Power lithium battery SOH and SOC joint estimation method based on data and model fusion

Abstract

The invention discloses a power lithium battery SOH and SOC joint estimation method based on data and model fusion, and the method comprises the steps: carrying out the modeling of a second-order RC equivalent circuit model of a lithium battery, so as to simulate the dynamic behaviors of the lithium battery under different charging and discharging conditions; the method comprises the following steps of: constructing a hybrid model fusing a convolutional neural network and a transformer by combining a multi-head differential attention mechanism under a macro scale, and inputting a plurality of extracted battery health factors into the model to obtain an SOH estimated value; and under the microscopic scale, based on a battery equivalent model, SOC estimation is carried out by using a weighted multi-information adaptive unscented Kalman filtering algorithm based on dynamic noise, and the SOC estimation process is corrected through the obtained SOH estimation value, so that an SOC estimation value is obtained. According to the method, the coupling relation between the SOH and the SOC is considered, and multi-time scale estimation is adopted, so that the calculation amount of the BMS can be effectively reduced, and the accuracy and robustness of the joint estimation of the SOH and the SOC of the power lithium battery can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion batteries, and in particular to a method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion. Background Art

[0002] Lithium-ion batteries have the advantages of low cost, high energy density and long cycle life. Against the backdrop of the intensifying global energy and environmental crisis, they are gradually becoming an important energy storage device for electric vehicles. The electrochemical reactions inside power lithium batteries are highly nonlinear, environmentally sensitive and age-degraded, making it difficult to accurately evaluate the various states of the system, which is also the bottleneck of current battery energy storage technology. Therefore, the study of the joint estimation of SOH and SOC of power lithium batteries has important theoretical significance and practical application value. SOH is defined as the ratio of the current maximum available capacity to the nominal capacity, which is a description of the current state on a long time scale. Its accurate estimation is conducive to the health diagnosis of the battery and the timely replacement of aging batteries. SOC is considered to be the ratio of the current charge capacity of the battery to the maximum available capacity, which is a state change of the battery on a short time scale. Real-time SOC estimation can predict the system operation time and formulate a reasonable charging and discharging strategy. Both ensure the smooth operation of the system from different aspects.

[0003] At present, the estimation methods of SOH and SOC are mainly concentrated in two directions: model-driven and data-driven. Model-based SOH estimation methods mainly include electrochemical models and empirical degradation models. The electrochemical model describes the internal working mechanism of the battery in more detail. It describes the physical and chemical mechanism of the battery's capacity decay by establishing a series of partial differential equations. However, the electrochemical model parameter identification is difficult and is not suitable for online estimation of the BMS system. The empirical degradation model can model the capacity decay trend of the battery throughout its entire cycle. The parameter identification is simple, but it is difficult to adapt to the different capacity decay trends caused by individual differences in the battery. The data-driven SOH estimation method does not need to analyze the internal mechanism of the battery. It extracts and analyzes the external health factors (HF) that are closely related to the battery capacity decay, and uses a machine learning algorithm to establish a nonlinear mapping relationship between HF and the battery SOH, avoiding physical modeling and parameter identification problems. It is more flexible and widely used.

[0004] The model-based SOC estimation method needs to establish an equivalent circuit model, such as the Rint model, Thevenin model, PNGV model, etc., to simulate the external working state of the battery, and combine the filtering algorithm to perform closed-loop SOC estimation. However, as the battery ages, the model parameter identification value will produce errors, and the current available capacity or health status will have a significant impact on the SOC estimation result, so it is not suitable to perform SOC estimation alone. The data-driven SOC estimation method uses an algorithm to learn the mapping relationship between measurable values ​​such as voltage, current, and temperature and SOC. The training and calculation amount is large, and it is not easy to apply online.

[0005] In summary, how to achieve high-precision and strong convergence joint estimation of SOH and SOC of power lithium batteries is a technical problem that needs to be solved urgently. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion. Firstly, a hybrid model integrating a multi-head differential attention mechanism (MHDA), a convolutional neural network (CNN) and an inverted transformer (iTransformer) is trained as a SOH state estimator, and a health factor is input to estimate the SOH state. Next, based on the second-order RC equivalent circuit model of the lithium-ion power battery, a variable forgetting factor recursive least squares (VFFRLS) method is used for parameter identification, and a weighted multi-innovation and dynamic noise adaptive unscented Kalman filter (WMI-DNAUKF) algorithm based on dynamic noise is used to construct a SOC state estimator. Finally, the established SOH state estimator and SOC state estimator are coupled to realize joint estimation of SOH and SOC based on multiple time scales.

