A battery health state estimation method based on semi-supervised regression and collaborative training

Abstract

The present invention relates to a battery health state estimation method based on semi-supervised regression and collaborative training, belonging to the field of battery technology. The method comprises the following steps: S1: collecting operating data of electric vehicles and establishing a battery operation database; S2: calculating battery health state labels using the Ampere integral formula and open circuit voltage correction strategy; S3: extracting a set of health indicators that are highly correlated with the battery health state based on the Pearson correlation coefficient and the grey correlation gradient index, respectively; S4: using the selected health indicator set to estimate pseudo labels of unlabeled data using semi-supervised collaborative training to enhance the training data set, and constructing a battery health state estimation model based on a multi-layer perceptron; S5: using the test data after extracting the health indicator set as input to the battery health state estimation model to obtain an estimated battery health state. The present invention improves the accuracy of battery health state estimation when labeled data is limited.

Description

Technical Field

[0001] The present invention belongs to the technical field of batteries and relates to a battery health state estimation method based on semi-supervised regression and collaborative training. Background Art

[0002] Battery State of Health (SOH) estimation methods can generally be categorized into model-based and data-driven approaches. Model-based approaches describe battery aging by building mathematical models, typically including equivalent circuit models and electrochemical models. Equivalent circuit model-based approaches employ advanced filtering algorithms, such as the least squares method, to estimate the battery SOH online. Electrochemical model-based SOH estimation methods directly calculate the battery SOH by identifying electrochemical parameters associated with battery aging. Data-driven approaches have garnered significant attention in recent years due to their reliance on sufficient data to achieve high-accuracy estimates. Data-driven approaches typically formulate SOH estimation as a supervised regression problem, employing advanced machine learning algorithms to build models that learn the nonlinear relationship between input variables and SOH without requiring prior knowledge of battery parameters. Typically, input variables are statistical features of directly measured parameters that reflect the battery SOH. Both lightweight and deep machine learning models have gained popularity in battery SOH estimation. However, the successful application of these approaches relies on the availability of a large number of battery SOH labels, which is often unavailable with field data. In contrast to supervised learning methods that rely entirely on labeled data for model training, semi-supervised learning methods leverage both labeled and unlabeled data by assigning pseudo labels to unlabeled samples. Current research is primarily developed and evaluated using laboratory data under controlled conditions, lacking implementation of battery health state estimation in real-world scenarios.

[0003] To address the above issues, no effective semi-supervised learning and collaborative training methods based on field data have been proposed to efficiently and accurately estimate the health status of battery packs. Summary of the Invention

[0004] In view of this, the object of the present invention is to provide a battery health state estimation method based on semi-supervised regression and collaborative training, which can accurately estimate the battery health state based on field data.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A battery health state estimation method based on semi-supervised regression and collaborative training, the method comprising the following steps:

[0007] S1: Collect the operating data of electric vehicles, including the operating data of each cell in the battery pack, and establish a battery operation database;

[0008] S2: Analyzes the charging data of electric vehicles, selects charging segments that meet the requirements, and calculates the battery health status label using the ampere integral formula and open circuit voltage correction strategy;

[0009] S3: Extract the health indicator set that is highly correlated with the battery health status based on the Pearson correlation coefficient PCC and the grey relational gradient GRG indicator;

[0010] S4: Based on the selected health indicator set, semi-supervised co-training is used to estimate pseudo labels of unlabeled data to enhance the training dataset, and a battery health state estimation model is constructed based on a multi-layer perceptron (MLP).

[0011] S5: The test data after extracting the health indicator set is used as the input of the battery health state estimation model, and all regressors are integrated to obtain the estimated battery health state.

[0012] Furthermore, the S1 is specifically:

[0013] S11: Collect operating data of a certain electric vehicle battery pack, including charge and discharge current, voltage, temperature, time, and power parameters;

[0014] S12: Collecting cell operating data in the battery pack, including cell voltages of all cells and temperature data sets of each temperature probe placed near the cells;

[0015] S13: Based on the collected operating data of the battery cells and battery packs, a battery pack operating database for a certain vehicle is established.

