A battery state of health (SOH) estimation method for an electric vehicle extracts polarization features from real-world dynamic conditions

Abstract

The present application relates to a kind of electric vehicle battery pack SOH estimation method of extracting polarization feature from real vehicle dynamic working condition, belong to battery technical field.For the problems that existing method relies on laboratory data and complete charging segment, cannot effectively utilize polarization feature caused by current switching in dynamic working condition, the present application extracts charging segment by collating battery operation data and preprocessing, based on the SOC distribution of current switching time in multistage constant current charging curve analysis, filters feature extraction interval and sets dynamic threshold, extracts polarization voltage difference, current variation and other features from representative monomer, constructs two-stage machine learning model to realize SOH estimation.This method breaks through the limitation of traditional static feature, utilizes the instantaneous polarization effect of current switching under dynamic working condition, significantly improves the utilization rate of real vehicle data and feature representation ability, effectively improves battery pack health status evaluation precision, realizes electric vehicle large-scale rapid detection demand.

Description

Technical Field

[0001] This invention belongs to the field of battery technology and relates to a method for estimating the state of harmonics (SOH) of electric vehicle battery packs by extracting polarization features from real vehicle dynamic conditions. Background Technology

[0002] Currently, methods for estimating the state of harm (SOH) of lithium-ion batteries mainly include model-based methods and data-driven methods. Model-based methods, based on in-depth research into electrochemical mechanisms, establish mathematical models of battery degradation phenomena and use optimization methods such as least squares to identify model parameters. Then, they estimate battery SOH using methods such as Kalman filtering. These models include equivalent circuit models and electrochemical models. Data-driven methods treat the battery as a black box, establishing a mapping relationship between features reflecting battery aging information (i.e., extracting battery health characteristics as input) and using machine learning to build an estimation model. This method directly uses historical monitoring data to predict battery degradation trends, avoiding the need for complex physics-based models and offering greater flexibility and applicability. Although many SOH estimation methods exist, most are based on laboratory data, making them difficult to apply to real vehicles. Furthermore, in terms of feature engineering, current methods rely on complete charging segments and do not consider the dynamic response characteristics of the battery caused by current switching during vehicle operation in the real world. During actual vehicle operation, the back-and-forth movement of lithium ions between the positive and negative terminals during current switching causes a short-term imbalance in the concentration of lithium ions inside the battery, resulting in polarization. This phenomenon becomes more pronounced as the battery ages and can be used to evaluate the battery's state of harmonic equilibrium (SOH). Current research mostly focuses on depolarization after full charging. This method requires a relatively long resting period after full charging. However, in the real world, after a full charge, the vehicle is often turned off and no longer records data, making it difficult to obtain relaxation voltage data; or the vehicle starts driving immediately, making large-scale application difficult. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a method for estimating the state of harm (SOH) of an electric vehicle battery pack by extracting polarization features from real vehicle dynamic conditions. This method can extract polarization mechanism features characterizing the battery aging process from multi-stage constant current charging curves, achieve accurate SOH estimation of the battery pack, and be applied to rapid SOH detection on a large scale in electric vehicles.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for estimating the state of harmonics (SOH) of an electric vehicle battery pack by extracting polarization features from real-world vehicle dynamic conditions, the method comprising the following steps:

[0006] S1: Organize the operating data from the battery packs of hybrid electric vehicles and establish a battery dataset;

[0007] S2: Analyze and preprocess the battery dataset, extract relatively complete charging segments, and calculate the tag capacity using the ampere-hour integration method;

[0008] S3: Based on dynamic operating conditions, analyze the changes in current and SOC in the multi-stage constant current charging curve, set feature filtering conditions, and select the current switching time for feature extraction.

[0009] S4: Analyze the condition of each cell in the battery pack, select representative cells, and extract the polarization characteristics of representative cells based on the selected current switching time.

[0010] S5: Establish a machine learning estimation model, take the extracted polarization feature set as input, and obtain the SOH estimation result of the battery pack.

[0011] Furthermore, S2 specifically refers to:

[0012] S21: Analyze the charging data in the battery dataset, preprocess it, and select relatively complete charging segments that are suitable for subsequent tag capacity calculation;

[0013] S22: Based on the extracted charging segment data, calculate the current maximum usable capacity of the battery pack using the ampere-hour integration method to obtain the capacity label for subsequent model training; the mathematical expression for calculating the label capacity based on the ampere-hour integration method is:

[0014]

[0015] Where SOC(t) and SOC(t0) are the SOCs of the battery pack at time steps t and t0, respectively, and I(t) is the current of the battery pack at time step t; C act It is the maximum available capacity of the battery pack.

