Method for diagnosing short circuit of new energy in vehicle power battery

Abstract

The invention relates to the technical field of power battery fault diagnosis, in particular to a method for diagnosing short circuit of new energy in a vehicle power battery, which comprises the following steps of: S1, acquiring real-time working data of the power battery; s2, constructing an FFRLS combined UKF algorithm, identifying the working parameters of the battery, and estimating the SOC of the battery in real time; s3, constructing a fault feature function of the SOC curve difference between each battery monomer and the reference battery; and S4, substituting the identification data of each battery into the fault characteristic function, and carrying out internal short circuit fault detection. According to the method, the sensitivity and accuracy of battery internal short circuit fault diagnosis can be effectively improved, meanwhile, the fault detection precision and the system safety are improved through the online parameter identification technology, and the method has good adaptability and robustness and is suitable for battery internal short circuit detection under the dynamic working condition.

Description

Technical Field

[0001] The present invention relates to the technical field of power battery fault diagnosis, and particularly to a method for diagnosing internal short circuits in power batteries of new energy in-use vehicles. Background Art

[0002] The on-line internal short circuit fault of a power battery refers to the phenomenon that an abnormal conduction occurs between the positive and negative electrodes inside the battery due to various reasons (such as diaphragm damage, metal impurities, manufacturing defects, etc.). This kind of fault will cause local overheating, increased self-discharge, and may even cause thermal runaway, seriously threatening the safety and service life of the battery.

[0003] Existing methods for diagnosing internal short circuit faults collect voltage and current data of the battery or detect changes in the battery internal resistance, and combine with a fault mode recognition model for fault diagnosis, which have various deficiencies:

[0004] During the operation of the battery, the voltage signal may be affected by factors such as the external environment and circuit interference, resulting in noise. The signal noise collected by the voltage sensor is relatively large, causing obvious fluctuations in the voltage curve, making it difficult to directly characterize the correlation between batteries and may thus affect the accuracy of diagnosis.

[0005] Existing fault detection methods need to determine the change data of the battery internal resistance. However, in practical applications, the test method of the battery internal resistance is relatively complex and often requires suspending the operation of the battery, making it difficult to achieve on-line detection of the battery internal resistance and not suitable for the requirements of real-time monitoring and diagnosis.

[0006] Commonly used parameter estimation algorithms, such as the least squares method and the recursive least squares method, may be interfered by historical data, resulting in a slower response speed or a decrease in accuracy. To a certain extent, these data may cause the algorithm to be less sensitive to new data, reducing the sensitivity of real-time detection.

[0007] The dynamic changes in working conditions such as the temperature and load of the battery will cause changes in some battery parameters (such as SOC and internal resistance). Existing models and algorithms do not fully consider these changes, and the diagnostic results may not be accurate enough. Summary of the Invention

[0008] The present invention provides a method for diagnosing internal short circuits in power batteries of new energy in-use vehicles, which can effectively improve the sensitivity and accuracy of diagnosing internal short circuit faults of the battery. At the same time, through on-line parameter identification technology, the accuracy of fault detection and the safety of the system are improved, and it has good adaptability and robustness, and is suitable for detecting internal short circuits of the battery under dynamic working conditions.

[0009] The present application provides the following technical solutions:

[0010] A method for diagnosing internal short circuits in power batteries of new energy in-use vehicles, comprising the following steps:

[0011] S1. Collect the real-time working data of the power battery;

[0012] S2. Construct the FFRLS combined with the UKF algorithm to identify the battery working parameters and estimate the SOC of the battery in real time;

[0013] S3. Construct a fault feature function for the difference in the SOC curves of each battery cell and the reference battery;

[0014] S4. Substitute the identification data of each battery cell into the fault feature function for internal short circuit fault detection.

[0015] Technical principle: Construct the FFRLS combined with the UKF algorithm to identify the working parameters of the battery, convert the voltage data into an SOC curve, construct a fault feature extraction function for the SOC curve, and diagnose the internal short circuit fault of the power battery.