[0007] In order to achieve the above-mentioned object of the invention, the present invention provides a method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion, the method comprising the following steps:

[0008] S1: Extract health factors from the battery aging dataset and input them into a hybrid model that combines MHDA, CNN, and iTransformer to obtain the SOH estimate;

[0009] S2: Construct a second-order RC equivalent circuit model of the power lithium battery and perform parameter identification. Use the WMI-DNAUKF algorithm to estimate the battery SOC for the battery current and voltage data, and use the SOH estimate of S1 to correct the battery capacity to obtain the SOC estimate.

[0010] S3: At the macro scale, the SOH obtained by the estimator is updated to the state space equation of S2, and the scale is switched to the micro scale to estimate the SOC. After the charge and discharge cycle reaches the set threshold, the macro scale is switched to estimate the SOH, and the cycle is repeated to achieve the joint estimation of SOH and SOC.

[0011] Furthermore, the specific steps of establishing the SOH state estimator in step S1 are:

[0012] S11: Based on the battery aging data set, the battery SOH is calculated as the true value, and the time interval HF of the equal discharge voltage difference interval [U1, U2] is selected t and the curve area HF area As a health factor, the correlation between the feature and the label is determined by calculating the Pearson correlation coefficient;

[0013] S12: Use two 1D convolutional layers to extract local features and identify deep features that may be ignored or difficult to capture by iTransformer;

[0014] S13: Treat each deep feature as an independent variable feature and embed the time information into the time series data as an independent dimension;

[0015] S14: Apply MHDA on the variable dimension to enhance the focus on relevant information and reduce the interference of irrelevant information.

[0016] Furthermore, the specific steps of establishing the SOC state estimator in step S2 are:

[0017] S21: Construct the discretized state space equation and observation equation expressions of the second-order RC equivalent circuit model and determine the parameters of the equivalent circuit model;

[0018] S22: Correct the battery capacity C in the state space expression according to the SOH estimate of S1 n ;

[0019] S23: Establish a weighted multi-innovation unscented Kalman filter. The weights and weighting matrix are calculated as shown in formula (1).

[0020]

[0021] In formula (1), ε k Represents the system error, Wi is the error-based weight matrix;

[0022] S24: Introduce dynamic noise weight ρ and process noise Q into the weighted multi-innovation unscented Kalman filter established in S23 k and measurement noise R k The calculation of is shown in formula (2):

[0023]

[0024] In formula (2), F k represents the approximation of the real-time covariance of the innovation, ε k is the voltage information of the battery model at any time, M is the window size of the covariance, K x,k represents the system gain at the current moment, is the weight coefficient of the observed covariance, Y i,k|k-1 is the predicted value of the observed quantity, is the estimated value of the observed quantity.

[0025] Furthermore, the specific steps of switching the time scale of the battery state estimator in step S3 are:

[0026] S31: At the macro scale, the extracted health factor is input into the SOH estimator to perform aging capacity update, the SOH is estimated, and the macro time step is updated to S+1;

[0027] S32: At the micro scale, the SOC estimator is used to update the SOC in real time, and the number of charge and discharge cycles is counted. After each cycle, the micro time step is updated to H+1. When the scale switching setting threshold is reached, return to step S31.

[0028] The beneficial effects of the present invention are:

[0029] When estimating SOH, CNN with powerful feature extraction capabilities and iTransformer with global perception capabilities are combined, and MHDA is used to improve the model's attention to relevant information and reduce the interference of irrelevant information. By embedding time information as an independent dimension, the model's understanding of time can be enhanced, and a more comprehensive hybrid model can be generated, thereby improving the accuracy of SOH estimation.

