is collected by current and voltage sensors. chg,meas (k) and charging voltage U chg,meas (k), and record the actual charging time T chg,meas (k), as state feedback. Combined with the battery temperature T collected by the temperature sensor meas (k), as well as the current state of charge SOC(k) and health state SOH(k) given by the SOC and SOH estimation algorithms, update the status information X of the stacked battery stack (k): X stack (k)=[SOC(k),SOH(k),T meas (k),I chg,meas (k),U chg,meas (k),T chg,meas (k),......]. After the state information is updated, it is stored in the state observation vector and used for state estimation and objective function calculation in the next control cycle.

[0092] At the end of the current control cycle, the stacked battery has completed one charging step and needs to make the next charging decision. To achieve rolling optimization control, the optimal control sequence needs to be updated as the initial value for the next round of sequential quadratic programming.

[0093] First, remove the first element U*(k|k) of the optimal control sequence and move the remaining M-1 elements forward one position: Then, a new variable to be optimized, U(k+M|k+1), is added to the end of the control sequence. Its initial value can be the suboptimal solution U(k+M|k) obtained in the previous round of optimization, or an extrapolated value of the current control variable U(k) (such as maintaining the same change trend as the previous elements). Finally, the updated control sequence is:

[0094] {U*(k+1|k+1),U*(k+2|k+1),......,U*(k+M-1|k+1),U*(k+M|k+1)}, the sequence length is still M, but each element has rolled forward one control cycle, and the last element is the newly added variable to be optimized. The updated control sequence can be directly used for a new round of sequential quadratic programming solution to generate a new optimal charging strategy in the next control cycle k+1. This application realizes the coordination between the output results of the model predictive control algorithm and the underlying execution unit through the extraction of optimal control quantities and the generation of charging control strategies; through the update of state information and the update of control sequences, rolling optimization based on feedback correction is realized, which ensures the adaptive ability of the control strategy to the changes in the state of stacked batteries.

Claims

1. A stacked battery parameter control method, characterized in that: include: Collect battery temperature, voltage and current parameters of each single cell in the stacked battery; Preprocessing the collected parameters to obtain a preprocessed data set; Divide the preprocessed dataset into training set and test set; The support vector machine (SVM) model is trained using the battery temperature, voltage, current, first-order current difference, and second-order current difference in the training set as input and the corresponding single-cell battery health status as output to obtain a single-cell battery health status assessment model. Using the trained single-cell battery health status assessment model, the collected single-cell battery parameters are evaluated for health status to obtain the health index of each single-cell battery; The health index of the current stacked battery is obtained by weighted average based on the health index of all single batteries; Obtain the total charging current, total charging voltage, and accumulated charging time of the stacked battery as current charging state parameters; Obtain the total discharge current, total discharge voltage, and accumulated discharge time of the stacked batteries as current discharge state parameters; The obtained stacked battery health index, current charge state parameter, and current discharge state parameter are linearly combined to obtain a multi-dimensional characteristic scalar of the stacked battery; Taking multidimensional feature scalars as input, a model predictive control algorithm is used to generate a charging control strategy for the stacked battery; wherein the charging control strategy includes charging current, charging voltage and charging time.

2. The stacked battery parameter control method according to claim 1, wherein: Preprocess the collected parameters, including: Use Kalman filter algorithm to filter the battery temperature data; The 3σ principle is used to detect outliers in voltage data; the detected outliers are corrected using a linear interpolation algorithm; The current data is smoothed using a smoothing filter algorithm based on polynomial fitting, and first-order and second-order differences are performed to obtain smoothed current data, first-order differential current data, and second-order differential current data; Combining the smoothed current data, the first-order differential current data, and the second-order differential current data to obtain an expanded current data set; The filtered battery temperature data, the corrected voltage data, and the expanded current data are aligned according to the timestamps to obtain the preprocessed data set.

3. The stacked battery parameter control method according to claim 2, wherein: Obtain smoothed current data, as well as first-order differential current data and second-order differential current data, including: According to each data point ii in the current data sequence, with the data point ii as the center, multiple data points before and after the data point ii are selected to form a sliding window; In each sliding window, the least squares method is used to perform polynomial fitting on multiple data points in the corresponding sliding window, and the fitting order is the set filter order; The polynomial fitting value of the center point in the sliding window is used as the filtering result of the corresponding data point ii to obtain the smoothed current data sequence; According to the fitted polynomial, the first-order derivative value of each data point ii is calculated as the first-order differential current, which reflects the current change rate; According to the fitted polynomial, the second-order derivative value of each data point is calculated as the second-order differential current, which reflects the current change acceleration.

4. The stacked battery parameter control method according to claim 2, wherein: Get the health index of the current stack battery, including: Extract the battery temperature, voltage, smoothed current, first-order differential current and second-order differential current data of each single battery to form the input feature vector X of the corresponding single battery i ,i=1,2,.....,N, where N is the number of single cells in the stacked battery; The input feature vector X of each single battery i Input the single battery health status assessment model, and use the classification hyperplane to classify the input feature vector X i Perform health status classification to obtain the health status H of the corresponding single battery at the current moment i ,i=1,2,......,N; where the health status H i Includes health, sub-health and unhealth; According to the preset health status quantification rules, the health status H of each single battery is i Quantified into the corresponding health index value I i ; According to the electrical connection topology of each single cell in the stacked battery, the weight coefficient w of the single cell is set i ,i=1,2,.....,N; Using the weight coefficient w i Health Index Value I i Perform weighted averaging to obtain the health index I of the stacked batteries at the current moment. stack : Among them, I stack is the health index of the stacked battery.