[0016] Beneficial effects: The fluctuation of the SOC curve is smaller than that of the voltage curve. Using the SOC curve for fault diagnosis can effectively reduce the probability of false fault alarms;

[0017] The UKF algorithm can more accurately capture the nonlinear characteristics of the battery system through unscented transformation, and realize the optimal estimation of the battery SOC through the UKF algorithm recursion, which can effectively deal with the nonlinear and uncertainty problems of the battery model.

[0018] Through the online parameter identification technology, the accuracy of fault detection and the safety of the system are improved, and it has good adaptability and robustness, and is suitable for the internal short circuit detection of the battery under dynamic working conditions.

[0019] Furthermore, the S2 includes:

[0020] S21. Through the real-time collected voltage sequence data, select the battery cell with the smallest standard deviation of the voltage sequence as the reference battery and calculate the voltage curve of the reference battery;

[0021] S22. Establish an equivalent circuit model and a discretized equation corresponding to the equivalent circuit model;

[0022] S23. According to the discretized equation of the equivalent circuit model, use the FFRLS method with a variable forgetting factor to identify the online parameters of the equivalent circuit model;

[0023] S24. Based on the unscented Kalman filter algorithm UKF, online estimate the battery SOC.

[0024] Furthermore, the S23 includes:

[0025] S231. Obtain the transfer function of the equivalent circuit model according to the dynamic equation of the equivalent circuit model;

[0026] S232. Calculate the variable forgetting factor and the gain matrix K0;

[0027] S233. Update the covariance matrix P0(k);

[0028] S234. Update the parameter to be estimated θ(k);

[0029] S235. Return the parameter identification value at the next moment to S232, and iteratively execute S232 - S235. When the change in the covariance matrix is less than the set threshold, end the FFRLS algorithm.

[0030] Furthermore, in S232, the simulated annealing algorithm is used to solve the optimal forgetting factor.

[0031] Beneficial effects: Dynamically update the forgetting factor according to the change of the system state to adapt to different working conditions, and at the same time avoid the performance degradation caused by a fixed forgetting factor, making the RLS algorithm more robust. The simulated annealing algorithm can avoid falling into local optima and find the global optimal solution of the forgetting factor.

[0032] Furthermore, the S24 includes:

[0033] S241. Assign initial values to the model parameters;

[0034] S242. Determine the state equation vector and the observation vector equation of the battery system;

[0035] S243. After obtaining the state equation and the observation equation of the battery system, realize the optimal estimation of the battery SOC according to the recursive steps of the UKF algorithm.

[0036] Furthermore, in the S3 step, the data points from t0 - t e are defined as

[0037]

[0038] where n is the number of voltage data collected by the sensor;

[0039] Calculate the mean value at each moment through the data points as:

[0040]

[0041]

[0042]

[0043] Sort the data points at each moment from largest to smallest and define them as Then calculate the median of the voltage data at each moment:​

[0044]

[0045] Then the reference data at each moment is defined as:

[0046] And

[0047] where n sel is the number of voltage data points that satisfy the inequality;

[0048] Then the fault feature function is:

[0049] Beneficial effect: The fault feature function based on the exponential function can amplify the small differences in the SOC curves between the target battery and the reference battery, which is beneficial to improving the accuracy of fault detection.

[0050] Furthermore, in S4, a functional data outlier detection algorithm based on the deformation abnormality is adopted to determine the internal short - circuit fault of the power battery.

[0051] Beneficial effect: Using the functional data outlier detection algorithm based on the deformation abnormality to determine the characteristic curve of the abnormal cell can effectively improve the sensitivity and accuracy of the diagnosis of the internal short - circuit fault of the battery and enhance the robustness of the system.

[0052] Furthermore, S4 includes:

[0053] S41. Calculate the discrete Fréchet distance between the SOC curves of each battery and the reference battery, and generate a monomer curve graph that changes with time;

[0054] S42. Determine the deformation abnormality detection reference curve according to the monomer curve graphs of all batteries;

[0055] S43. Calculate the deformation abnormality of the SOC characteristic curve of each battery monomer, then combine the deformation abnormality with the depth value MBD, change the vertical distance from the original depth point (MEL, MBD) to the parabola, and compare with the reference curve to finally identify the outlier.