[0016] Furthermore, S3 specifically refers to:

[0017] S31: Based on the rule of current change with SOC during charging, analyze the SOC distribution corresponding to the current switching of the battery pack charging segment, and divide the feature extraction interval according to SOC.

[0018] S32: Based on the selected feature extraction interval, statistically analyze the current change during current switching within this interval, and set feature filtering conditions according to the current change and SOC.

[0019] S33: Count the number of charging segments at each current switching moment, and select the current switching moment for feature extraction.

[0020] Furthermore, S4 specifically includes:

[0021] S41: Based on the selected current switching time, analyze the situation of each cell in the battery pack and find several representative cells that can represent the capacity degradation of the battery pack at each time.

[0022] S42: Extract the polarization characteristics of each representative cell from the selected current switching time to form a polarization characteristic set.

[0023] Furthermore, the polarization characteristics include the maximum voltage difference among all cells during current switching, the voltage of the cell corresponding to the maximum voltage difference during current switching at the moment before switching, the minimum voltage difference among all cells during current switching, the voltage of the cell corresponding to the minimum voltage difference during current switching at the moment before switching, the maximum cell voltage value at the moment before current switching, the voltage difference corresponding to the maximum cell voltage value during current switching, the minimum cell voltage value at the moment before current switching, the voltage difference corresponding to the minimum cell voltage value during current switching, the average voltage of all cells at the moment before current switching, the voltage difference corresponding to the average voltage of all cells during current switching, the current value corresponding to the moment before current switching, and the current difference corresponding to current switching, etc.

[0024] Furthermore, S5 specifically includes:

[0025] S51: Establish a machine learning estimation model, input the training data of the extracted polarization feature set into the model for training, and perform SOH estimation of the battery pack;

[0026] S52: Combine the selected current switching times to verify the effectiveness and versatility of the proposed polarization characteristics, enabling rapid detection of vehicle SOH.

[0027] Furthermore, the feature selection criteria are set such that the current change at the current switching moment is greater than 2A, and the SOC value is selected from at least two consecutive intervals among (61.5%, 66%), (71%, 76%), (80%, 85%), and (95%, 97.1%).

[0028] Furthermore, the selected current switching time corresponds to three SOC values ​​of 71%, 85%, and 95%, and the number of charging segments corresponding to each feature point is no less than 100 historical charging cycles.

[0029] Furthermore, the selection criteria for the representative monomers include:

[0030] The top 3 cells with the largest voltage changes during current switching;

[0031] The three cells with the smallest voltage change amplitude at the current switching moment;

[0032] Individual cells whose voltage difference before and after current switching exceeds a set threshold;

[0033] Abnormal monomers with a historical capacity decay rate exceeding the group average decay rate by 20%.

[0034] Furthermore, the machine learning estimation model adopts a two-stage architecture based on feature importance. The first stage uses a random forest to select a subset of polarized features, and the second stage uses a Long Short-Term Memory (LSTM) network with an attention mechanism to predict the SOH sequence. The attention weight allocation function is as follows:

[0035]

[0036] in, This represents the hidden state vector of the LSTM at time step t, with dimension d. h The characterization model represents the memory encoding results of historical polarization feature sequences;

[0037] This represents a trainable parameter matrix used to learn the importance mapping relationship of features at different time steps;

[0038] This represents a scalar value indicating the matching degree between the feature at time step t and the global attention pattern.

[0039] This indicates that the matching degree for all time steps s = 1, 2, ..., T is normalized to ensure the attention weight α t ∈(0,1) and

[0040] α t The attention weight at time step t represents the strength of the contribution of the polarization features at that moment to the current SOH prediction.

[0041] The beneficial effects of this invention are as follows:

[0042] (1) A method for extracting features across intervals from dynamic operating conditions is proposed. This method uses dynamic operating conditions to replace stable operating conditions, which solves the problem that the operating conditions of real vehicles are mostly dynamic due to the influence of user behavior, charging piles, environment and other factors. Through dynamic operating condition feature screening and key SOC point selection, the utilization rate of real vehicle data is increased from 62% to 89% of the traditional method.

[0043] (2) The polarization feature set is extracted from the current switching moment of the multi-level constant current charging curve. This feature set can effectively characterize the battery aging from the perspective of mechanism.

[0044] (3) Compared with traditional feature extraction methods, the proposed feature extraction method only requires a few seconds of charging time before and after the current switching moment to complete the battery pack SOH estimation. It solves the problems of large data requirements and voltage confusion caused by current switching in traditional feature extraction methods, and can be widely applied to large-scale rapid SOH detection of electric vehicles.

