A method and system for health state detection based on the battery capacity increment curve

Abstract

The present invention provides a health state detection method and system based on a capacity increment curve in the field of battery detection technology. The method includes the following steps: Step S10, obtaining a large amount of charging data of a lithium battery and constructing a charging data set based on each of the charging data; Step S20, calculating a capacity increment curve based on the charging data set; Step S30, preprocessing and filtering the capacity increment curve; Step S40, using a time series feature extraction tool to extract features from the filtered capacity increment curve; Step S50, sorting and screening the importance of the extracted features to obtain a feature data set; Step S60, creating a health state detection model, training the health state detection model using the feature data set, and performing online health state detection of the lithium battery using the trained health state detection model. The advantages of the present invention are as follows: greatly improving the accuracy and generalization of the health state detection of lithium batteries.

Description

Technical Field

[0001] The present invention relates to the technical field of battery detection, and particularly to a method and system for detecting the state of health based on the battery capacity increment curve. Background Art

[0002] The state of health (SOH) of a lithium battery is defined as the percentage between the current available capacity and the initial capacity of the lithium battery. During the charge and discharge cycles of the lithium battery, some irreversible chemical reactions will occur inside, which will lead to the aging of the lithium battery and inevitably some potential safety hazards. Therefore, the state of health of the lithium battery is a key indicator.

[0003] Since there is a voltage plateau with a slow change in the amount of electricity during the charging and discharging processes of the lithium battery, it is not conducive to observing minute changes (changes in voltage / current) during the voltage plateau period, which will affect the detection of the state of health of the lithium battery. The incremental capacity method (IC curve method) can convert the voltage plateau into a dQ / dV peak that is easy to observe, so that minute changes that are not easily found on the voltage curve can be reflected on the incremental capacity curve. Therefore, the incremental capacity method is widely used in the detection of the state of health of lithium batteries.

[0004] However, the traditional incremental capacity method has the following disadvantages: 1. Manually extracting features such as the peak value and its position of the incremental capacity curve, the distance between different peaks / valleys, the peak area, and the half-height area of the right peak for subsequent state of health detection has the problems of being unable to fully explore the essential features of the incremental capacity curve and unable to ensure that the features are definitely located within the SOC interval of online charging, which directly affects the detection accuracy; 2. Since the charging condition is generally more controllable and stable than the discharging condition, the state of health of the lithium battery is generally detected under the charging condition; however, when the lithium battery is charging, especially when a new energy vehicle is charging, there are situations where the charging SOC interval is not fixed, the starting point of the charging SOC is not fixed, and the charging rate is not fixed, which may lead to the situation that the feature calculation area does not exist (such as the curve peak does not fall within the charging SOC interval) during the state of health detection process, resulting in the failure of the state of health detection.

[0005] Therefore, how to provide a method and system for detecting the state of health based on the battery capacity increment curve to improve the accuracy and generalization of the state of health detection of lithium batteries has become an urgent technical problem to be solved. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for detecting the state of health based on the battery capacity increment curve to improve the accuracy and generalization of the state of health detection of lithium batteries.

[0007] In a first aspect, the present invention provides a method for detecting the health state based on the battery capacity increment curve, including the following steps:

[0008] Step S10: Obtain a large amount of charging data of the lithium battery, and construct a charging data set based on each of the charging data;

[0009] Step S20: Calculate the capacity increment curve based on the charging data set;

[0010] Step S30: Preprocess and filter the capacity increment curve;

[0011] Step S40: Use a time series feature extraction tool to extract features from the filtered capacity increment curve;

[0012] Step S50: Sort and screen the importance of the extracted features to obtain a feature data set;

[0013] Step S60: Create a health state detection model, train the health state detection model using the feature data set, and use the trained health state detection model to perform online health state detection of the lithium battery.

[0014] Further, in the step S10, the charging data includes the voltage value, current value, SOC value, and SOH value of the lithium battery during full charge and full discharge;

[0015] The charging data set is constructed by randomly extracting the charging data in any SOC interval under the charging industrial control;

[0016] In the step S20, the calculation formula of the capacity increment curve is as follows:

[0017] IC = dQ / dV;

[0018] Wherein, IC represents the value of the capacity increment curve; Q represents the charging power; V represents the voltage; dV represents the preset voltage interval; dQ represents the capacity increment change value within the preset voltage interval;

[0019] The step S30 is specifically:

[0020] Set a voltage range, replace the values in the capacity increment curve that exceed the voltage range with 0, thereby completing the preprocessing of the capacity increment curve;

[0021] Filter the preprocessed capacity increment curve by the moving average filtering method or the Gaussian filtering method.