[0056] Beneficial effect: Characterized by the discrete Fréchet distance of the monomer SOC curve estimated by the UKF algorithm, it can sensitively capture the small differences between the SOC characteristic curves of different battery monomers, detect battery faults at an early stage, can better adapt to the complex behavior of the battery system, effectively filter noise and interference, and is applicable to the application scenario of real - time online detection. Description of the Drawings

[0057] Figure 1 It is a flowchart of a method for diagnosing the internal short - circuit of a power battery in a new - energy vehicle in use;

[0058] Figure 2 To construct a schematic diagram of the process for estimating the battery SOC using the FFRLS combined with the UKF algorithm;

[0059] Figure 3 For the voltage curve of the 5Ω internal short - circuit resistance under the DST working condition;

[0060] Figure 4 For the schematic diagram of the second - order RC equivalent circuit model;

[0061] Figure 5 For the schematic diagram of the SOC estimation result under dynamic conditions;

[0062] Figure 6 For the internal short - circuit detection result within the 5Ω short - circuit resistance. Specific implementation manners

[0063] The following is a further detailed description through specific implementation manners:

[0064] Example 1

[0065] A new - energy in - use vehicle power - battery internal short - circuit diagnosis algorithm, as Figure 1 shown, includes the following steps:

[0066] S1. Collect the real - time working data of the power battery.

[0067] When the battery is in the charging or discharging state, through the on - vehicle monitoring or the monitoring data uploaded by the vehicle to the cloud data platform in real - time, collect the voltage (V) of each battery cell, the total current (I) flowing into the battery, and the sampling time (t) in real - time, providing a data basis for subsequent analysis and fault diagnosis.

[0068] Classify the battery working data according to different models of the power battery and establish several sets. Each set includes the working data of several power batteries of the same model but different serial numbers. The subsequent modeling analysis and fault diagnosis algorithms are all targeted at a specific set.

[0069] S2. Construct the FFRLS combined with the UKF algorithm, identify the battery working parameters, and estimate the SOC of the battery in real - time. The algorithm process is as Figure 2 shown, and includes the following steps:

[0070] S21. Through the voltage sequence data collected in real - time, select the battery cell with the smallest standard deviation of the voltage sequence as the reference battery, and calculate the voltage curve of the reference battery;

[0071] Figure 3 Shown as the voltage curve of the 5Ω internal short - circuit resistance under the DST working condition.

[0072] S22. Establish an equivalent circuit model and a discretized equation corresponding to the equivalent circuit model.

[0073] The equivalent circuit model uses some basic circuit components to form a circuit model that simulates the working characteristics of a battery, and also establishes the relationship between the external characteristics and the internal state of the battery during operation. This method has the characteristics of small computational amount, easy parameter identification, high accuracy, etc., making the verification of the model easier to achieve. Therefore, it is widely used in practical engineering applications.

[0074] In this embodiment, the second-order RC model as shown in Figure 4 is adopted. In this model, U t represents the terminal voltage of the lithium-ion battery, U ocv represents the open-circuit voltage of the lithium-ion battery, R0 represents the ohmic resistance, which is used to describe the ohmic internal resistance inside the battery, and the RC network, i.e., the resistor-capacitor network, is used to describe the resistance generated inside the battery due to polarization reactions.

[0075] The dynamic equations of the second-order RC equivalent circuit model are summarized as follows:

[0076]

[0077] In the formula: I represents the current, which is specified as negative during discharge and positive during charge, with the unit of A;

[0078] Ut represents the terminal voltage, with the unit of V;

[0079] U ocv represents the open-circuit voltage, with the unit of V;

[0080] R0 represents the ohmic internal resistance, which simulates the change relationship between the terminal voltage and the open-circuit voltage during the charge and discharge process, with the unit of Ω;

[0081] R1 and R2 respectively represent the polarization resistances of the first and second RC networks, with the unit of Ω;

[0082] C1 and C2 respectively represent the polarization capacitances of the first and second RC networks, with the unit of F;

[0083] respectively represent the currents passing through resistors R1 and R2, with the unit of A;

[0084] U1(t) and U2(t) respectively represent the voltage values across the two RC networks, with the unit of V;

[0085] η(t) represents the Coulomb efficiency. Generally, it is considered to be equal to 1 during discharge and not greater than 1 during charge.