[0030] When estimating SOC, the unscented Kalman filter algorithm, the dynamic noise adaptive filter algorithm and the weighted multi-information recognition theory are combined. The dynamic noise weight is used to maintain historical information and adjust the statistical characteristics of the noise in real time. The multi-information error weight makes the new information with larger errors account for a larger proportion in the correction process, thereby greatly improving the accuracy of the unscented Kalman filter algorithm in estimating SOC. When the scale switching threshold is reached, the value output by the SOH estimator is used to correct the model parameters used by the SOC estimator. Effective correction of the errors caused by aging of the model parameters can effectively improve the accuracy of the model-based WMI-DNAUKF algorithm throughout the battery life while reducing the amount of algorithm calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0032] Figure 1 It is a flow chart of the method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion proposed by the present invention;

[0033] Figure 2 This is a schematic diagram of the structure of the SOH estimation model used in the present invention;

[0034] Figure 3 A schematic diagram of the flow of the SOC estimation algorithm used in the present invention;

[0035] Figure 4 This is a schematic flow chart of the joint estimation method used in the present invention;

[0036] Figure 5 It is a schematic diagram of SOH estimation results;

[0037] Figure 6 Schematic diagram of SOC estimation results. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages of the embodiment of the present invention clearer, the technical scheme in the embodiment of the present invention will be clearly and completely described in conjunction with the drawings in the embodiment of the present invention. It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as those commonly understood by ordinary technicians in the technical field to which the present invention belongs.

[0039] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0040] The specific embodiments of the present invention will be further described below in conjunction with the accompanying drawings. Figure 1 The flowchart of the method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion proposed by the present invention is shown, including the following steps S1-S3:

[0041] S1: Figure 2 The schematic diagram of the SOH estimation model structure used in the present invention is shown. The health factor is extracted from the battery aging data set and input into the hybrid model integrating MHDA, CNN and iTransformer to obtain the SOH estimation value. The detailed steps are as follows:

[0042] S11: Calculate the battery SOH as the true value based on the battery aging data set. Select the time interval HF of the equal discharge voltage difference interval [U1, U2] t and the curve area HF area As a health factor, the maximum value U2 and the minimum value U1 of the equal discharge voltage sampling are 3.9V and 3.6V respectively, HF t Defined as formula (1):

[0043] HF t =|t U2 -t U1 |,i=1,2,…,k (1)

[0044] In formula (1), k is the number of cycles, t U2 and t U1 They are the upper and lower limit time of voltage sampling respectively. area Defined as formula (2):

[0045]

[0046] In formula (2), n represents the number of data points between 3.9 V and 3.6 V, k is the number of cycles, and U j and U j+1 are two adjacent time points t j ,t j+1 Voltage value.

[0047] The correlation between the feature and the tag is determined by Pearson correlation analysis. Formula (3) is the battery time interval HF t The Pearson correlation coefficient between SOH and the discharge curve area HF of the battery is given by formula (4). area Pearson correlation coefficient between θ and SOH:

[0048]

[0049] In formulas (3) and (4), n represents the number of cycles of the lithium-ion battery; HF t,i HF area,i represents the value of the health factor at the i-th cycle; It represents the mean value of health factor under all cycles.

[0050] S12: Use two one-dimensional convolutional layers to extract local features and identify deep features that may be ignored or difficult to capture by iTransformer. Formula (5) is the calculation process of the output feature of the nth layer convolution operation,

[0051]

[0052] In formula (5), relu(·) represents the activation function, and They represent the bias and weight of the nth layer features, respectively. v and x represent the filter index in the convolution and the output vector of the previous step, respectively.

[0053] S13: Treat each deep feature as an independent variable feature and embed the time information into the time series data as an independent dimension.

[0054] The data of dimension (B, T, N) is input into the fully connected layer, and the data of each time step is mapped to a higher-dimensional feature space. The output dimension is (B, T, D), where B is the batch size, that is, the number of samples input into the model at one time; N is the feature dimension, that is, the number of variables contained in each time step; T is the number of time steps, that is, the length of the time series or the size of the observation window, and D is the dimension in which the data is embedded.

[0055] The Rearrange function is used to rearrange the dimensions of the data into the form of (B, D, T), which can better integrate the information of variables and time steps so that the subsequent self-attention mechanism can better capture the relationship between variables in time series data.