5. The stacked battery parameter control method according to any one of claims 2 to 4, characterized in that: The multi-dimensional characteristic scalars of the stacked battery are obtained, including: The health index of the stacked battery I stack , current charging state parameters and current discharging state parameters are arranged in a preset order to form a characteristic parameter set F: F={I stack ,I chg ,U chg ,T chg ,I dis ,U dis ,T dis } Among them, I chg ,U chg ,T chg They are the total charging current, total charging voltage and cumulative charging time obtained respectively; I dis ,U dis ,T dis They are the total discharge current, total discharge voltage and cumulative discharge time obtained respectively; Normalize each parameter in the feature parameter set F to obtain the normalized feature parameter set F norm : <h2 style=";text-align:left;direction:ltr">F<h2 style=";text-align:left;direction:ltr"> norm <h2 style=";text-align:left;direction:ltr"> (f1,f2,......,f7) Among them, f i is the normalized ith characteristic parameter, and the calculation formula is: Among them, max(F i ) and min(F i ) are the maximum and minimum values of the i-th feature parameter in the historical data set respectively; Set each normalized feature parameter f i The weight coefficient μ i , the weight coefficient satisfies the normalization condition: Using the normalized feature parameter f i and its weight coefficient μ i , through weighted linear combination, the multidimensional characteristic scalar X of the stacked battery is obtained stack :

6. The stacked battery parameter control method according to claim 5, characterized in that: Generate a charging control strategy for the stacked battery using a model predictive control algorithm, including: Get the multidimensional characteristic scalar X of the stacked battery stack (k), where k represents the current control period; According to the design parameters of the stacked battery and the user's charging requirements, the objective function f(X target ), where X target is the target state vector of the stacked battery, which contains the target state of charge SOC of the stacked battery target Target health status SOH target and target temperature T target ; Based on the objective function f(X target ), construct an optimization problem, where the optimization variables in the optimization problem are the charging control sequences {U(k|k),U(k+1|k),......,U(k+M-1|k)} for the next M control cycles; The sequential quadratic programming algorithm is used to solve the optimization problem and obtain the optimal charging control sequence {U*(k|k),U*(k+1|k),......,U*(k+M-1|k)} for the next M control cycles; The first element U*(k|k) of the optimal control sequence is used as the optimal charge control quantity of the stacked battery in the current control cycle k, and decoded into the corresponding charging current I chg (k), charging voltage U chg (k) and charging time T chg (k), as the charging control strategy for the stacked batteries in the current control period.

7. The stacked battery parameter control method according to claim 6, characterized in that: Objective function f(X target ), the expression is as follows: f(X target )=w1×||SOC target -SOC||+w2×||SOH target -SOH||+w3×||T target -T|| Among them, SOC target ,SOH target ,T target They are the target state of charge, target state of health and target temperature of the stacked battery, corresponding to the target state vector X target The three components of the stacked battery are SOC, SOH and T, respectively, which correspond to the current multidimensional characteristic scalar X stack (k); w1, w2, w3 are the weighting coefficients of the three state components.

8. The stacked battery parameter control method according to claim 7, characterized in that: Construct an optimization problem, including: Take the target state vector X target To control the target, according to the current state X of the stacked battery stack (k) and the state prediction value X in the next M control cycles stack (k+i|k), i=1,2,.....,M, construct the objective function J, which is used to evaluate the degree to which the stacked battery state deviates from the target state in the next M control cycles; The charging control sequence of the next M control cycles {U(k|k),U(k+1|k),......,U(k+M-1|k)} is used as the optimization variable, where the charging control variable U(k) of a single control cycle includes the charging current I chg (k), charging voltage U chg (k) and charging time T chg (k); The value range of the charging control variable U(k) is used as the constraint condition of the optimization problem, and the objective function J is used as the performance indicator of the optimization problem to construct the optimization problem.

9. The stacked battery parameter control method according to claim 8, characterized in that: The objective function J is expressed as follows: Among them, SOC(k+i|k), SOH(k+i|k) and T(k+i|k) are the predicted values of the state of charge, health state and temperature in the future i-th control cycle, respectively, of the k-th control cycle, and the corresponding state prediction value X stack The three components of (k+i|k); w1, w2, w3 are the weighted coefficients of state deviation; I chg (k+i|k),U chg (k+i|k),T chg (k+i|k) are the charging current, charging voltage and charging time of the k+i control cycle, corresponding to the three components of the charging control variable U(k+i|k); λ I ,λ U ,λ T It is the weighting coefficient of the change of charging current, voltage and time in adjacent control cycles.

10. The stacked battery parameter control method according to claim 9, characterized in that: The optimization problem is expressed as follows: minJ I min ≤I chg (k+i|k)≤I max ,i=0,1,.....,M-1 U min ≤U chg (k+i|k)≤U max ,i=0,1,.....,M-1 T min ≤T chg (k+i|k)≤T max ,i=0,1,.....,M-1 Among them, I chg (k+i|k),U chg (k+i|k),T chg (k+i|k) are the charging current, charging voltage and charging time of the k+i control cycle respectively, I min ,I max ,U min ,U max ,T min ,T max It is the value range constraint of charging current, voltage and time.

Images