Battery health state rapid assessment method based on time-frequency domain joint characteristics
Through the battery health status evaluation method of time-frequency domain combined characteristics, combined with time-frequency domain data and machine learning model, the problem of insufficient combination of time-domain and frequency domain characteristics in the existing technology is solved, and the rapid and accurate evaluation of battery health status is achieved. , reduces cost and time and improves evaluation efficiency.
Patent Information
- Application Number
- CN202411050476.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-12
AI Technical Summary
In the evaluation of battery health status, the combination of time domain and frequency domain characteristics is insufficient, resulting in the inability to build a comprehensive battery internal state evaluation model, and the application of electrochemical impedance spectroscopy technology has high cost and inconvenience.
Design a rapid evaluation method for battery health status based on time-frequency domain joint characteristics, through the test process design of step one, the time-frequency domain data acquisition in step two, the joint feature extraction in step three, the construction of evaluation model in step four, and the construction of an evaluation model in step five The deployment application uses machine learning regression estimator and XGBoost model, combines time-frequency domain data to build battery aging rules, remove impurity elements, and deploy them in a fast impedance spectrum analyzer.
It realizes a more comprehensive acquisition of battery status information, improves the robustness and accuracy of the evaluation model, reduces time and labor costs, and improves the efficiency and accuracy of battery health status evaluation.
Smart Images

Figure CN118938058A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of lithium battery status assessment, and in particular to a method for rapid assessment of battery health status based on joint features in time and frequency domains. Background Art
[0002] In the field of new energy batteries, the electrochemical kinetic reaction mechanism characterization characteristics of the battery are often studied, and the equivalent circuit ECM is used to simulate the internal chemical reaction of the battery. The electrochemical impedance spectroscopy technology EIS is used to simulate the electrical characteristics and interface processes of the battery by applying an AC signal to the simulation system and measuring the voltage and current response of the system. The battery voltage, current, internal resistance, temperature and other state information of the battery during the charging and discharging process in a shorter SOC range are obtained, and the internal state of the battery is presented in the above frequency and time domain methods. The above method is generally used to extract the battery health factor and construct a battery aging model. However, in actual applications, the battery often exhibits a more complex form. The health factors extracted by the above method rarely take the two into consideration, which makes it impossible to build a comprehensive evaluation model that characterizes the internal state of the battery.
[0003] In terms of battery health factor exploration, frequency domain features use the high precision and non-destructive (non-invasive) advantages of the original impedance spectrum to characterize battery characteristics. According to the principle of relaxation time distribution (DRT), the EIS frequency domain features are transformed into the time domain to further extract the impedance characteristics of the battery, such as ohmic resistance and Gamma. The DRT method can better distinguish the main time constants of the battery's electrochemical process, accelerate the impedance analysis process, and improve the interpretability and accuracy of time-scale dynamics. The time domain feature studies voltage-related characteristics, which are mainly divided into three categories: voltage, incremental capacity, and temperature-related characteristics. The voltage characteristic analyzes the voltage change, the slope of the voltage curve, the initial and final charging voltage, and the capacity value of the constant current and constant voltage step. The incremental capacity feature is based on the differential relationship between the voltage and capacity at the previous and next moments (which can be adjacent moments or multiple moments) during the charging and discharging process to amplify the change characteristics of the voltage and capacity during the stable operation of the battery. The peak displacement (voltage position) and peak value of the incremental capacity curve are related to the battery capacity attenuation. The battery operating temperature will also affect the battery's health status. Excessive battery temperature will accelerate the generation of battery chemical reactions and accelerate the loss of battery materials.
[0004] In terms of related technical applications, the application of electrochemical impedance spectroscopy (EIS) technology in recent years is different from the past. It is integrated into complex equipment or high-cost electrochemical workstations. In this case, there are often problems such as high cost, bulky equipment that makes it inconvenient to carry, and difficulty in promoting and applying the technology on a large scale. Therefore, the application of this technology in the battery industry is relatively low. Summary of the invention
[0005] The purpose of the present invention is to provide a method for quickly evaluating the health status of a battery based on joint features in the time and frequency domains, so as to solve the problems raised in the above-mentioned background technology.