[0016] Furthermore, the S2 is specifically:

[0017] S21: analyzing the charging data of the electric vehicle and screening the charging segments that meet the rules of the open circuit voltage correction strategy;

[0018] S22: Collecting the open circuit voltage-state of charge lookup table provided by the battery manufacturer, adopting the open circuit voltage correction strategy, and correcting the initial battery state of charge according to the minimum battery voltage at the initial time;

[0019] S23: Based on the historical operating data of the battery pack, the maximum available capacity of the battery pack is calculated using the ampere integral formula to obtain a battery health status label.

[0020] Furthermore, the S3 is specifically:

[0021] S31: performing a charging curve analysis based on the maximum voltage of the battery for the charging segment of the battery pack to determine a voltage range for extracting the health indicator set;

[0022] S32: Calculate statistical characteristics of the charging data, including the average value and standard deviation, based on the selected voltage range, and extract a health indicator set;

[0023] S33: A set of health indicators that are highly correlated with the battery health status is screened out through correlation analysis based on the Pearson correlation coefficient PCC and the grey relational gradient GRG.

[0024] Further, the health indicator set includes current, battery pack voltage, state of charge, maximum and minimum battery voltage, maximum and minimum battery probe temperature, battery voltage difference, probe temperature and battery capacity calculated by the ampere integral formula;

[0025] In order to ensure the diversity of health indicator combinations, two health indicator sets are selected to simultaneously establish a battery health state estimation model to achieve battery health state estimation.

[0026] Furthermore, the S4 is specifically:

[0027] S41: establishing independent pseudo battery health state estimation models based on the two selected health indicator sets;

[0028] S42: Based on a multi-layer perceptron, the estimated pseudo battery health status is integrated into the enhanced training set of another battery health status estimation model, and an independent battery health status estimation model is established using a semi-supervised regression and collaborative training method;

[0029] Furthermore, the multilayer perceptron is composed of a plurality of fully connected layers, wherein each layer is composed of a plurality of units, and these units are interconnected with all units in the previous layer.

[0030] Furthermore, the S5 is specifically as follows:

[0031] S51: Input the test data after the health indicator set is extracted into the trained battery health state estimation model to estimate the battery health state;

[0032] S52: The average value estimated by the battery health state estimation model using the enhanced training set is considered as the final estimated battery health state.

[0033] The beneficial effects of the present invention are:

[0034] (1) The semi-supervised regression and collaborative training method can effectively utilize labeled and unlabeled data to improve the accuracy and robustness of battery health status estimation in scenarios with limited labeled data.

[0035] (2) The method of combining the ampere-hour integral formula and the open circuit voltage correction strategy in the present invention can reduce the battery health status calculation error caused by inaccurate state of charge.

[0036] (3) The present invention can overcome the obstacles of on-site battery health status datasets caused by the complex conditions of electric vehicles and the high cost of data collection, accurately estimate the battery health status, and thus ensure timely maintenance and safe operation of the battery pack.

[0037] (4) The present invention retains the diversity of health indicators and provides multi-dimensional correlation criteria to determine the correlation between battery health status and multiple health indicators.

[0038] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, in which:

[0040] Figure 1 A flow chart of the overall method of the present invention;

[0041] Figure 2 This is a schematic diagram of the analysis results of the correlation between the health indicator set and the battery health status in the embodiment; Figure 2 (a) is a schematic diagram of the correlation analysis results based on the Pearson correlation coefficient PCC; Figure 2 (b) is a schematic diagram of the correlation analysis results based on grey relational gradient (GRG);

[0042] Figure 3 It is the overall framework diagram of the embodiment method;

[0043] Figure 4 Create a flow chart for the embodiment model. DETAILED DESCRIPTION

[0044] The following describes the embodiments of the present invention by means of specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways 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 illustrations of the basic concept of the present invention, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0045] Among them, the accompanying drawings are only for illustrative purposes and represent only schematic diagrams rather than actual pictures, and should not be understood as limiting the present invention. In order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the dimensions of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions may be omitted in the accompanying drawings.