[0022] Further, the step S40 is specifically:

[0023] The time series feature extraction tool tsfresh is used to extract features from the filtered capacity increment curve, and then the input X is converted into the output X';

[0024] X = {IC, SOC};

[0025] X' = {tsfresh(IC) 1-64 , SOC min , SOC max};

[0026] Among them, tsfresh(IC) 1-64 represents extracting features of the tsfresh time series from the capacity increment curve to obtain 64 - dimensional features; SOC min represents the starting value of SOC during charging; SOC max represents the ending value of SOC during charging.

[0027] Furthermore, the step S50 specifically includes:

[0028] Step S51: After randomly arranging the extracted features, merge them with the initial features;

[0029] Step S52: Calculate the importance degree of each feature through the random forest algorithm, and judge whether the number of repetitions is greater than the preset threshold. If so, enter step S53; if not, enter step S51;

[0030] Step S53: Construct a feature data set based on the features whose importance degree has changed.

[0031] Furthermore, in the step S60, the health state detection model is created based on LightGBM or xgboost.

[0032] In the second aspect, the present invention provides a health state detection system based on a battery capacity increment curve, including the following modules:

[0033] A charging data set construction module, configured to obtain a large amount of charging data of a lithium - ion battery and construct a charging data set based on each of the charging data;

[0034] A capacity increment curve calculation module, configured to calculate a capacity increment curve based on the charging data set;

[0035] A capacity increment curve processing module, configured to pre - process and filter the capacity increment curve;

[0036] A feature extraction module, configured to extract features from the filtered capacity increment curve by using a time series feature extraction tool;

[0037] A feature dataset construction module for sorting and screening the importance of the extracted features to obtain a feature dataset;

[0038] A health status detection module for creating a health status detection model, training the health status detection model using the feature dataset, and performing online health status detection of the lithium battery using the trained health status detection model.

[0039] Further, in the charging dataset construction module, the charging data includes voltage values, current values, SOC values, and SOH values during full charge and full discharge of the lithium battery;

[0040] The charging dataset is constructed by randomly extracting charging data in any SOC interval under charging industrial control;

[0041] In the capacity increment curve calculation module, the calculation formula of the capacity increment curve is as follows:

[0042] IC = dQ / dV;

[0043] Where IC represents the value of the capacity increment curve; Q represents the charging power; V represents the voltage; dV represents the preset voltage interval; dQ represents the capacity increment change value within the preset voltage interval;

[0044] The capacity increment curve processing module is specifically:

[0045] Set a voltage range, replace the values in the capacity increment curve that exceed the voltage range with 0, and thus complete the preprocessing of the capacity increment curve;

[0046] Filter the preprocessed capacity increment curve by moving average filtering method or Gaussian filtering method.

[0047] Further, the feature extraction module is specifically:

[0048] Use the time series feature extraction tool tsfresh to extract features from the filtered capacity increment curve, and then convert the input X to the output X';

[0049] X = {IC, SOC};

[0050] X' = {tsfresh(IC) 1-64 , SOC min , SOC max};

[0051] Where tsfresh(IC) 1-64 represents extracting features of the tsfresh time series from the capacity increment curve to obtain 64 - dimensional features; SOC minrepresents the starting value of SOC during charging; SOC max represents the ending value of SOC during charging.

[0052] Further, the feature dataset construction module specifically includes:

[0053] A feature recombination unit, which is used to randomly arrange the extracted features and then merge them with the initial features;

[0054] An importance calculation unit, which is used to calculate the importance of each feature through the random forest algorithm, judge whether the number of repetitions is greater than a preset threshold. If so, it enters the feature screening unit; if not, it enters the feature recombination unit;

[0055] A feature screening unit, which is used to construct a feature dataset based on the features whose importance has changed.

[0056] Further, in the health state detection module, the health state detection model is created based on LightGBM or xgboost.

[0057] The advantages of the present invention are as follows:

[0058] By calculating the capacity increment curve from the charging dataset and performing preprocessing and filtering, using the time series feature extraction tool to automatically extract features from the filtered capacity increment curve, replacing the traditional manual feature extraction, and sorting and screening the extracted features to obtain a feature dataset, and then using the feature dataset to train the health state detection model to ensure the accuracy and generalization of the health state detection model, and converting the feature extraction of the time series signal of the charging data into the feature extraction of the capacity increment curve, which helps to eliminate the problem of inconsistent data acquisition frequencies during online health state detection, and finally greatly improves the accuracy and generalization of the lithium battery health state detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The present invention will be further described below with reference to the accompanying drawings in conjunction with embodiments.