[0086] According to the dynamic equations, the discretized equations of the second-order RC equivalent battery model are established as follows:

[0087]

[0088] S23. Identify the online parameters of the equivalent circuit model using the FFRLS method according to the discretized equation of the equivalent circuit model.

[0089] An important basis for ensuring the accuracy of SOC estimation is the accurate identification of the relevant parameters of the second-order RC equivalent circuit model, which can obtain the key parameters of the system in real time, thereby reflecting the current operating state of the device or system, promptly detecting whether the parameter changes deviate from the normal range, and providing a basis for early fault warning.

[0090] In this embodiment, the recursive least squares method with a variable forgetting factor (FFRLS) is introduced for dynamic parameter identification, which is particularly suitable for online estimating the parameters of the equivalent circuit model. This method can adapt to the situation where system parameters change over time, and at the same time balance the importance of historical data and current data by adjusting the forgetting factor. The algorithm includes the following steps:

[0091] S231. Obtain the transfer function of the equivalent circuit model according to the dynamic equation of the equivalent circuit model:

[0092] When identifying the parameters of the established equivalent circuit model based on FFRLS, the transfer function of the established circuit model needs to be obtained. Converting the expression of the second-order RC equivalent circuit model to the least squares form of the system input and output gives:

[0093]

[0094] where y(k) = E(k), representing the system output (such as voltage, current, etc.);

[0095] θ(k) = [θ1 θ2 θ3 θ4 θ5] T , is the parameter vector to be identified;

[0096]

[0097] v k represents the measurement noise.

[0098] S232. Calculate the variable forgetting factor λ and the calculation gain matrix K0:

[0099] On the premise of ensuring accuracy, the recursive least squares algorithm can minimize the computational complexity as much as possible. The basic principle of this algorithm is to use the estimated value at the previous moment and the measured value at the current moment for calculation, and recursively obtain the estimated value at the current moment. However, as the recursive process progresses, new data will be interfered by some historical data, resulting in slow algorithm response speed and decreased calculation accuracy. Therefore, a variable forgetting factor is introduced on the basis of the RLS algorithm to reduce the influence of historical data on the recursive algorithm and improve the role of new data and calculation accuracy.

[0100] The objective function of FFRLS can be defined as the output voltage function E obtained each time the parameters are updated, and the calculation formula is as follows:

[0101]

[0102] In the formula, E(θ) is the actual output value, and E(θ,λ) is the model prediction value based on the current forgetting factor.

[0103] In the formula, λ is the forgetting factor, which is used to allocate the weights of new and old data. Usually, the forgetting factor λ of FFRLS is assigned a value between 0.9 and 0.999. The smaller the assignment, the stronger the tracking ability of the algorithm, but in this case, there may be a parameter jump phenomenon; the larger the assignment, the slower the algorithm speed; when assigned a value of 1, FFRLS degenerates into recursive least squares. P0(k) in the above formula is the covariance matrix at time k, and its value is usually assigned according to actual experience.

[0104] In order to adaptively set the forgetting factor λ during the calculation of the gain matrix, a simulated annealing algorithm is introduced to find the global random optimal solution of the forgetting factor.

[0105] The steps of the simulated annealing algorithm include:

[0106] First, set the initial forgetting factor λ0, initial temperature T0, and cooling rate α related to the calculation.

[0107] To meet the boundary requirements of FFRLS, the range of the forgetting factor randomly generated each time is restricted to the interval [0.9, 0.999]. In each iteration, the objective function and the initial value are substituted into the acceptance probability formula, and whether to accept the new solution is determined according to the current objective function value f(λ) and the objective function value f(λ′) of the new forgetting factor: if f(λ)≥f(λ′), then accept the new solution, otherwise generate a random number r in the range [0,1]. If r < λ accept then accept the new solution, otherwise retain the current solution λ. The acceptance probability formula is as follows: [[ID=2S]]

[0108]

[0109] Next, update the temperature value:

[0110] T new = α·T

[0111] At high temperatures, the probability of accepting a worse solution during the calculation process is relatively high, allowing the algorithm to explore a wider solution space. As the temperature decreases, the probability of accepting a worse solution decreases, and the algorithm tends to converge to a better solution.