[0045] (4) The multi-dimensional single-cell screening mechanism enables the detection time of abnormal aging cells in the battery pack to be advanced by 1200 kilometers of driving distance.

[0046] (5) The two-stage model architecture breaks through the limitations of traditional machine learning methods in temporal correlation modeling. In tests with 1500 sets of real vehicle data, R... 2 The value reached a historical best level of 0.976.

[0047] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0048] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0049] Figure 1 This is a flowchart illustrating the overall method of the present invention;

[0050] Figure 2 This is a general framework diagram of the method in the embodiment;

[0051] Figure 3 Here is a detailed flowchart of embodiment S2;

[0052] Figure 4 Here is a detailed flowchart of embodiment S3;

[0053] Figure 5 Here is a detailed flowchart of embodiment S4;

[0054] Figure 6 This is a detailed flowchart of embodiment S5. Detailed Implementation

[0055] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0056] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0057] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0058] Please see Figure 1 and Figure 2 A method for estimating the State of Harm (SOH) of an electric vehicle battery pack by extracting polarization features from real-world vehicle dynamic conditions can be divided into the following steps:

[0059] S1: Organize the operating data from the battery packs of hybrid electric vehicles and establish a battery dataset;

[0060] S2: Analyze and preprocess the battery dataset, extract relatively complete charging segments, and calculate the tag capacity using the ampere-hour integration method;

[0061] S3: Based on dynamic operating conditions, analyze the changes in current and SOC in the multi-stage constant current charging curve, set feature filtering conditions, and select the current switching time for feature extraction.

[0062] S4: Analyze the condition of each cell in the battery pack, select representative cells, and extract the polarization characteristics of representative cells based on the selected current switching time.

[0063] S5: Establish a machine learning estimation model, take the extracted polarization feature set as input, and obtain the SOH estimation result of the battery pack.

[0064] Please see Figure 3 The steps for calculating tag capacity using the ampere-hour integration method are as follows:

[0065] S21: Analyze the charging data in the battery dataset, preprocess it, and select relatively complete charging segments that are suitable for subsequent tag capacity calculation;

[0066] S22: Based on the extracted charging segment data, the current maximum usable capacity of the battery pack is calculated using the ampere-hour integration method to obtain the capacity label for subsequent model training. The mathematical expression for calculating the label capacity based on the ampere-hour integration method is:

[0067]

[0068] Where SOC(t) and SOC(t0) are the SOCs of the battery pack at time steps t and t0, respectively, and I(t) is the current of the battery pack at time step t; C act It is the maximum available capacity of the battery pack.

[0069] Please see Figure 4 The steps for dividing the feature extraction interval and setting feature filtering conditions are as follows:

[0070] S31: Based on the rule of current change with SOC during charging, analyze the SOC distribution corresponding to the current switching of the battery pack charging segment, and divide the feature extraction interval according to SOC.

[0071] S32: Based on the selected feature extraction interval, statistically analyze the current changes during current switching within this interval. Set feature filtering conditions based on current changes and SOC. In this example, the optional feature filtering conditions include, but are not limited to, the following:

[0072] Current: The current change is greater than 2A at the moment of current switching;

[0073] SOC: Current switching time SOC∈(61.5, 66, 71, 76, 80, 85, 95, 97.1).

[0074] S33: Count the number of charging segments at each current switching moment, and select the current switching moment for feature extraction. In this example, the selectable current switching moments include, but are not limited to, the following:

[0075] SOC = 71%, 85%, 95%

[0076] By setting a threshold condition of current change > 2A, false current switching signals caused by sensor noise can be effectively filtered out, ensuring the reliability of feature extraction. When selecting SOC intervals (61.5%, 66%), (71%, 76%), (80%, 85%), and (95%, 97.1%), it was found that these intervals correspond to the SOC operating range where the battery polarization voltage is most sensitive to aging. Experimental data shows that this screening condition can improve the effectiveness of features by 23%. Three feature points, SOC = 71%, 85%, and 95%, were specifically selected because they correspond to the inflection point of the mid-range polarization effect, the capacity mutation region in the high SOC interval, and the polarization saturation region before full charge, respectively. Furthermore, each feature point is required to contain no less than 100 historical charging cycle data points to ensure statistical significance and avoid model overfitting problems caused by data sparsity.