[0046] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "back", etc. indicating directions or positional relationships, they are based on the directions or positional relationships 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 direction, be constructed and operate in a specific direction. Therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0047] See also Figure 1 ,A battery health state estimation method based on semi-supervised regression and ,co-training can be divided into the following steps:

[0048] S1: Collect the operating data of electric vehicles, including the operating data of each cell in the battery pack, and establish a battery operation database;

[0049] S2: Analyzes the charging data of electric vehicles, selects charging segments that meet the requirements, and calculates the battery health status label using the ampere integral formula and open circuit voltage correction strategy;

[0050] S3: Extract a set of health indicators that are highly correlated with the battery health status based on the Pearson correlation coefficient (PCC) and the grey relational gradient (GRG) indicator respectively;

[0051] S4: Based on the selected health indicator set, semi-supervised co-training is used to estimate pseudo labels of unlabeled data to enhance the training dataset, and a battery health state estimation model is constructed based on a multi-layer perceptron (MLP);

[0052] S5: The test data after extracting the health indicator set is used as the input of the battery health state estimation model, and all regressors are integrated to obtain the estimated battery health state.

[0053] As an optional embodiment, the complete technical roadmap of this solution is as follows: Figure 3 shown.

[0054] As an optional embodiment, the above S1 specifically includes S11-S13:

[0055] S11: Collect operating data of a certain electric vehicle battery pack, including charge and discharge current, voltage, temperature, time, and power parameters;

[0056] S12: Collecting cell operating data in the battery pack, including cell voltages of all cells and temperature data sets of each temperature probe placed near the cells;

[0057] S13: Based on the collected operating data of the battery cells and battery packs, a battery pack operating database for a certain vehicle is established.

[0058] As an optional embodiment, the above S2 specifically includes S21-S23:

[0059] S21: analyzing the charging data of the electric vehicle and screening the charging segments that meet the rules of the open circuit voltage correction strategy;

[0060] S22: Collecting the open circuit voltage-state of charge lookup table provided by the battery manufacturer, adopting the open circuit voltage correction strategy, and correcting the initial battery state of charge according to the minimum battery voltage at the initial time;

[0061] S23: Based on the historical operating data of the battery pack, the maximum available capacity of the battery pack is calculated using the ampere integral formula to obtain a battery health status label.

[0062] As an optional embodiment, an open circuit voltage-state of charge lookup table provided by the battery manufacturer is collected, and an open circuit voltage correction strategy is adopted to correct the initial battery state of charge according to the minimum battery voltage at the initial time, specifically including:

[0063] First, a 2-hour standing charge process was selected to eliminate the influence of polarization effects and obtain an accurate open-circuit voltage value. Then, the terminal state of charge was set to 100% to ensure that the battery was fully charged and the terminal state of charge was accurate. In addition, the initial state of charge was set to less than 30% to mitigate the impact of voltage fluctuations on the voltage platform. Finally, the open-circuit voltage-state of charge lookup table was used to correct the initial state of charge according to the minimum battery voltage at the initial time.

[0064] The battery health status tag is obtained in the following ways:

[0065] The health status of a battery pack is defined as the ratio of the current maximum available capacity of the battery pack to its rated capacity, as shown in the following formula:

[0066]

[0067] The charging process using field data is calculated based on the Ampere integral formula Get the battery health status label:

[0068]

[0069] Where ΔQ(i) and ΔSOC(i) represent the cumulative charge and state-of-charge change in the i-th charging segment, respectively. SOC(t1) and SOC(t2) represent the initial state-of-charge at t1 and the final state-of-charge at t2, respectively.