[0112] When the set maximum number of iterations is reached or the temperature drops to the threshold, stop the algorithm and output the current best forgetting factor.

[0113] The calculation formula for the corresponding gain matrix \(K_0\) is as follows:

[0114]

[0115] S233. Update the covariance matrix \(P_0(k)\), and the formula is as follows:

[0116]

[0117] S234. Update the parameter to be estimated \(\theta(k)\), and the formula is as follows:

[0118]

[0119] In the formula, the initial value \(\theta(0)\) takes a real number matrix that is small enough.

[0120] S235. Return the parameter identification value at the next moment to S232, and iteratively execute S232 - S235 until the change in the covariance matrix is less than the set threshold, and the FFRLS algorithm ends.

[0121] It should be noted that in the least - squares recursive algorithm containing genetic factors, the open - circuit voltage at time \(k\) is obtained by fitting the SOC - OCV corresponding curve. According to the open - circuit voltage identification experiment, the fitting polynomials of SOC and OCV under charge and discharge states can be obtained. Further, the FFRLS algorithm can be implemented through MATLAB programming.

[0122] S24. Online estimate the battery SOC based on the unscented Kalman filter algorithm UKF.

[0123] In the actual engineering application process, most systems or models are non - linear. In order to implement filtering in a non - linear system, the unscented Kalman filter algorithm approximates the probability density distribution of the non - linear function, uses a series of deterministic samples to approximate the posterior probability density of the state, and processes the non - linear transfer problems of the mean and covariance through the UKF unscented transformation, avoiding the process of linearizing the non - linear function and not requiring approximation of the non - linear function and derivation of the Jacobian matrix, thus performing Kalman filtering more accurately. It includes the following steps:

[0124] S241. Assign initial values to the model parameters.

[0125] Before the algorithm starts, initial values should be assigned to \(x_0\), \(P_0\), \(Q\) k and \(R\) k where \(x_0\) is the initial value of the polarization capacitor voltage, \(P_0\) is the initial covariance matrix, \(Q\) k is the process noise covariance matrix, and \(R\) k is the measurement noise covariance matrix. The state variables in the battery model are composed of the state of charge SOCk and the polarization capacitor voltages U1(k) and U2(k). In MATLAB programming, the open-circuit voltage U of the battery is used ocv to determine the SOC k value. The polarization capacitor voltage is generally zero, so x0 can be determined, and the fault tolerance of x0 is relatively high; P0 is the identity matrix, and Q k and R k are taken according to experience.

[0126] S242. Determine the state equation vector and the observation vector equation of the battery system.

[0127] In order to successfully apply the UKF algorithm to the SOC estimation of lithium-ion batteries, it is necessary to determine the state vector, the observation equation, and the output vector of the controlled battery system, etc.

[0128] Select the battery SOC, and the voltages U1 and U2 of the two RC circuits as the state vectors of the system model. Take the change equation of the battery model terminal voltage as the observation equation of the system. In addition, since these variables are all affected by the current, the current is used as the input of the entire system.

[0129] According to the established equivalent circuit model and related expressions, the state equation of the discrete battery system is as follows: [[ID=2!]]

[0130]

[0131] The observation equation of the discrete battery system is as follows:

[0132] U t (k) = U OCV (SOC(k)) - R0I(k) - U1(k) - U2(k)

[0133] According to the state equation and the observation equation of the system, the state vector equation and the observation vector equation can be obtained.

[0134] The state vector equation is as follows:

[0135]

[0136] Let:

[0137]

[0138] u(k - 1) = I(k - 1)

[0139] Then the state equation of the battery is obtained as follows:

[0140] x k = Ax k-1 + Bu k-1 + wk-1

[0141] The observation vector equation is as follows:

[0142] U t (k) = U OCV (SOC(k)) - R0I(k) - U1(k) - U2(k)

[0143] Similarly, let:

[0144]

[0145] D = [-r0]

[0146] Then the state equation of the battery is obtained as follows:

[0147] Y k = cX k + dU k + V k

[0148] S243. After obtaining the state equation and the observation equation of the battery system, the optimal estimation of the battery SOC is realized according to the recursive steps of the UKF algorithm.

