High-voltage battery pack state estimation combined fault diagnosis method considering internal resistance

Abstract

The invention discloses a high-voltage battery pack state estimation combined fault diagnosis method considering internal resistance, and the method comprises the steps: firstly obtaining micro internal resistance data from a high-voltage battery pack end, and enabling the micro internal resistance data of a high-voltage battery pack and the long-term operation data of the high-voltage battery pack to form a high-voltage battery pack original data set; secondly, performing correlation analysis on data in the original data set of the high-voltage battery pack; and then building an SOH estimation network, and inputting the data obtained by the correlation analysis into the SOH estimation network to obtain an SOH estimation value. Finally, a fault diagnosis network is built, the high-voltage battery pack micro-internal resistance, the battery SOC and the battery SOH estimated value serve as input, and a fault diagnosis result is output. The high-voltage battery pack joint SOH accurate estimation and fault diagnosis under multiple scenes can be achieved, the battery state monitoring and prediction efficiency can be improved, and the fault diagnosis efficiency is improved. And the service life and the operation safety of the battery can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power battery management, and specifically refers to a high-voltage battery pack state estimation and fault diagnosis method taking internal resistance into consideration. Background Art

[0002] In actual operation, the performance of 400V / 800V high-voltage battery packs for electric vehicles and 2000V / 2500V high-voltage battery packs for energy storage power stations will gradually degrade with the extension of usage time, which is manifested as capacity attenuation, increased internal resistance, and decreased charging and discharging efficiency. This aging process may lead to operational safety issues of high-voltage battery packs. Therefore, it is of great significance to estimate and diagnose the SOH (state of health) of high-voltage battery packs for electric vehicles / energy storage power stations, which can not only help users understand the battery status and optimize the battery pack operation strategy, but also improve the overall safety and reliability of electric vehicles / energy storage power stations. At present, the research on state estimation of high-voltage battery packs mainly focuses on two types of methods: model-based methods and data-driven methods. Model-based methods attempt to estimate SOH and fault diagnosis by establishing a physical model that reflects the battery degradation process. However, the internal reaction mechanism of the battery is complex and the environmental conditions are changeable, which makes it difficult to accurately obtain many parameters in the model, and the model is difficult to fully reflect the behavior characteristics of the battery under actual conditions, thus limiting its practical application.

[0003] In contrast, data-driven methods learn the mapping relationship between input and output by analyzing battery operation data without relying on complex physical models, so they are more suitable for actual working conditions. In recent years, the superior performance of convolutional neural networks (CNN) in feature extraction and the ability of recurrent neural networks (RNN) and their variants (such as LSTM and GRU) in time series data modeling have made them a hot topic in SOC and SOH estimation research. However, existing data-driven methods still have limitations. For example, traditional recurrent neural networks have insufficient performance when processing long-term dependencies, limited storage capacity, and cannot achieve parallel processing. At the same time, most of the current research focuses on state estimation, lacking a joint modeling framework for state estimation and fault diagnosis, which leads to insufficient complementarity and integrity of battery management. Therefore, developing a joint state estimation and fault diagnosis method that can combine the advantages of multiple networks can not only improve the estimation accuracy of a single state, but also achieve coordinated optimization of the two, thereby meeting the application requirements of high-voltage battery packs under complex working conditions. Most importantly, none of the current data-driven methods take into account the micro-internal resistance of the battery pack, which is one of the important changing parameters affecting battery aging. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and propose a high-voltage battery pack state estimation and joint fault diagnosis method considering the internal resistance. First, the high-voltage battery pack data is preprocessed to obtain the original data set of the high-voltage battery pack. Secondly, a multi-dimensional input data correlation analysis is performed to obtain the parameters with the highest correlation with SOH, and a SOH estimation network based on MLP is built. Then, the estimated SOH and internal resistance, SOC change data are used as input to build a fault diagnosis network based on extended LSTM. It has the potential to be applied to the battery management system of high-voltage battery packs of electric vehicles / energy storage power stations.

[0005] In order to solve the above technical problems, the technical solution of the present invention is:

[0006] S1. Obtain micro-internal resistance data from the high-voltage battery pack. The four-wire method is used to measure the micro-internal resistance of the high-voltage battery pack. The instrument excitation end outputs an AC current and transmits it to the high-voltage battery pack. The ohmic internal resistance of the high-voltage battery pack responds with an AC voltage that returns to the instrument measurement end. The ohmic internal resistance of the high-voltage battery pack is obtained by calculating the current and voltage.