[0060] Figure 1 is a flowchart of a health state detection method based on the battery capacity increment curve of the present invention.

[0061] Figure 2 is a structural schematic diagram of a health state detection system based on the battery capacity increment curve of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The overall idea of the technical solution in the embodiments of this application is as follows: Automatically extract features through a timing feature extraction tool instead of traditional manual feature extraction, and use the feature dataset obtained by importance ranking and screening to train a health status detection model to ensure the accuracy and generalization of the health status detection model; Convert the feature extraction of the timing signal of the charging data into the feature extraction of the capacity increment curve, eliminate the problem of inconsistent data acquisition frequencies, and further ensure the accuracy of the lithium battery health status detection.

[0063] Please refer to Figures 1 to 2 as shown in the figure, a preferred embodiment of a health status detection method based on a battery capacity increment curve according to the present invention includes the following steps:

[0064] Step S10: Obtain a large amount of charging data of the lithium battery and construct a charging dataset based on each piece of the charging data;

[0065] Step S20: Calculate a capacity increment curve based on the charging dataset;

[0066] Step S30: Preprocess and filter the capacity increment curve;

[0067] Step S40: Use a timing feature extraction tool to automatically extract features from the filtered capacity increment curve;

[0068] Step S50: Rank and screen the importance of the extracted features to obtain a feature dataset;

[0069] Step S60: Create a health status detection model, use the feature dataset to train the health status detection model, and use the trained health status detection model to perform online health status detection of the lithium battery, that is, use the charging data in any SOC interval collected during online charging as the input of the health status detection model, and output the predicted value of the health status detection model as the online health status detection value of the lithium battery.

[0070] In the step S10, the charging data includes the voltage value, current value, SOC value, and SOH value of the lithium battery during full charge and full discharge;

[0071] The charging dataset is constructed by randomly extracting the charging data in any SOC interval under the charging industrial control;

[0072] Since the discharge conditions of lithium batteries in actual applications are often more complex than the charging conditions, and the charging data of the charging conditions is also easier to obtain (for example, obtaining charging condition data when charging new energy vehicles), the health state detection of the present invention is based on the charging conditions of lithium batteries. Since the true value of the current health state of the lithium battery cannot be obtained during on-line charging, and the laboratory test contains a large amount of time series data such as voltage and current and the true value of SOH under different full charge and full discharge cycle numbers, the charging data of the present invention is obtained based on laboratory data.

[0073] In the step S20, the calculation formula of the capacity increment curve is as follows:

[0074] IC = dQ / dV;

[0075] where IC represents the value of the capacity increment curve; Q represents the charging quantity; V represents the voltage; dV represents the preset voltage interval, and the preferred value is 0.05V; dQ represents the capacity increment change value within the preset voltage interval;

[0076] The step S30 is specifically as follows:

[0077] Set a voltage range, and replace the values in the capacity increment curve that exceed the voltage range with 0, thereby completing the preprocessing of the capacity increment curve;

[0078] Different charging SOC intervals correspond to different voltage rising intervals. Therefore, in the calculation of the capacity increment curve, by setting the maximum voltage range, multiple IC curve segments with different voltage interval ranges and their corresponding SOH values are obtained, and the IC values in the non-charging voltage rising interval range are automatically filled with 0; through the preprocessing of the capacity increment curve, the sampling frequencies (1s, 3s, 10s, etc.) of different charging devices can be converted into IC curves within the same SOC voltage range;

[0079] Filter the preprocessed capacity increment curve by the moving average filtering method or the Gaussian filtering method;

[0080] Since the obtained capacity increment curve is a set of discrete points, it is necessary to filter to obtain a smooth capacity increment curve.

[0081] The step S40 is specifically as follows:

[0082] The time series feature extraction tool tsfresh is used to extract features from the filtered capacity increment curve, and then the input X is converted into the output X'; tsfresh is a Python time series data feature extraction module that can automatically calculate a large number of time series features, including a variety of statistical and transformation-based time series extraction methods, such as absolute energy value, absolute sum of first-order differences, aggregated statistical features of autocorrelation coefficients of various orders, approximate entropy, autoregressive coefficient, ADF test, lag-order autocorrelation, etc.;

[0083] X = {IC, SOC};

[0084] X'={tsfresh(IC) 1-64 ,SOC min ,SOC max};

[0085] Among them, tsfresh(IC) 1-64 It means extracting the feature of tsfresh time series from the capacity increment curve to obtain 64-dimensional features; SOC min Indicates the starting value of SOC during charging; SOC max Indicates the end value of SOC during charging.