[0149] For a nonlinear discrete system, its state equation and observation equation are as follows:

[0150]

[0151] where the nonlinear functions F(X k , U k ) and g(x k , u k ) are differentiable at all sampling points. X k is the system state variable at time k, u k represents the system input at time k, Z k represents the system output at time k, w k is the system process noise, v k is the system observation noise, w k , v k are independent Gaussian white noises, and the covariance matrices of the process noise and the observation noise are Q and R respectively. UKF obtains the mean and covariance P of the current system state to generate sigma points, and obtains the mean and covariance of the predicted state through weighted averaging of the sigma points and updates the state estimate and covariance matrix using the Kalman gain. The specific steps are as follows:

[0152] (1) Obtain a set of sampling points (referred to as the Sigma point set) using the unscented transformation formula:

[0153]

[0154] The weight of each sampling point is:

[0155]

[0156] (2) Calculate the prediction of the 2n + 1 Sigma point set:

[0157] X (i) (k + 1|k) = f[k, x (i) (K|k)], i = 1 to 2n + 1

[0158] (3) Calculate the prediction of the system state quantity and the covariance matrix:

[0159]

[0160] (4) Generate a new Sigma point set using the unscented transform according to the prediction result:

[0161]

[0162] (5) Substitute the predicted Sigma point set into the observation equation to obtain the predicted observed quantity:

[0163] Z (i) (k + 1|k) = h[X (i) (k + 1|k)], i = 1 to 2n + 1

[0164] (6) Obtain the mean value and covariance of the system prediction through weighted summation of the observation prediction values:

[0165]

[0166] (7) Calculate the cross covariance and the Kalman gain:

[0167]

[0168] (8) Calculate the state update and covariance update of the system:

[0169]

[0170] P(k + 1|k + 1) = P(k + 1|k) - K(k + 1)P xz K T (k + 1) The SOC estimation result under dynamic conditions according to the UKF algorithm is as Figure 5 shown.

[0171] S3. Construct a fault feature function for the difference in the SOC curves of each battery cell and the reference battery.

[0172] Construct a fault feature function by utilizing the explosive growth characteristics of the exponential function to amplify the small differences in the SOC curves between each battery cell and the reference battery, which helps to better identify potential internal short - circuit faults. The SOC value does not follow sudden changes such as current and has less fluctuation compared to voltage or current parameters, which can effectively reduce the false alarm probability. Specifically:

[0173] Define the data points from t0 - t e as

[0174]

[0175] where n is the number of voltage data collected by the sensor. Calculate the mean value at each moment through the data points as:

[0176]

[0177] Then the standard deviation of the data points at each moment is:

[0178]

[0179] Sort the data points at each moment from largest to smallest and define them as After that, calculate the median of the voltage data at each moment:

[0180]

[0181] Then define the reference data at each moment as:

[0182] and

[0183] where n sel is the number of voltage data points that satisfy the inequality.

[0184] Then the fault feature function is:

[0185]

[0186] S4. Substitute the identification data of each battery into the fault feature function for internal short - circuit fault detection. It includes:

[0187] S41. Calculate the discrete Fréchet distance between the SOC curves of each battery and the reference battery, and generate a monomer curve graph that changes with time.

[0188] The Fréchet distance is a quantization method for path - space similarity. It considers the time - series space between curves and quantifies the curve similarity by calculating the distance between two curves. Therefore, it is used for fault feature extraction.

[0189] Suppose the curve consisting of p data points, the curve consisting of q data points, passing through curve X p and curve Y q a sequence - pair data set L can be obtained, and the data set is expressed as:

[0190]

[0191] where a1 = 1, b1 = 1, a m = p, b m = q, for any i = 1, 2, …, m, there is a i+1 = a i or a i+1 = a i + 1, and b i+1 = b i . The curves X p and Y q are defined by the sequence - pair data set L, and the distance between them is

[0192]

[0193] where is the Euclidean distance of the sequence - pair, and ||L(X p , Y q )|| is the maximum value of the Euclidean distance of the sequence - pair.