[0007] S2. The internal resistance data of the high-voltage battery pack obtained in step S1 and the long-term operation data of the high-voltage battery pack (high-voltage battery pack temperature, external temperature, battery pack voltage, maximum voltage of battery pack monomers, minimum voltage of battery pack monomers, battery pack current, battery pack temperature, battery pack SOC, and battery pack SOH) are used to form the original data set of the high-voltage battery pack. First, remove the outliers. When the difference between the adjacent current moment parameters and the two adjacent parameters is greater than 20%, the current moment parameters are eliminated. SOC is calculated by the ampere-hour integration method. The label of the battery pack SOH is calculated online and combined with offline indicators for analysis. The SOH calculation method is as follows: the equivalent cycle number is used for calculation, that is, when the battery is discharged from SOC 60% to 20% (discharge 40%), then from SOC 20% to 100% (charge 80%), then from SOC 100% to 40% (discharge 60%), and then from SOC 40% to 60% (charge 20%). This whole process is counted as an equivalent cycle, and the battery pack capacity value of this cycle is calculated. The ratio of the battery pack capacity value to the battery pack initial capacity is defined as the battery pack SOH.

[0008] S3. Correlation analysis of input data. Calculate the correlation between high-voltage battery pack temperature, external temperature, battery pack voltage, battery pack cell maximum voltage, battery pack cell minimum voltage, battery pack micro internal resistance, battery pack current, battery pack temperature, battery pack SOC and battery pack SOH. Use the Kendall correlation coefficient method to obtain the three parameters with the greatest correlation as the original input of the SOH estimation network model. Specifically, select the high-voltage battery pack micro internal resistance, pack voltage and cell maximum voltage as network inputs, and the three parameter values ​​of all cycles correspond to the values ​​of the battery pack SOH in each cycle.

[0009] S4. Build a SOH estimation network. Take the high-voltage battery pack micro-internal resistance, pack voltage and single cell maximum voltage as input to get the SOH estimation value. The SOH estimation network includes an MLP modeling module for feature dimension, an MLP modeling module for time dimension and a task layer. Between the three modules, a data dimension transposition operation is added to realize feature processing of different dimensions of the original data. The task layer realizes a many-to-one mapping from input to output.

[0010] S5. Fault diagnosis network construction. Taking the high-voltage battery pack micro-internal resistance, battery SOC and SOH estimation values ​​as input, the fault diagnosis network includes a time series modeling module based on an extended LSTM network and a task layer. Among them, the feature extraction module is used to extract battery aging characteristics and realize feature dimension upgrading. The extended LSTM network module is used to realize battery time series modeling, and the task layer realizes classification. The fault labels are as follows: the rapid drop, cliff drop, rapid rise, cliff rise (dynamic fault) of the high-voltage battery pack micro-internal resistance, battery SOC and battery SOH, and the battery SOH below the threshold (aging fault) are defined as faults.

[0011] S6. Perform back propagation training on the SOH estimation network and the fault diagnosis network. Model training update optimization. Set x = [(R, V pack ,V max ), SOH] as the input of the SOH estimation network, where R, V pack 、V max They represent the micro-internal resistance, pack voltage and single cell maximum voltage of the high-voltage battery pack. In the SOH estimation task, the three parameter curves of all time stamps in each cycle correspond to one SOH. To train the model, SOH = [SOH 1 ,SOH 2 ,...,SOH n ], where SOH t Represents the SOH of the high-voltage battery pack for t equivalent cycles. x = [(ΔR, ΔSOC, ΔSOH), fault] is used as the input of the fault diagnosis network, where ΔR and ΔSOC represent the changes in the micro-internal resistance and SOC of the battery pack, respectively, and ΔSOH represents the change in SOH and the situation below the threshold.

[0012] The present invention has the following characteristics and beneficial effects:

[0013] Adopting the above technical scheme, the present invention uses a joint network of SOH estimation and fault diagnosis to process high-voltage battery pack data in actual application scenarios of electric vehicles / energy storage power stations, optimizes input features through outlier removal and data correlation analysis, and selects battery pack internal resistance, battery pack voltage and battery pack single cell maximum voltage as network input variables. The SOH estimation network adopts a hierarchical modeling method of feature dimension and time dimension, combined with dimension transposition operation to improve feature extraction capability; the fault diagnosis network enhances the ability to capture battery fault characteristics by extending LSTM timing modeling. High-precision SOH estimation and accurate fault diagnosis of high-voltage battery packs can be achieved in multiple scenarios, further improving the operating safety of battery packs. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0015] Figure 1 A framework diagram of a high-voltage battery pack state estimation and fault diagnosis method taking internal resistance into consideration according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The present invention provides a high-voltage battery pack state estimation combined fault diagnosis method considering internal resistance, such as Figure 1 As shown, the following steps are included:

[0017] Step (1) obtains micro-internal resistance data from the high-voltage battery pack. The four-wire method is used to measure the micro-internal resistance of the high-voltage battery pack, with SOURCE+- forming the instrument excitation end and SENSE+- forming the instrument measurement end. The instrument excitation end outputs an AC current and transmits it to the high-voltage battery pack. The ohmic internal resistance of the high-voltage battery pack responds with an AC voltage that returns to the instrument measurement end, and the ohmic internal resistance of the battery is obtained by calculating the current and voltage. The measurement current flows through the SOURCE end, and the voltage generated by the high-voltage battery pack is measured from the SENSE end. The input impedance of the voltmeter for measuring voltage is high, and almost no current flows through it. Therefore, even if there is some wiring resistance or contact resistance, the voltage hardly drops, so that only the voltage generated by the object being measured can be measured.

[0018] Step (2) The long-term operation data of the 800V high-voltage battery pack is uploaded to the cloud platform, and the outliers are removed first. SOC is calculated by the ampere-hour integration method. The label of the battery pack SOH is calculated online and analyzed in combination with offline indicators. The SOH calculation method is as follows: the equivalent cycle number is used for calculation, that is, when the battery is discharged from SOC 60% to 20% (discharge 40%), then charged from SOC 20% to 100% (charge 80%), then discharged from SOC 100% to 40% (discharge 60%), and then charged from SOC 40% to 60% (charge 20%). This whole process is counted as an equivalent cycle, and the battery pack capacity value of this cycle is calculated. This ratio and the ratio of the initial capacity of the battery pack are defined as the battery pack SOH.

[0019] After the above processing, the original data obtained include high-voltage battery pack temperature, external temperature, battery pack voltage, battery pack single cell maximum voltage, battery pack single cell minimum voltage, battery pack micro internal resistance, battery pack current, battery pack temperature, battery pack SOC, and battery pack SOH. These data constitute the high-voltage battery pack original data set.

[0020] Step (3) performs correlation analysis on the high-voltage battery pack input data. Calculate the correlation between the high-voltage battery pack temperature, external temperature, battery pack voltage, battery pack single cell maximum voltage, battery pack single cell minimum voltage, battery pack micro-internal resistance, battery pack current, battery pack temperature, battery pack SOC and battery pack SOH. The Kendall correlation coefficient method is used to obtain the three parameters with the greatest correlation as the original input of the network model. The Kendall correlation coefficient calculation formula is as follows:

[0021]

[0022] Among them, the same-order logarithm refers to the observation pairs with the same order, the reverse-order logarithm refers to the observation pairs with the opposite order, and n represents the total number of paired observations, that is, the number of high-voltage battery pack temperature, external temperature, battery pack voltage, battery pack single cell maximum voltage, battery pack single cell minimum voltage, battery pack micro-internal resistance, battery pack current, battery pack temperature and battery SOC, battery SOH respectively. The value range of the Kendall correlation coefficient is between -1 and 1. When τ is 1, it means that the two random variables have consistent rank correlation; when τ is -1, it means that the two random variables have completely opposite rank correlation. Take the parameters with the absolute value of the Kendall correlation coefficient close to 1 as the network input. Among them, each charge and discharge cycle corresponds to a SOH. The input of the entire cycle in the SOH estimation task corresponds to a SOH. Finally, the battery pack internal resistance, battery pack voltage and battery pack single cell maximum voltage are selected as the SOH estimation network input.

[0023] Step (4) SOH estimation network model building. With the high-voltage battery pack micro-internal resistance, pack voltage and single cell maximum voltage as input, an MLP-based SOH estimation network is built. The SOH estimation network includes an MLP modeling module for the feature dimension, an MLP modeling module for the time dimension, and a task layer. Between the three modules, a data dimension transposition operation is added to realize feature processing of different dimensions of the original data.

[0024] The SOH estimation network is built only by the MLP network. Specifically, the data first passes through an MLP network:

[0025]

[0026] Among them, the superscript represents the first MLP network, represents the features after the first MLP, X i represents input data, MLP(·) represents MLP operation, ReLU(·) represents activation function, W represents weight, and b represents bias. Then the data is transposed (feature dimension and time dimension are exchanged) to obtain Then pass through an MLP network again:

[0027]

[0028] in, is the output of the second MLP. Then, the data dimension is transposed to the initial dimension form, and we get Finally, it is mapped to the battery SOH output through the output layer:

[0029]

[0030] Among them, F(·) represents the Flatten layer and D[·] represents the Dense layer.