[0070] As an optional embodiment, the above S3 specifically includes S31-S33:

[0071] S31: performing a charging curve analysis based on the maximum voltage of the battery for the charging segment of the battery pack to determine a voltage range for extracting the health indicator set;

[0072] S32: Calculate statistical characteristics of the charging data, including the average value and standard deviation, based on the selected voltage range, and extract a health indicator set;

[0073] S33: A set of health indicators that are highly correlated with the battery health status is screened out through correlation analysis based on the Pearson correlation coefficient (PCC) and grey relational gradient (GRG).

[0074] As an optional embodiment, the health indicator set includes current, battery pack voltage, state of charge, maximum and minimum battery voltages, maximum and minimum battery probe temperatures, battery voltage difference, probe temperature, and battery capacity calculated by an ampere integral formula;

[0075] In order to ensure the diversity of health indicator combinations, two health indicator sets are selected to simultaneously establish a battery health state estimation model to achieve battery health state estimation.

[0076] As an optional embodiment, two commonly used indicators, Pearson correlation coefficient (PCC) and grey relation gradient (GRG), are used to evaluate the correlation between the battery health status and the health indicator set, such as Figure 2 FIG. 1 is a schematic diagram showing the analysis of the correlation between the health indicator set and the battery health status according to an embodiment; Figure 2 (a) is a schematic diagram of the correlation analysis results based on the Pearson correlation coefficient (PCC); Figure 2 (b) is a schematic diagram of the correlation analysis results based on grey relational gradient (GRG);

[0077] The Pearson correlation coefficient (PCC) provides information about the direction and strength of linear correlation. The calculation formula is as follows:

[0078]

[0079] Where x i represents the i-th health indicator, y represents the health status of the marker, and Represent the average values ​​of the corresponding health indicator series and health status series respectively.

[0080] Grey relational gradient (GRG) provides a quantitative assessment of system evolution, and the calculation formula can be defined as follows:

[0081]

[0082]

[0083] Where, ξ i (k) is the grey correlation coefficient between the i-th health index and the marked health status at time k, and ρ represents the discrimination coefficient, which is usually set to 0.5.

[0084] As an optional embodiment, in order to select effective health indicators with high correlation as inputs of the health status estimation model, the high correlation threshold is set to 0.8.

[0085] As an optional embodiment, the above S4 specifically includes S41-S42:

[0086] S41: establishing independent pseudo battery health state estimation models based on the two selected health indicator sets;

[0087] S42: Based on a multi-layer perceptron, the estimated pseudo battery health status is integrated into the enhanced training set of another battery health status estimation model, and an independent battery health status estimation model is established using a semi-supervised regression and collaborative training method;

[0088] As an optional embodiment, the multilayer perceptron is composed of multiple fully connected layers, wherein each layer is composed of multiple units, and these units are interconnected with all units of the previous layer, that is, a unit j∈(1,2,…,J l ) In any hidden layer l∈(0,1,…,L), the mathematical formula for propagation from one unit to another is as follows:

[0089]

[0090] h j,l =a l (z j,l )

[0091] Where w, b, and a represent unit weight, bias factor, and activation function, respectively.

[0092] As an optional embodiment, the mean absolute error and the root mean square error are used to evaluate the battery health status estimation effect.

[0093] The specific technical implementations described in S2-S4 are as follows Figure 4 shown.

[0094] As an optional embodiment, S5 specifically includes S51-S52:

[0095] S51: Input the test data after the health indicator set is extracted into the trained battery health state estimation model to estimate the battery health state;

[0096] S52: The average value estimated by the battery health state estimation model using the enhanced training set is considered as the final estimated battery health state.