[0194] The discrete Fréchet distance is defined as:

[0195] DFD(X p , Y q ) = min||L(X p , Y q )||

[0196] Then the discrete Fréchet distance between the data point and the mean voltage is

[0197]

[0198] To diagnose faults in real - time, a forgetting mechanism based on a sliding window is used to calculate the discrete Fréchet distance at each sampling moment.

[0199] S42. Determine the deformation abnormality detection reference curve according to the single - cell curve graphs of all batteries.

[0200] At the data level, abnormal single - entity feature curves occur from time to time. To detect the abnormal fluctuations of the feature curves, a functional data outlier detection algorithm based on deformation abnormality is used for determination. The deformation abnormality detection curve reflects the degree to which its shape deviates from the normal curve. The greater the deviation value, the more obvious the abnormality of the curve.

[0201] When selecting the reference normal curve, to reduce the influence of outliers, the curves are sorted according to the depth values, and only the part with larger depth values is retained for average calculation. The average curve obtained in this way is less affected by outliers and can better represent the curve fluctuations of normal data. The specific steps are as follows:

[0202] (1) Calculate the MBD, MEI, and parabolic value P of the i - th curve i :

[0203] MBD mainly measures the frequency of curve i in the band - like region formed by other curves. The main formula is:

[0204]

[0205] where n is the total number of curves, m is the number of time points, and usually m = [n / 2]+1. Considering all pairs of curves to form bands, calculate the proportion of the target curve included in the band - like region formed by the minimum and maximum values at each time point. I(·) is the indicator function. If x i (t) satisfies min(t) ≤ x i (t k ) ≤ max(t) for all t, then the count is incremented by 1. After calculating all the proportions, take the average of all curve pairs.

[0206] MEI mainly combines the positional relationship between the upper and lower envelopes of the curve and reflects the degree of the curve located in the upper part of the data set. The main formula is:

[0207]

[0208] Calculate the frequency of curve x i exceeding other curves x j at each time point, and then take the average of all curves and time points. For the convenience of subsequent calculations, the calculations of MBD and MEI are described as:

[0209]

[0210] The formula for the parabolic value is:

[0211] (2) Determine the reference curve:

[0212] For x1,...,x nFind the derivative curve \(x'(t)\). If the data is discrete, perform fitting.

[0213] Sort \(x'(t)\) from large to small according to the statistical depth value to obtain \(x' (1) ,..., \(x' (n) 。

[0214] After removing the [an] (the largest integer not greater than an) curves with the smallest depth value, calculate the mean to obtain the reference curve \(x' mean (t):

[0215]

[0216] S43. Calculate the deformation abnormality degree of the SOC characteristic curve of each battery cell, and then combine the deformation abnormality degree with the depth value MBD to establish a depth point. Change the vertical distance from the original depth point to the parabola and compare it with the reference curve to finally identify the abnormal value. The specific steps are as follows:

[0217] (1) Calculate the shape fluctuation deviation degree \(s i :

[0218] s i =∫ I (x' i =x' mean )ω(t)dt;

[0219] (2) Normalize the deviation degree \(s i , and calculate the deformation abnormality degree \(S i :

[0220]

[0221] (3) Determine the deformation abnormality degree coefficient \(β\), and add the deformation abnormality degree to \(mb i to obtain \(Amb i :

[0222]

[0223] Amb i =mb i -βs i .

[0224] (4) Calculate the vertical distance \(d i from the point \((MEL, MBD)\) to the parabola, and calculate the cutoff:

[0225] D i =p i -amb i , \(i = 1,\ldots,n

[0226] Among them, MEL represents the horizontal position of the measurement point; MBD represents the depth value of the measurement point.

[0227] (5) Calculate d i 's upper quartile q d3 and interquartile range IQR d , as parameters for subsequent calculation of the abnormal boundary cutoff:

[0228] cutoff = q d3 + 1.5 × IQR d

[0229] (6) Identify outliers based on the abnormal boundary cutoff:

[0230] SO = {i|d i ≥ cutoff}

[0231] When the deformation abnormality degree of a certain battery relative to the SOC curve of the reference battery exceeds the abnormal boundary cutoff, it is determined that the battery has an internal short - circuit fault, as Figure 6 shown. Generate an alarm message for the battery cell determined to have an internal short - circuit fault.