[0031] Step (5) Fault network model construction. Taking the high-voltage battery pack micro-internal resistance, battery SOC and battery SOH estimated in step (4) as input, a fault diagnosis network based on extended LSTM is built. The fault diagnosis network includes three stacked time series modeling modules based on the extended LSTM network and a task layer. Among them, the extended LSTM network module is used to realize battery time series modeling, and the task layer realizes classification output. The fault diagnosis network is composed of three layers of extended LSTM networks stacked together. The covariance matrix and normalized state are introduced in the extended LSTM network to expand the changes in internal states. The extended LSTM network has six gates, and three gate mechanisms that are more complex than LSTM are introduced in the update process, namely query gate (Query), key gate (Key) and value gate (Value). Different from LSTM, the extended LSTM network introduces the covariance matrix C tTo capture the high-order relationship between input features, what is stored is no longer a one-dimensional cell state, but a multi-dimensional matrix. Assuming the input is x, the fault diagnosis process is as follows:

[0032] Fault = D[KLSTM(KLSTM(KLSTM(x)))] (5)

[0033] Where D[·] represents the Dense layer and the softmax layer. KLSTM(·) represents a layer of extended LSTM operation. Taking a layer of extended LSTM network as an example, its specific operation is as follows. First, the input gate, forget gate, and output gate are pre-activated, and the activation function is applied, just like LSTM:

[0034]

[0035] act [i,f,o] =[exp(·),σ(·),σ(·)] (8)

[0036] in, Represent the updated output of the input gate, the output of the forget gate, and the output of the output gate, respectively. t 、f t and t Represents the output of the input gate, the output of the forget gate, and the output of the output gate at the previous moment. W and B represent the weight matrix and the bias term, respectively. exp(·) and σ(·) represent the activation function. Then, the new query gate, key gate, and value gate are pre-activated:

[0037] q t ←W q x+b q (9)

[0038]

[0039] v t ←W v x+b v (11)

[0040] Among them, q t , k t and v t Represent the outputs of the query gate, key gate, and value gate respectively. To reduce digital fluctuations, k t Scaled to Then, update the stable state m t :

[0041] m t ←max(log(f t )+m t-1 ,log(i t )) (12)

[0042] Among them, max(·) represents the maximum value, m t-1 Represents the stable state of the previous moment. Then, update the covariance matrix Ct:

[0043]

[0044] in, represents the outer product operation between input features, C t Used to capture the correlation between features. Then update the normalized state n t :

[0045] n t ←f t ⊙n t-1 +i t ′·k t (15)

[0046] Finally, the hidden state is calculated:

[0047] norm_inner←diag(n t ·q t T ) (16)

[0048] div←max(|norm_inner|,1) (17)

[0049]

[0050] Among them, norm_inner represents the normalized inner product, diag(·) represents the extraction of the diagonal matrix, and q t T Represents q t The transposed matrix of div represents the normalization factor, which is used to control the numerical range, and h t represents the final hidden state.

[0051] Step (6) Model training update optimization. Set x = [(R, V pack ,V max ), SOH] as the network input, where R, V pack 、V max They represent the micro-internal resistance, pack voltage and single cell maximum voltage of the high-voltage battery pack. In the SOH estimation task, the three parameter curves of all time stamps in each cycle correspond to one SOH. To train the model, SOH = [SOH 1 ,SOH 2 ,...,SOH n ], where SOH tRepresents the SOH of the high-voltage battery pack for t equivalent cycles. Take x = [(R, SOC, SOH), fault] as the network input, where R and SOC represent the micro-internal resistance and SOC value of the battery pack respectively, and SOH represents the estimated value of SOH.

[0052] Step (7) The present invention conducts a comparative experiment on the battery pack SOH estimation effect and fault diagnosis effect on the actual high-voltage battery pack data (the ratio of the training set to the test set is 3:1), wherein MAE, RMSE and diagnostic accuracy are used as evaluation indicators. The comparison method adopts the currently widely used CNN method and LSTM method, and the test results are shown in Table 1.

[0053] Table 1

[0054] Estimation Model SOH MAE (%) SOH RMSE (%) Fault diagnosis accuracy CNN 3.6 7.0 85.1％ LSTM 2.9 5.7 87.2％ The present invention 1.9 3.1 93.5％

[0055] Among them, MAE and RMSE are mean absolute error and root mean square error loss functions respectively, and fault diagnosis accuracy represents the ratio of the number of diagnosed faults to the number of actual faults. It can be seen from Table 1 that the high-voltage battery pack state estimation combined fault diagnosis method considering internal resistance proposed in the present invention can well achieve high-precision SOH estimation of high-voltage battery packs.