[0077] Please see Figure 5 The steps for extracting the polarization feature set are as follows:

[0078] S41: Based on the selected current switching time, analyze the condition of each cell in the battery pack, and find several representative cells that can represent the capacity degradation of the battery pack at each time. In this example, the selectable representative cells include, but are not limited to, the following:

[0079]

[0080] S42: Extract the polarization characteristics of each representative cell from the selected current switching moment to form a polarization feature set. Taking a selected current switching moment as an example, the selectable polarization characteristics include, but are not limited to, the following:

[0081]

[0082] Representative monomers were selected using multi-dimensional criteria:

[0083] The top 3 cells with the largest / smallest voltage changes: capturing extreme cases of polarization response within the battery pack and reflecting the maximum performance differences within the pack;

[0084] Monomers with voltage differences exceeding a threshold (e.g., >50mV): Identify abnormally aged monomers; experiments show that such monomers contribute up to 35% to the overall SOH prediction.

[0085] Abnormal cells with a historical degradation rate exceeding the group average by 20%: Early detection of potential faulty cells, providing early warning before the battery pack capacity drops significantly.

[0086] By using composite screening criteria, the model's ability to characterize the discretization of individual cell performance within the battery pack is improved by 41%, especially in the later stages of battery pack aging (SOH < 80%), where the prediction error is reduced to less than 1.8%.

[0087] Please see Figure 6 The steps for estimating SOH based on machine learning are as follows:

[0088] S51: Establish a machine learning estimation model, input the training data of the extracted polarization feature set into the model for training, and perform SOH estimation of the battery pack;

[0089] S52: Combine the selected current switching times to verify the effectiveness and versatility of the proposed polarization characteristics, enabling rapid detection of vehicle SOH.

[0090] The innovation of the two-stage model architecture lies in:

[0091] Random Forest Feature Selection Stage: By calculating the feature importance score (Gini importance > 0.15), 12 key features are selected from the original 24 polarization features, reducing redundant computation by 48%.

[0092] LSTM prediction stage with attention mechanism:

[0093] The time attention mechanism is expressed through the formula Dynamically assigning weights to each time step, experiments show that the attention given to the characteristics at the time SOC=95% is 3.2 times that at other times, which is highly consistent with the polarization saturation characteristics in this interval;

[0094] in, This represents the hidden state vector of the LSTM at time step t, with dimension d. h The characterization model represents the memory encoding results of historical polarization feature sequences;

[0095] This represents a trainable parameter matrix used to learn the importance mapping relationship of features at different time steps;

[0096] This represents a scalar value indicating the matching degree between the feature at time step t and the global attention pattern.

[0097] This indicates that the matching degree for all time steps s = 1, 2, ..., T is normalized to ensure the attention weight α t ∈(0,1) and

[0098] α t The attention weight at time step t represents the strength of the contribution of the polarization features at that moment to the current SOH prediction.

[0099] Through W a Due to its trainability, the model can adaptively focus on polarization features at times with high information content, such as SOC = 95% (experiments show that α at such times...). tThe value can reach 0.42, which is 3.2 times that of other times;

[0100] Hidden state h t With weighted moments W a Interactive computation enables the model to accurately capture temporal correlations under dynamic battery pack conditions, reducing prediction error variance by 58% compared to traditional LSTM models.

[0101] By introducing time-series correlation analysis of historical capacity decay trajectories, the predicted stability index (MAE / variance) under real vehicle dynamic conditions is improved by 37% compared with the traditional static feature method.

[0102] Bench tests verified that the SOH estimation error of this architecture remained within 2.5% under extreme conditions of -20℃ low temperature and 5C fast charging, which is significantly better than the single model method.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for estimating the state of harmonics (SOH) of an electric vehicle battery pack by extracting polarization features from real vehicle dynamic conditions, characterized in that: The method includes the following steps: S1: Organize the operating data from the battery packs of hybrid electric vehicles and establish a battery dataset; S2: Analyze and preprocess the battery dataset, extract relatively complete charging segments, and calculate the tag capacity using the ampere-hour integration method; S3: Based on dynamic operating conditions, analyze the changes in current and SOC in the multi-stage constant current charging curves, set feature filtering conditions, and select the current switching moments for feature extraction; specifically: S31: Based on the rule of current change with SOC during charging, analyze the SOC distribution corresponding to the current switching of the battery pack charging segment, and divide the feature extraction interval according to SOC. S32: Based on the selected feature extraction interval, statistically analyze the current change during current switching within this interval, and set feature filtering conditions according to the current change and SOC. S33: Count the number of charging segments at each current switching moment, and select the current switching moment for feature extraction; S4: Analyze the condition of each cell in the battery pack, select representative cells, and extract the polarization characteristics of the representative cells based on the selected current switching time; specifically: S41: Based on the selected current switching time, analyze the situation of each cell in the battery pack and find several representative cells that can represent the capacity degradation of the battery pack at each time. S42: Extract the polarization characteristics of each representative cell from the selected current switching moment to form a polarization characteristic set; The selection criteria for the representative monomers include: The top 3 cells with the largest voltage changes during current switching; The three cells with the smallest voltage change amplitude at the current switching moment; Individual cells whose voltage difference before and after current switching exceeds a set threshold; Abnormal monomers with a historical capacity decay rate exceeding the group average decay rate by 20%; S5: Establish a machine learning estimation model, take the extracted polarization feature set as input, and obtain the SOH estimation result of the battery pack.