[0056] S14: Apply MHDA on the variable dimension to enhance the focus on relevant information and reduce the interference of irrelevant information.

[0057] The key to the differential attention mechanism is to use a pair of softmax functions to eliminate the noise of the attention scores. Given the input X, formula (6) is the process of projecting them into query matrix, key matrix and value matrix Q1, Q2, K1, K2, V. Formula (7) is the process of calculating the output of the differential attention operator DiffAttn(·):

[0058] [Q1;Q2] = XW Q , [K1; K2] = XW K , V = XW V (6)

[0059]

[0060] In formula (6) and (7), W Q , W K , W V is a parameter, d k is represented by the number of dimensions of the key matrix, and λ is a learnable scalar. In order to synchronize the learning dynamics, the scalar λ is reparameterized as shown in formula (8),

[0061] λ=exp(λ q1 ·λ k1 )-exp(λ q2 ·λ k2 )+λ init (8)

[0062] In formula (8), λ q1 , k1 , q2 , k2 is a learnable vector, λ init ∈(0,1) is a constant used to initialize λ.

[0063] S2: Construct the second-order RC equivalent circuit model of the power lithium battery and perform parameter identification. Figure 3 The flowchart of the SOC estimation algorithm used in the present invention is shown. The WMI-DNAUKF algorithm is used to estimate the battery SOC for the battery current and voltage data, and the battery capacity is corrected using the SOH estimation value of S1. The detailed steps for obtaining the SOC estimation value are as follows:

[0064] S21: Combining the battery circuit model and the ampere-hour integral calculation formula of the state of charge, the discretized state space equation and observation equation expressions of the second-order RC equivalent circuit model are obtained, as shown in formula (9):

[0065]

[0066] In formula (9), Δt represents the sampling time interval, τ is the time constant, τ1=R1C1, τ2=R1C1, η is the battery coulomb efficiency, C n is the rated capacity, w k-1 and v k-1 are system state noise and measurement noise, respectively, both are zero-mean Gaussian white noise;

[0067] Each time new observation data is obtained, the gain vector K(k) is calculated using the VFFRLS formula, and then the parameters to be identified are updated. and the covariance matrix P(k) of the state estimate, the model parameter vector θ(k) to be identified includes R0, R1, R2, C1, C2 and U OCV The expression of VFFRLS is formula (10),

[0068]

[0069] In formula (10), ρ is the sensitivity factor, e(k) is the estimation error at time k, λ(k) is the variable forgetting factor, K(k) is the least squares gain, and P(k) is the covariance matrix.

[0070] S22: Correct the battery capacity C in the state space expression according to the SOH estimate of S1 n .

[0071] S23: Establish a weighted multi-innovation unscented Kalman filter based on the state space equation and observation equation of S21. The detailed steps are as follows:

[0072] Set the initial value of the state variable and use it as the posterior estimate of the state at time 0.

[0073]

[0074] Formula (11) is the initial value of the covariance of the state variable, and formula (12) is the initial value of the covariance of the state estimate.

[0075] Construct the sigma point X at time k-1 based on the posterior estimate of the state at time k-1 i and weights

[0076] In equations (13) and (14), n represents the dimension of the extended state variables, which include polarization voltages U1, U2 and battery SOC. Here, n is 3. is the scale adjustment coefficient, and its size is modified to reduce the prediction error. α is the dispersion factor, which is generally a smaller number between 0 and 1. β is the state distribution parameter.

[0077] Based on the state equation of the state space expression described in S21, determine the prior estimate of the state at time k And the prior estimate of the covariance matrix P xx,k|k-1 ,

[0078]

[0079] In formula (16), Q k-1 is the system noise w k-1 The covariance matrix of .

[0080] According to the sigma point and weight at time k, and based on the observation equation of the state space expression described in S21, the predicted value of the observation at time k is determined and the observed covariance P yy,k|k-1 ,

[0081]

[0082]

[0083] In formula (18), R k-1 is the system noise v k-1 The covariance matrix of .