[0232] The above are only embodiments of the present invention. The invention is not limited to the fields involved in this embodiment case. Common knowledge such as specific structures and characteristics known in the solution is not described in detail here. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can still be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be based on the content of its claims, and the specific implementation manners described in the specification can be used to interpret the content of the claims.

Claims

1. A method for diagnosing internal short circuit in a power battery of a new energy vehicle in use, characterized in that: It includes the following steps: S1. Collect the real-time working data of the power battery; S2. Construct the FFRLS combined with the UKF algorithm to identify the battery working parameters and estimate the SOC of the battery in real time; S3. Construct a fault feature function for the difference in the SOC curves of each battery and the reference battery; S4. Substitute the identification data of each battery into the fault feature function to perform internal short-circuit fault detection.

2. A method for diagnosing internal short circuit in a power battery of a new energy vehicle in use according to claim 1, characterized in that: The S2 includes: S21. Through the voltage sequence data collected in real time, select the single cell with the smallest standard deviation of the voltage sequence as the reference battery, and calculate the voltage curve of the reference battery; S22. Establish an equivalent circuit model and a discretized equation corresponding to the equivalent circuit model; S23. According to the discretized equation of the equivalent circuit model, use the FFRLS method with a variable forgetting factor to identify the online parameters of the equivalent circuit model; S24. Based on the unscented Kalman filter algorithm UKF, estimate the battery SOC online.

3. A method for diagnosing internal short circuit in a power battery of a new energy vehicle in use according to claim 2, characterized in that: The S23 includes: S231. Obtain the transfer function of the equivalent circuit model according to the dynamic equation of the equivalent circuit model; S232. Calculate the variable forgetting factor λ and the gain matrix K0; S233. Update the covariance matrix P0(k); S234. Update the parameter to be estimated θ(k); S235. Return the parameter identification value at the next moment to S232, and iteratively execute S232 to S235. When the change in the covariance matrix is less than the set threshold, end the FFRLS algorithm.

4. A method for diagnosing internal short circuit in a power battery of a new energy vehicle in use according to claim 3, characterized in that: The S232 uses the simulated annealing algorithm to solve the optimal forgetting factor.

5. A method for diagnosing internal short circuit in a power battery of a new energy vehicle in use according to claim 4, characterized in that: The S24 includes: S241. Assign initial values to the model parameters; S242. Determine the state equation vector and the observation vector equation of the battery system; S243. After obtaining the state equation and the observation equation of the battery system, realize the optimal estimation of the battery SOC according to the recursive steps of the UKF algorithm.

6. A method for diagnosing internal short circuit in a power battery of a new energy vehicle in use according to claim 1, characterized in that: In the step S3, the data points at time t0 - t e are defined as Wherein, n is the number of voltage data collected by the sensor; Calculate the mean value at each moment through data points It is: Then the standard deviation of each data point at each moment is: Sort the data points at each moment from largest to smallest and define them as After that, calculate the median of the voltage data at each moment: Then the reference data at each moment is defined as: where n sel is the number of voltage data points that satisfy the inequality; Then the fault feature function is as follows:

7. A method for diagnosing internal short circuit in a power battery of a new energy vehicle in use according to claim 1, characterized in that: The S4 uses a functional data outlier detection algorithm based on the deformation abnormality degree to determine the internal short-circuit fault of the power battery.

8. A method for diagnosing internal short circuit in a power battery of a new energy vehicle in use according to claim 7, characterized in that: The S4 includes: S41. Calculate the discrete Fréchet distance of the SOC curves between each battery and the reference battery, and generate a single-cell curve graph that changes with time; S42. Determine the deformation abnormality detection reference curve according to the single-cell curve graphs of all batteries; S43. Calculate the deformation abnormality degree of the SOC characteristic curve of each battery cell, then combine the deformation abnormality degree with the depth value MBD, change the vertical distance from the original depth point to the parabola, and compare with the reference curve to finally identify the outlier.

Images