[0086] The step S50 specifically includes:

[0087] Step S51, after arranging the extracted features in random order, merge them with the initial features;

[0088] Step S52, calculate the importance of each feature through the random forest algorithm, and determine whether the number of repetitions is greater than a preset threshold. If so, proceed to step S53; if not, proceed to step S51; the threshold is preferably 100 times;

[0089] Step S53, constructing a feature data set based on the features whose importance has changed;

[0090] That is, the extracted features are sorted and screened by importance through the Boruta method; since the number of output tsfresh time series features is large and the amount of calculation is large, in order to improve the algorithm operation efficiency and avoid extracting irrelevant features, it is necessary to sort the calculated features by importance. The general method uses the Pearson correlation coefficient to sort the feature importance. The closer the obtained correlation coefficient is to 1, the higher the importance; however, the feature importance sorting based on the Pearson correlation coefficient can only calculate the correlation of a single feature, and cannot mine the correlation between multiple features, and each feature needs to be calculated separately, which has calculation redundancy. Therefore, the present invention adopts a feature importance selection strategy based on the Boruta method.

[0091] The goal of Boruta is to select all the feature sets related to the dependent variable, rather than selecting the feature set that can minimize the model cost function for a specific model. The significance of the Boruta algorithm lies in being able to more comprehensively understand the influencing factors of the dependent variable, so as to perform feature selection better and more efficiently.

[0092] In the step S60, the health status detection model is created based on LightGBM or xgboost.

[0093] A preferred embodiment of a health status detection system based on the battery capacity increment curve according to the present invention includes the following modules:

[0094] A charging data set construction module, configured to obtain a large amount of charging data of a lithium battery and construct a charging data set based on each of the charging data;

[0095] A capacity increment curve calculation module, configured to calculate a capacity increment curve based on the charging data set;

[0096] A capacity increment curve processing module, configured to preprocess and filter the capacity increment curve;

[0097] A feature extraction module, configured to automatically extract features from the filtered capacity increment curve by using a time series feature extraction tool;

[0098] A feature data set construction module, configured to rank and screen the importance of the extracted features to obtain a feature data set;

[0099] A health status detection module, configured to create a health status detection model, train the health status detection model by using the feature data set, and perform online health status detection of the lithium battery by using the trained health status detection model, that is, using the charging data of any SOC interval collected during online charging as the input of the health status detection model, and outputting the predicted value of the health status detection model as the online health status detection value of the lithium battery.

[0100] In the charging data set construction module, the charging data includes the voltage value, current value, SOC value, and SOH value of the lithium battery during full charge and full discharge;

[0101] The charging data set is constructed by randomly extracting the charging data of any SOC interval under the charging industrial control;

[0102] Since the discharge conditions of lithium batteries in actual applications are often more complex than the charging conditions, and the charging data of the charging conditions is also easier to obtain (for example, obtaining charging condition data when charging new energy vehicles), the health state detection of the present invention is based on the charging conditions of lithium batteries. Since the true value of the current health state of the lithium battery cannot be obtained during online charging, and the laboratory test contains a large amount of time series data such as voltage and current and the true value of SOH under different full charge and full discharge cycle times, the charging data of the present invention is obtained based on laboratory data.

[0103] In the capacity increment curve calculation module, the calculation formula of the capacity increment curve is as follows:

[0104] IC = dQ / dV;

[0105] Where, IC represents the value of the capacity increment curve; Q represents the charging amount; V represents the voltage; dV represents the preset voltage interval, and the preferred value is 0.05V; dQ represents the capacity increment change value within the preset voltage interval;

[0106] The capacity increment curve processing module is specifically:

[0107] Set a voltage range, and replace the values in the capacity increment curve that exceed the voltage range with 0, thereby completing the preprocessing of the capacity increment curve;

[0108] Different charging SOC intervals correspond to different voltage rise intervals. Therefore, in the calculation of the capacity increment curve, by setting the maximum voltage range, multiple groups of IC curve segments with different voltage interval ranges and their corresponding SOH values are obtained, and the IC values in the non-charging voltage rise interval range are automatically filled with 0; through the preprocessing of the capacity increment curve, the sampling frequencies (1s, 3s, or 10s, etc.) of different charging devices can be converted into IC curves within the same SOC voltage range;

[0109] Filter the preprocessed capacity increment curve by the moving average filtering method or the Gaussian filtering method;

[0110] Since the obtained capacity increment curve is a set of discrete points, it is necessary to filter to obtain a smooth capacity increment curve.