[0006] To achieve the above object, the present invention provides the following technical solution: a method for rapid evaluation of battery health status based on joint features in time and frequency domains, comprising step one, designing a test process; step two, acquiring data in time and frequency domains; step three, extracting joint features; step four, building an evaluation model; and step five, deploying an application;
[0007] In the above step 1, according to the parameter information such as battery type and capacity, the details of voltage regulation experiment, static test and EIS test experiment are set, and the battery health status assessment test process is designed;
[0008] In the above step 2, the battery is charged to 20% SOC through the capacity distribution cabinet to obtain voltage regulation data, that is, time domain data, and the battery is tested with a broadband current through an electrochemical AC fast impedance spectrum analysis device to obtain EIS data, that is, frequency domain data;
[0009] In the above step 3, the indirect and direct features of voltage, EIS, and DRT are extracted to construct a list of battery preparation health factors, and the impurity elements in the list are removed using the correlation coefficient method, and the list is updated as the time-frequency domain joint features of the model input;
[0010] In the above step 4, a machine learning regression estimator is used as a learning model, and the constructed time-frequency domain features are used as input to learn the battery aging law, and a residual value estimation model based on the joint features of the time-frequency domain is established;
[0011] In the above step five, the time-frequency domain joint feature method and capacity estimation algorithm are deployed in the data analysis module of the fast impedance spectrum analyzer so as to use the learned model to quickly evaluate the battery health status.
[0012] As a further technical solution of the present invention, in the step one, the entire evaluation test process consists of three parts: a charging and discharging stage, a standing stage, and an EIS test stage. In the charging and discharging stage, the battery is first discharged to ensure that it is at 0% SOC, and then charged to 20% SOC after standing. After charging, the battery needs to be stood for 1 hour to allow the internal state of the battery to stabilize. After the standing period is over, the EIS test is performed. The entire test process takes 1 hour and 24 minutes relative to a single battery.
[0013] As a further technical solution of the present invention, in the step 2, the battery is charged to 20% SOC by the charging and discharging equipment, and the working step information is: constant current discharge 1-standing 2-constant current discharge 3-standing 4-constant current charging 5-standing 6, wherein the standing time of standing 2 is 5 minutes, the standing time of standing 4 is 1 minute, and the standing time of standing 6 is 30 seconds. The constant current discharge 1 current is 1C, the constant current discharge 3 current is 0.1C, the charging current is 0.6C, and the pre-charge capacity = rated capacity * 20% SOC. This part of the operation coincides with the battery state of charge required by the gradient battery when it leaves the factory, so it is not necessary to use this part as an additional part. The fast impedance spectrum equipment is set to a frequency range of 10kHz-0.1Hz and uses a set current excitation. The excitation current is set to 5A. The frequency domain data under different aging states are obtained by testing gradient batteries with different aging degrees.
[0014] As a further technical solution of the present invention, in the step three, the training samples are grouped by the DBSCAN clustering algorithm, and the clustering results with less than 3 samples in the group are eliminated. The samples with less than the threshold in the cluster group are regarded as abnormal samples, and the time-frequency domain features of the eliminated samples are extracted to characterize the health factors of the battery. The voltage regulation charging and discharging step in step two is used as the time domain feature extraction method, and the complex impedance information in the frequency range of 10kHz-0.1Hz is obtained by the sweeping frequency method. The frequency logarithmic distribution is taken, and 10 frequency points are taken for each logarithmic scale, for a total of 51 groups. The final combined features in the time-frequency domain are 300-0.1Hz real impedance, charging time between 3.2-3.35V, voltage mean, voltage variance, and IC capacity increment peak.
[0015] As a further technical solution of the present invention, in step 3, the time-frequency domain feature selection method uses the Pearson correlation coefficient method to obtain the positive and negative correlation between each feature and the capacity, and uses the features and capacity to visualize the correlation degree, and eliminates the features with low correlation. The Pearson correlation coefficient calculation formula is:
[0016]
[0017] In the formula and They represent the sample means of the two respectively.
[0018] As a further technical solution of the present invention, in the step three, the time-frequency domain features are eliminated and the range is reduced according to the prior knowledge of the battery equivalent circuit ECM and the correlation analysis results obtained by using the Pearson correlation coefficient method. The prepared health factor feature list before elimination consists of the real and imaginary impedance of the electrochemical impedance spectrum, the γ in the DRT feature calculated by the relaxation time distribution method, the charging time of the voltage regulation curve, the voltage mean, the voltage variance, and the IC incremental capacity peak. The correlation score of each health factor and the capacity is calculated according to the Pearson method. The correlation score is [-1, 1], a negative number is negatively correlated with the capacity, and a positive number is positively correlated with the capacity, so as to further screen out better features.
[0019] As a further technical solution of the present invention, in the step 4, after constructing a feature list by using a combination of multiple features, the data is divided into a training set and a validation set with a division ratio of 7:3. When the data is small, the leave-one-out method and K-fold cross-validation method can be used for division, and an extreme gradient regression tree XGBoost model is constructed to learn the battery aging law and output the existing capacity residual value of the battery. The verification estimation results are used through a variety of model evaluation indicators to indicate the degree of model learning. The evaluation indicators used here are: model accuracy r2, relative root mean square percentage error RRMSE, mean absolute percentage error MAPE, and maximum relative error.