2. The method for estimating the state of harmonics (SOH) of an electric vehicle battery pack by extracting polarization features from real vehicle dynamic conditions according to claim 1, characterized in that: Specifically, S2 is: S21: Analyze the charging data in the battery dataset, preprocess it, and select relatively complete charging segments that are suitable for subsequent tag capacity calculation; S22: Based on the extracted charging segment data, calculate the current maximum usable capacity of the battery pack using the ampere-hour integration method to obtain the capacity label for subsequent model training; the mathematical expression for calculating the label capacity based on the ampere-hour integration method is: in, and These are the battery pack at the time step. and SOC, It is the battery pack in time step The current; It is the maximum available capacity of the battery pack.

3. The method for estimating the state of harmonics (SOH) of an electric vehicle battery pack by extracting polarization features from real vehicle dynamic conditions according to claim 1, characterized in that: The polarization characteristics include the maximum voltage difference among all cells during current switching, the voltage of the cell corresponding to the maximum voltage difference during current switching at the moment before switching, the minimum voltage difference among all cells during current switching, the voltage of the cell corresponding to the minimum voltage difference during current switching at the moment before switching, the maximum cell voltage value at the moment before current switching, the voltage difference corresponding to the maximum cell voltage value during current switching, the minimum cell voltage value at the moment before current switching, the voltage difference corresponding to the minimum cell voltage value during current switching, the average voltage of all cells at the moment before current switching, the voltage difference corresponding to the average voltage of all cells during current switching, the current value corresponding to the moment before current switching, and the current difference corresponding to current switching.

4. The method for estimating the state of harmonics (SOH) of an electric vehicle battery pack by extracting polarization features from real vehicle dynamic conditions according to claim 3, characterized in that: Specifically, S5 is: S51: Establish a machine learning estimation model, input the training data of the extracted polarization feature set into the model for training, and perform SOH estimation of the battery pack; S52: Combine the selected current switching times to verify the effectiveness and versatility of the proposed polarization characteristics, enabling rapid detection of vehicle SOH.

5. The method for estimating the state of harmonics (SOH) of an electric vehicle battery pack by extracting polarization features from real vehicle dynamic conditions according to claim 1, characterized in that: The feature selection criteria are set as follows: the current change at the current switching moment is greater than 2A, and the SOC value is selected from at least two consecutive intervals among (61.5%, 66%), (71%, 76%), (80%, 85%), and (95%, 97.1%).

6. The method for estimating the state of harmonics (SOH) of an electric vehicle battery pack by extracting polarization features from real vehicle dynamic conditions according to claim 1, characterized in that: The selected current switching time corresponds to three SOC values: 71%, 85%, and 95%, and the number of charging segments corresponding to each feature point is no less than 100 historical charging cycles.

7. The method for estimating the state of harmonics (SOH) of an electric vehicle battery pack by extracting polarization features from real vehicle dynamic conditions according to claim 3, characterized in that: The machine learning estimation model adopts a two-stage architecture based on feature importance. The first stage uses a random forest to select a subset of polarized features, and the second stage uses a Long Short-Term Memory (LSTM) network with an attention mechanism to predict the SOH sequence. The attention weight allocation function is as follows: in, This indicates that the LSTM is at time step t The hidden state vector, with dimension The characterization model represents the memory encoding results of historical polarization feature sequences; , represents the trainable parameter matrix, used to learn the importance mapping relationship of features at different time steps; Indicates the calculation time step t The scalar value of the match between the feature and the global attention pattern; Indicates all time steps s =1,2,..., T The matching degree is normalized to ensure the attention weight. and ; Indicates time step t The attention weights reflect the strength of the contribution of the polarization features at that moment to the current SOH prediction.

Images