[0111] The feature extraction module is specifically:

[0112] The time series feature extraction tool tsfresh is used to extract features from the filtered capacity increment curve, and then the input X is converted into the output X'; tsfresh is a Python time series data feature extraction module that can automatically calculate a large number of time series features, including a variety of statistical and transformation-based time series extraction methods, such as absolute energy value, absolute sum of first-order differences, aggregated statistical features of autocorrelation coefficients of various orders, approximate entropy, autoregressive coefficient, ADF test, lag-order autocorrelation, etc.;

[0113] X = {IC, SOC};

[0114] X'={tsfresh(IC) 1-64 ,SOC min ,SOC max};

[0115] Among them, tsfresh(IC) 1-64 It means extracting the feature of tsfresh time series from the capacity increment curve to obtain 64-dimensional features; SOC min Indicates the starting value of SOC during charging; SOC max Indicates the end value of SOC during charging.

[0116] The feature data set construction module specifically includes:

[0117] A feature recombination unit is used to shuffle the extracted features and then merge them with the initial features;

[0118] The importance calculation unit is used to calculate the importance of each feature through the random forest algorithm, and determine whether the number of repetitions is greater than a preset threshold. If so, it enters the feature screening unit; if not, it enters the feature recombination unit; the threshold is preferably 100 times;

[0119] A feature screening unit, used to construct a feature data set based on features whose importance has changed;

[0120] That is, the extracted features are sorted and screened by importance through the Boruta method; since the number of output tsfresh time series features is large and the amount of calculation is large, in order to improve the algorithm operation efficiency and avoid extracting irrelevant features, it is necessary to sort the calculated features by importance. The general method uses the Pearson correlation coefficient to sort the feature importance. The closer the obtained correlation coefficient is to 1, the higher the importance; however, the feature importance sorting based on the Pearson correlation coefficient can only calculate the correlation of a single feature, and cannot mine the correlation between multiple features, and each feature needs to be calculated separately, which has calculation redundancy. Therefore, the present invention adopts a feature importance selection strategy based on the Boruta method.

[0121] The goal of Boruta is to select all the feature sets related to the dependent variable, rather than selecting the feature set that can minimize the model cost function for a specific model. The significance of the Boruta algorithm lies in that it can comprehensively understand the influencing factors of the dependent variable, so as to perform feature selection better and more efficiently.

[0122] In the health status detection module, the health status detection model is created based on LightGBM or xgboost.

[0123] To sum up, the advantages of the present invention are as follows:

[0124] Calculate the capacity increment curve through the charging data set and perform preprocessing and filtering. Use the time series feature extraction tool to automatically extract features from the filtered capacity increment curve, replacing the traditional manual feature extraction. Then, rank and screen the importance of the extracted features to obtain the feature data set. Finally, use the feature data set to train the health status detection model to ensure the accuracy and generalization of the health status detection model. Moreover, converting the feature extraction of the time series signal of the charging data into the feature extraction of the capacity increment curve helps to eliminate the problem of inconsistent data acquisition frequencies during online health status detection, and ultimately greatly improves the accuracy and generalization of the lithium battery health status detection.

[0125] Although the specific implementation manners of the present invention have been described above, those skilled in the art of this technology should understand that the specific embodiments we described are illustrative rather than used to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be covered by the scope protected by the claims of the present invention.

Claims

1. A method for detecting the health state based on the battery capacity increment curve, characterized in that: It includes the following steps: Step S10: Obtain a large amount of charging data of the lithium battery, and construct a charging data set based on each piece of the charging data; the charging data includes the voltage value, current value, SOC value, and SOH value of the lithium battery during full charge and full discharge; the charging data set is constructed by randomly extracting the charging data in any SOC interval under the charging industrial control; Step S20: Calculate the capacity increment curve based on the charging data set; Step S30: Preprocess and filter the capacity increment curve; Step S40: Use a time series feature extraction tool to extract features from the filtered capacity increment curve; Step S50: Sort and screen the extracted features according to importance to obtain a feature data set; Step S60: Create a health state detection model, use the feature data set to train the health state detection model, and use the trained health state detection model to perform on-line health state detection of the lithium battery.