[0020] As a further technical solution of the present invention, in step 4, the calculation formula of the model accuracy determination coefficient is:
[0021]
[0022] The relative RMS percentage error is calculated as:
[0023]
[0024] The mean absolute percentage error is calculated as:
[0025]
[0026] The maximum relative error calculation formula is:
[0027]
[0028] As a further technical solution of the present invention, in the step 4, in the battery capacity estimation algorithm part, the xgboost regression model is trained by serially connecting multiple weak learners (i.e., regression trees), and each learner will perform further correction learning based on the learning residual results of the previous learner to gradually improve the estimation ability of the model. Each round of iteration constructs a new regression tree model by optimizing the objective function. During the construction process, the greedy optimization method is used to minimize the objective function, and regularization is used to avoid overfitting.
[0029] As a further technical solution of the present invention, in step five, the online application process consists of sampling, data processing, algorithm estimation, and result visualization output. The corresponding hardware and software implementation consists of an electrochemical impedance spectroscopy excitation sampling module, a Raspberry Pi: data storage, transmission and algorithm deployment, and a visualization result host computer. The best model is selected and saved as a pkl format file through the pickle library function. The model is deployed on the edge device Raspberry Pi, the model is called by building a FastAPI interface, and the algorithm is tested online to evaluate the battery capacity. A large amount of battery information is collected, and the model is continuously retrained and optimized.
[0030] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention collects the time-frequency domain data characterizing the internal state of the battery by respectively performing voltage regulation curve test and electrochemical AC impedance spectrum test through charging and discharging equipment and Zhiyun battery cell AC impedance analyzer, and intercepts the voltage mean, charging time, indirect health factor of voltage variance and incremental capacity IC of the 3.2-3.3V relatively obvious slope change stage data of the battery voltage regulation curve interval, calculates the correlation with capacity by the Pearson correlation coefficient method, further eliminates interference factors, optimizes the list of health factors, and constructs a battery residual value estimation method by integrating time-frequency domain features. The model can learn more comprehensive status information and adopts xgboost integrated regression tree as the battery aging model. The integration characteristics and loss inheritance characteristics of the algorithm make the model prediction results more robust, accurate, fast training and better adaptable to nonlinear characteristics. The learned aging model is deployed in the Raspberry Pi of the Zhiyun battery cell AC impedance analyzer to realize real-time battery data collection and call the data and model API interface through the host computer to quickly estimate the battery capacity residual value. Compared with the traditional capacity division method to obtain the battery capacity residual value, the AC impedance device greatly saves time and labor costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 is a flow chart of the method of the present invention;
[0032] Figure 2 It is a schematic diagram of the battery experimental working condition of the present invention;
[0033] Figure 3 It is the voltage regulation curve and the interception interval diagram;
[0034] Figure 4 Nyquist and Bode plots of electrochemical impedance spectroscopy;
[0035] Figure 5 is the relaxation time distribution characteristic diagram;
[0036] Figure 6 It is the heat map of the real part-capacity Pearson correlation coefficient score;
[0037] Figure 7 It is the heat map of the imaginary part-capacity Pearson correlation coefficient score;
[0038] Figure 8 Heat map of Ohmic impedance-capacity Pearson correlation coefficient scores;
[0039] Fig. 9 Heat map of the voltage-charge time-capacity Pearson correlation coefficient scores;
[0040] Fig.10 It is the heat map of voltage mean-capacity Pearson correlation coefficient score;
[0041] Fig.11 The voltage variance-capacity Pearson correlation coefficient score heat map;
[0042] Fig.12 It is the heat map of IC peak-capacity Pearson correlation coefficient score;
[0043] Fig.13 Evaluation results for the time-frequency domain feature model Figure 1 ;
[0044] Fig.14 Evaluation results for the time-frequency domain feature model Figure 2 ;
[0045] Fig.15 Evaluation results for the time-frequency domain feature model Figure 3 ;
[0046] Fig.16 Evaluation results for the time-frequency domain feature model Figure 4 ;
[0047] Fig.17 Evaluation results for the time-frequency domain feature model Figure 5 ;
[0048] Fig.18 Evaluation results for the time-frequency domain feature model Figure 6 . DETAILED DESCRIPTION
[0049] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0050] Please see attached Figure 1 -Attached Fig.18, an embodiment provided by the present invention: a method for rapid assessment of battery health status based on joint features in time and frequency domains, comprising step one, designing a test process; step two, acquiring data in time and frequency domains; step three, extracting joint features; step four, building an assessment model; step five, deploying an application;
[0051] In the above step 1, the entire evaluation test process consists of three parts: charging and discharging stage, static stage, and EIS test stage. In the charging and discharging stage, the battery is first discharged to ensure that it is at 0% SOC, and then charged to 20% SOC after static. After charging, it needs to be static for 1 hour to allow the internal state of the battery to stabilize. After static, the EIS test is performed. The entire test process takes 1h24min relative to a single battery.