[0097] As an optional embodiment, the semi-supervised regression and co-training method will perform semi-supervised co-training based on two selected health indicator sets, assign pseudo battery health status to unlabeled data to enhance the training set, and finally fuse all regressors to obtain the estimated battery health status.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A battery health state estimation method based on semi-supervised regression and collaborative training, characterized by: The method comprises the following steps: S1: Collect the operating data of electric vehicles, including the operating data of each cell in the battery pack, and establish a battery operation database; S2: Analyzes the charging data of electric vehicles, selects charging segments that meet the requirements, and calculates the battery health status label using the ampere integral formula and open circuit voltage correction strategy; S3: Extract a set of health indicators that are highly correlated with the battery health status based on the Pearson correlation coefficient PCC and the grey relational gradient GRG indicator; specifically: S31: performing a charging curve analysis based on the maximum voltage of the battery for the charging segment of the battery pack to determine a voltage range for extracting the health indicator set; S32: Calculate statistical characteristics of the charging data, including the average value and standard deviation, based on the selected voltage range, and extract a health indicator set; S33: Screen out a set of health indicators that are highly correlated with the battery health status through correlation analysis based on Pearson correlation coefficient PCC and grey relational gradient GRG respectively; S4: Based on the selected health indicator set, semi-supervised co-training is used to estimate pseudo labels of unlabeled data to enhance the training dataset, and a battery health state estimation model is constructed based on a multi-layer perceptron (MLP). Specifically: S41: establishing independent pseudo battery health state estimation models based on the two selected health indicator sets; S42: Based on a multi-layer perceptron, the estimated pseudo battery health status is integrated into the enhanced training set of another battery health status estimation model, and an independent battery health status estimation model is established using a semi-supervised regression and collaborative training method; S5: The test data after extracting the health indicator set is used as the input of the battery health state estimation model, and all regressors are integrated to obtain the estimated battery health state.

2. The method for estimating battery health status based on semi-supervised regression and collaborative training according to claim 1, characterized in that: The S1 is specifically: S11: Collect operating data of a certain electric vehicle battery pack, including charge and discharge current, voltage, temperature, time, and power parameters; S12: Collecting cell operating data in the battery pack, including cell voltages of all cells and temperature data sets of each temperature probe placed near the cells; S13: Based on the collected operating data of the battery cells and battery packs, a battery pack operating database for a certain vehicle is established.

3. The method for estimating battery health status based on semi-supervised regression and collaborative training according to claim 2, characterized in that: The S2 is specifically: S21: analyzing the charging data of the electric vehicle and screening the charging segments that meet the rules of the open circuit voltage correction strategy; S22: Collecting the open circuit voltage-state of charge lookup table provided by the battery manufacturer, adopting the open circuit voltage correction strategy, and correcting the initial battery state of charge according to the minimum battery voltage at the initial time; S23: Based on the historical operating data of the battery pack, the maximum available capacity of the battery pack is calculated using the ampere integral formula to obtain a battery health status label.

4. The method for estimating battery health status based on semi-supervised regression and collaborative training according to claim 1, characterized in that: The health indicator set includes current, battery pack voltage, state of charge, maximum and minimum battery voltage, maximum and minimum battery probe temperature, battery voltage difference, probe temperature and battery capacity calculated by the ampere integral formula; In order to ensure the diversity of health indicator combinations, two health indicator sets are selected to simultaneously establish a battery health state estimation model to achieve battery health state estimation.

5. The method for estimating battery health status based on semi-supervised regression and collaborative training according to claim 1, characterized in that: The multilayer perceptron consists of multiple fully connected layers, where each layer consists of multiple units, and these units are interconnected with all units in the previous layer.

6. The method for estimating battery health status based on semi-supervised regression and collaborative training according to claim 1, characterized in that: The S5 is specifically: S51: Input the test data after the health indicator set is extracted into the trained battery health state estimation model to estimate the battery health state; S52: The average value estimated by the battery health state estimation model using the enhanced training set is considered as the final estimated battery health state.

Images