[0084] According to the sigma point and state prior estimate at time k Observed value predicted value Y i,k|k-1 And the estimated value of the observation Determine the covariance matrix P of the state variables and the observed quantities xy,k|k-1 and the Kalman gain K x,k ,

[0085]

[0086] K x,k =P xy,k|k-1 (P yy,k|k-1 ) -1 (20)

[0087] The actual measured value at time k minus the predicted value is taken as the estimated error ε of the observed value at time k k , using the estimated error of the observation at each moment as a single new information to expand the new information matrix E l,k and Kalman gain K l,k Formula (21) is the extended new information matrix, and formula (22) is the extended Kalman gain matrix.

[0088]

[0089] K l,k =[K x,k K x,k-1… K x,k-l+1 ] (twenty two)

[0090] In formulas (21) and (22), l is the length of the new information, ε k Represents the new information error at the current moment, K x,k Indicates the system gain at the current moment.

[0091] The innovation errors at different times are different, and the large error should account for a large proportion in the correction process. The weight coefficient of the innovation is calculated, and the weighted multi-innovation matrix is ​​constructed. Formula (23) is the process of constructing the weighted multi-innovation matrix.

[0092]

[0093] In formula (23), w i is the weight coefficient of the current new information, W i is the weighted multi-innovation matrix.

[0094] According to the expanded gain K l,k and Xinxi E l,k Update the state and covariance matrix to obtain the estimated result and error of SOC. Formula (24) is the state updated at time k, and formula (25) is the covariance matrix updated at time k.

[0095]

[0096] S24: Based on the weighted multi-innovation unscented Kalman filter established in S23, an adaptive filtering algorithm is added to the nonlinear discrete system model, and a dynamic noise weight ρ is introduced to maintain historical information and adjust the statistical characteristics of the noise in real time. The process noise Q k and measurement noise R k The calculation of is shown in formula (26),

[0097]

[0098] In formula (26), F k represents the approximation of the real-time covariance of the innovation, ε i is the voltage information of the battery model at any time, and M is the window size of the covariance.

[0099] S3: At the macro scale, the SOH obtained by the estimator is updated to the state space equation of S2, and the scale is switched to the micro scale to estimate the SOC. After the charge and discharge cycle reaches the set threshold, the macro scale is switched to estimate the SOH, and the cycle is repeated to achieve the joint estimation of SOH and SOC. Figure 4 The schematic diagram of the joint estimation method used in the present invention is shown, and the detailed steps are as follows:

[0100] S31: At the macro scale, the extracted health factor is used as input, and the SOH estimator based on CNN-iTransformer-MHDA performs aging capacity update to estimate SOH. The macro time step is updated to S+1. Formula (27) is the aging capacity update process.

[0101]

[0102] S32: At the micro scale, the SOC estimator based on WMI-DNAUKF updates the SOC in real time, counts the number of charge and discharge cycles, and updates the micro time step to H+1 for each cycle. When the scale switching threshold is reached, it returns to step S31. Formula (28) is the SOC update process.

[0103]

[0104] In formula (28), H is the number of charge and discharge cycles in the joint estimation, and S is the SOH update frequency in the joint estimation.

[0105] The above article explains in detail the estimation process of the SOH estimator and the SOC estimator. The principle of the joint estimation of SOH and SOC of power lithium batteries based on data and model fusion is to achieve alternating updates of state variables and available capacity. The SOH estimator and the SOC estimator have an updating effect on different model parameters at both macro and micro time scales. Based on the corrected model, WMI-DNAUKF estimates SOC more accurately. In addition, the combination of the two estimators completely decouples the relationship between SOH and SOC, avoiding the coupling misalignment and error divergence problems that may occur when SOH and SOC are estimated simultaneously in one estimator.

[0106] The estimation accuracy of the proposed method is verified using the NASA random-step battery dataset.

[0107] The accuracy of the proposed algorithm is mainly reflected in two aspects: one is the estimation accuracy of SOH, and the other is the estimation accuracy of SOC.

[0108] In order to further explore the accuracy of the SOH estimator, the model proposed in the present invention is compared with the traditional model Transformer. Figure 5 The schematic diagram of SOH estimation results is shown, and Table 1 shows the comparison of SOH estimation errors.