[0056] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments including components are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

Claims

1. A high-voltage battery pack state estimation and fault diagnosis method considering internal resistance, characterized in that: The following steps are involved: S1. Obtain micro internal resistance data from the high-voltage battery pack; S2, combining the micro internal resistance data of the high-voltage battery pack and the long-term operation data of the high-voltage battery pack into an original data set of the high-voltage battery pack; S3. performing correlation analysis on the data in the original data set of the high-voltage battery pack; S4, build a SOH estimation network, input the data obtained from the correlation analysis into the SOH estimation network, and obtain the SOH estimation value; S5. Fault diagnosis network is built, taking the high-voltage battery pack micro-internal resistance, battery SOC and battery SOH estimation values ​​as input, and outputting fault diagnosis results; S6. Perform back-propagation training on the SOH estimation network and the fault diagnosis network.

2. The high-voltage battery pack state estimation and fault diagnosis method considering internal resistance according to claim 1, characterized in that: The method of obtaining micro-internal resistance data from the high-voltage battery pack end is specifically as follows: a four-wire method is used to measure the micro-internal resistance of the high-voltage battery pack, an AC current is output from the excitation end of the instrument and transmitted to the high-voltage battery pack; the ohmic internal resistance of the high-voltage battery pack responds with an AC voltage returned to the measurement end of the instrument, and the ohmic internal resistance of the high-voltage battery pack is obtained by calculating the current and voltage.

3. The high-voltage battery pack state estimation and fault diagnosis method considering internal resistance according to claim 2 is characterized in that: The specific implementation process of step S2 is as follows: The internal resistance data of the high-voltage battery pack and the long-term operation data of the high-voltage battery pack; the long-term operation data of the high-voltage battery pack includes the high-voltage battery pack temperature, the external temperature, the battery pack voltage, the maximum voltage of the battery pack cells, the minimum voltage of the battery pack cells, the battery pack current, the battery pack temperature, the battery pack SOC and the battery pack SOH, constitute the original data set of the high-voltage battery pack and perform preprocessing.

4. The high-voltage battery pack state estimation and fault diagnosis method considering internal resistance according to claim 3 is characterized in that: The step S2 further includes: removing abnormal values, and removing the current moment parameter when the difference between the adjacent current moment parameter and the two adjacent parameters is greater than 20%; SOC is calculated by the ampere-hour integration method; the label of the battery pack SOH is calculated online and combined with offline indicators for analysis; The SOH calculation method is as follows: the equivalent cycle number is used for calculation, that is, when the battery is discharged from SOC 60% to 20%, then charged from SOC20% to 100%, then discharged from SOC 100% to 40%, and then charged from SOC 40% to 60%; this whole process is counted as an equivalent cycle, and the battery pack capacity value of this cycle is calculated; the ratio of the battery pack capacity value to the initial capacity of the battery pack is defined as the battery pack SOH.

5. The high-voltage battery pack state estimation and fault diagnosis method considering internal resistance according to claim 4 is characterized in that: The step S3 is specifically as follows: calculating the correlation among high-voltage battery pack temperature, external temperature, battery pack voltage, battery pack single cell maximum voltage, battery pack single cell minimum voltage, battery pack micro internal resistance, battery pack current, battery pack temperature, battery pack SOC and battery pack SOH; using the Kendall correlation coefficient method to obtain the three parameters with the greatest correlation, namely, high-voltage battery pack micro internal resistance, pack voltage and single cell maximum voltage, as the original inputs of the SOH estimation network model, and the three parameter values ​​of all cycles correspond to the values ​​of the battery pack SOH in each cycle.

6. The high-voltage battery pack state estimation and fault diagnosis method considering internal resistance according to claim 5 is characterized in that: The SOH estimation network includes an MLP modeling module for feature dimension, an MLP modeling module for time dimension and a task layer. A data dimension transposition operation is added between the three modules to realize feature processing of different dimensions of the original data. The task layer realizes a many-to-one mapping from input to output.

7. The high-voltage battery pack state estimation and fault diagnosis method considering internal resistance according to claim 6, characterized in that: The fault diagnosis network includes a time series modeling module based on an extended LSTM network and a task layer; wherein the extended LSTM network realizes battery time series modeling, and the task layer realizes classification; the fault labels include: defining the rapid decline, cliff decline, rapid rise, cliff rise of the high-voltage battery pack micro-internal resistance, battery SOC and battery SOH, and the battery SOH being lower than the threshold as faults.

Images