2. A method for detecting the health state based on the battery capacity increment curve according to claim 1, characterized in that: In the step S20, the calculation formula of the capacity increment curve is as follows: IC = dQ / dV; where, IC represents the value of the capacity increment curve; Q represents the charging power; V represents the voltage; dV represents the preset voltage interval; dQ represents the capacity increment change value within the preset voltage interval; The step S30 is specifically: Set a voltage range, replace the values in the capacity increment curve that exceed the voltage range with 0, and thus complete the preprocessing of the capacity increment curve; Filter the preprocessed capacity increment curve by the moving average filtering method or the Gaussian filtering method.

3. A method for detecting the health state based on the battery capacity increment curve according to claim 1, characterized in that: The step S40 is specifically: Use the time series feature extraction tool tsfresh to extract features from the filtered capacity increment curve, and thus convert the input X into the output X'; X = {IC, SOC}; X' = {tsfresh(IC) 1-64 , SOC min , SOC max}; Among them, tsfresh(IC) 1-64 represents performing feature extraction on the capacity increment curve in the tsfresh time series to obtain 64-dimensional features; SOC min represents the starting value of SOC during charging; SOC max represents the ending value of SOC during charging.

4. A method for detecting the health state based on the battery capacity increment curve according to claim 1, characterized in that: The step S50 specifically includes: Step S51: After randomly arranging the extracted features, merge them with the original features; Step S52: Calculate the importance degree of each feature through the random forest algorithm, and judge whether the number of repetitions is greater than the preset threshold. If so, enter step S53; if not, enter step S51; Step S53: Construct a feature data set based on the features whose importance degree has changed.

5. A method for detecting the health state based on the battery capacity increment curve according to claim 1, characterized in that: In the step S60, the health state detection model is created based on LightGBM or xgboost.

6. A health state detection system based on the battery capacity increment curve, characterized in that: It includes the following modules: The charging data set construction module is used to obtain a large amount of charging data of lithium batteries and construct a charging data set based on each piece of the charging data; the charging data includes voltage values, current values, SOC values, and SOH values of the lithium batteries during full charge and full discharge; the charging data set is constructed by randomly extracting charging data in any SOC interval under charging industrial control; The capacity increment curve calculation module is used to calculate the capacity increment curve based on the charging data set; The capacity increment curve processing module is used to preprocess and filter the capacity increment curve; The feature extraction module is used to extract features from the filtered capacity increment curve by using a time series feature extraction tool; The feature data set construction module is used to rank and screen the importance of the extracted features to obtain a feature data set; The health state detection module is used to create a health state detection model, train the health state detection model by using the feature data set, and perform online health state detection of the lithium battery by using the trained health state detection model.

7. A health state detection system based on a battery capacity increment curve as described in claim 6, characterized in that: In the capacity increment curve calculation module, the calculation formula of the capacity increment curve is as follows: IC = dQ / dV; where, IC represents the value of the capacity increment curve; Q represents the charging amount; V represents the voltage; dV represents a preset voltage interval; dQ represents the capacity increment change value within the preset voltage interval; The capacity increment curve processing module is specifically: Set a voltage range, replace the values in the capacity increment curve that exceed the voltage range with 0, and thus complete the preprocessing of the capacity increment curve; Filter the preprocessed capacity increment curve by using a moving average filtering method or a Gaussian filtering method.

8. A health state detection system based on a battery capacity increment curve as described in claim 6, characterized in that: The feature extraction module is specifically: Use the time series feature extraction tool tsfresh to extract features from the filtered capacity increment curve, and thus convert the input X into the output X'; X = {IC, SOC}; X' = {tsfresh(IC) 1-64 , SOC min , SOC max}; Among them, tsfresh(IC) 1-64 represents the feature extraction of the capacity increment curve in the tsfresh time series, obtaining 64-dimensional features; SOC min represents the starting value of SOC during charging; SOC max represents the ending value of SOC during charging.

9. A health state detection system based on a battery capacity increment curve as described in claim 6, characterized in that: The feature data set construction module specifically includes: The feature recombination unit is used to randomly arrange the extracted features and then merge them with the initial features; The importance calculation unit is used to calculate the importance of each feature by using a random forest algorithm, judge whether the number of repetitions is greater than a preset threshold, if so, enter the feature screening unit; if not, enter the feature recombination unit; The feature screening unit is used to construct a feature data set based on the features whose importance has changed.

10. A health state detection system based on a battery capacity increment curve as described in claim 6, characterized in that: In the health state detection module, the health state detection model is created based on LightGBM or xgboost.

Images