[0109] Table 1 Comparison of SOH estimation errors

[0110]

[0111] from Figure 5As can be seen from Table 1, the accuracy of the CNN-iTransformer-MHDA model proposed in the present invention is higher than that of the traditional Transformer model, which verifies the effectiveness of the model proposed in the present invention.

[0112] To further explore the accuracy of the SOC estimator, when SOH = 0.9, the proposed algorithm is compared with the traditional Multi-Innovation Adaptive Unscented Kalman Filter (MI-AUKF) algorithm, and the effect of using the proposed algorithm alone to estimate SOC is also compared. The scale switching threshold is set to 1500, the dynamic noise weight parameter is 0.6, and the innovation length is 5. To test the convergence of the algorithm, the initial SOC value is set to 0.9. Figure 6 The schematic diagram of SOC estimation results is shown, and Table 2 shows the comparison of SOC estimation errors.

[0113] Table 2 Comparison of SOC estimation errors

[0114]

[0115] from Figure 6 As shown in Table 2, the accuracy of the joint estimation of SOC is higher than that of the single estimation of SOC, and compared with the traditional MI-AUKF algorithm, the error value of the WMI-DNAUKF algorithm is smaller, which verifies that the algorithm proposed in the present invention has high feasibility and superiority.

[0116] The above description is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept, which all belong to the protection scope of the present invention.

Claims

1. A method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion, characterized in that: include: S1: Extract health factors from the battery aging dataset and input a hybrid model that combines multi-head differential attention mechanism, convolutional neural network and transposed Transformer to obtain the SOH estimate; S2: Construct a second-order RC equivalent circuit model of the power lithium battery and perform parameter identification. Use a weighted multi-innovation adaptive unscented Kalman filter algorithm based on dynamic noise to estimate the battery SOC for the battery current and voltage data. Use the SOH estimate of S1 to correct the battery capacity and obtain the SOC estimate. S3: At the macro scale, the SOH obtained by the estimator is updated to the state space equation of S2, and the scale is switched to the micro scale to estimate the SOC. After the charge and discharge cycle reaches the set threshold, the macro scale is switched to estimate the SOH, and the cycle is repeated to achieve the joint estimation of SOH and SOC.

2. The method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11: Based on the battery aging data set, the battery SOH is calculated as the true value, and the time interval HF of the equal discharge voltage difference interval [U1, U2] is selected t and the curve area HF area As a health factor, the correlation between the feature and the label is determined by calculating the Pearson correlation coefficient; S12: Use two 1D convolutional layers to extract local features and identify deep features that may be ignored or difficult to capture by the transposed Transformer; S13: Treat each deep feature as an independent variable feature and embed the time information into the time series data as an independent dimension; S14: Apply a multi-head differential attention mechanism on the variable dimension to enhance the focus on relevant information and reduce the interference of irrelevant information.

3. The method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21: Construct the discretized state space equation and observation equation expressions of the second-order RC equivalent circuit model and determine the parameters of the equivalent circuit model; S22: Correct the battery capacity C in the state space expression according to the SOH estimate of S1 n ; S23: Establish a weighted multi-innovation unscented Kalman filter. The weights and weighting matrix are calculated as shown in formula (1). In formula (1), ε k Represents the system error, W i is the error-based weight matrix; S24: Introduce dynamic noise weight ρ and process noise Q into the weighted multi-innovation unscented Kalman filter established in S23 k and measurement noise R k The calculation of is shown in formula (2): In formula (2), F k represents the approximation of the real-time covariance of the innovation, ε k is the voltage information of the battery model at any time, M is the window size of the covariance, K x,k represents the system gain at the current moment, is the weight coefficient of the observed covariance, Y i,k|k-1 is the predicted value of the observed quantity, is the estimated value of the observed quantity.

4. The method for jointly estimating SOH and SOC of a power lithium battery based on data and model fusion according to claim 1, characterized in that: Step S3 specifically includes the following steps: S31: At the macro scale, the extracted health factor is input into the SOH estimator to perform aging capacity update, the SOH is estimated, and the macro time step is updated to S+1; S32: At the micro scale, the SOC estimator is used to update the SOC in real time, and the number of charge and discharge cycles is counted. After each cycle, the micro time step is updated to H+1. When the scale switching setting threshold is reached, return to step S31.

Images