[0052] In the above step 2, the battery is charged at 20% SOC through the capacity distribution cabinet to obtain voltage regulation data, that is, time domain data, and the battery is tested with a broadband current through an electrochemical AC fast impedance spectrum analysis device to obtain EIS data, that is, frequency domain data, to understand the battery information, and the battery is tested through charging and discharging equipment and electrochemical AC fast impedance spectrum analysis equipment to record process data. The experimental plan is arranged according to the battery information (this batch of experimental batteries is 176Ah-LFP batteries, and the battery capacity aging span is about 50Ah) and quantity (a total of 152 batteries), and the battery is charged and discharged The equipment is used to adjust the voltage of the battery to 20% SOC to obtain time domain data, and the Zhiyun electrochemical AC impedance analyzer is used to obtain frequency domain data In the voltage regulation experiment, the battery was first discharged, and then charged to 20% SOC at 0.6C current. The end condition was to charge the battery to 176*0.2Ah using the ampere-hour counting method. After the charging and discharging process, the battery was left to stand for 1min and 30s respectively. The equipment recorded the battery current, voltage, accumulated charging capacity and other data. In the EIS experiment, the battery was left to stand for 1h in a constant temperature environment to achieve a stable internal state of the battery. The test configuration was to excite the battery with a 5A current, and the frequency signal range was 0.1-10kHz. The real and imaginary impedances of 51 points of the frequency logarithm were obtained. The entire test process took 1h24min for the unit number of batteries.
[0053] In the above step three, the voltage indirect feature of the time domain voltage regulation curve is extracted, the impedance feature of the frequency point corresponding to 1000-0.1hz in the frequency domain is extracted, and a joint health feature extraction method based on the original data in the time and frequency domains is constructed; wherein, the time domain feature is composed of the voltage regulation data (charging process of 0-20% SOC), and the method of extracting the time domain health factor is to intercept a section of voltage, charging capacity and other data, and construct indirect features: charging time, voltage variance, voltage mean, IC incremental capacity curve, etc. The method of extracting the frequency domain health factor is to obtain the real impedance and imaginary impedance of EIS through the EIS calculation extraction algorithm through the acquired time domain voltage and current data, and transform the EIS data into a time constant through the relaxation time distribution to obtain the ohmic impedance, τ, γ, so the time and frequency domain health factor is extracted. The preliminary list of factors consists of: charging time, voltage variance, voltage mean, IC incremental capacity curve, real impedance, imaginary impedance, ohmic impedance, τ, and γ. τ is the time constant for elimination. The real and imaginary high-frequency impedance 1000-10000Hz represents the inductive reactance of the battery and is eliminated. The Pearson correlation coefficient algorithm is used to calculate the correlation score between the remaining health factors and the capacity. The calculation results show that the imaginary impedance and IC incremental capacity have a weak relationship with the capacity and are eliminated. The γ feature is characterized by multiple peaks within a certain period of time. The observation intercepts the more obvious peak area or peak-to-peak value as the health factor, but here the relationship between γ and capacity is weak and is eliminated. Therefore, the final list of health factors consists of charging time, voltage variance, voltage mean, real impedance, and ohmic impedance.
[0054] In the above step 4, a residual value estimation model based on the joint features of the time-frequency domain is established, and a battery health status assessment method based on health features and data-driven models is established by using a machine learning regression estimation algorithm; wherein, the xgboost regression model is trained by serially connecting multiple weak learners (i.e., regression trees), and each learner will further correct the learning based on the learning residual results of the previous learner, and gradually improve the estimation ability of the model. Each round of iteration constructs a new regression tree model by optimizing the objective function. During the construction process, the objective function is minimized by using a greedy optimization method, and regularization is used to avoid overfitting; the trained model is used to verify another 30% of the validation set data through the evaluation indicators in the above table, and 70 test set samples are prepared to test the model (not from the same batch of batteries). The evaluation indicators used here are: model accuracy r2 (range [1, -1], a positive result indicates that the model has learned the data law, and the closer it is to 1, the more obvious the data law is), relative root mean square percentage error RRMSE, mean absolute percentage error MAPE, and maximum relative error. The calculation formula is shown in Table 1, and the results of each feature combination model are compared in Table 2;
[0055] In the above step five, the feature extraction method and parameter estimation algorithm are deployed in the data analysis module of the fast impedance spectrum analyzer, the best model is selected and saved as a pkl format file through the pickle library function, the model is deployed on the edge device (Raspberry Pi), the model is called by building a FastAPI interface, and the battery capacity is evaluated by the online test algorithm, a large amount of battery information is collected, and the model is continuously retrained and optimized.
[0056] Table 1 Model evaluation indicators
[0057]
[0058] Table 2 Comparison of model evaluation results
[0059]
[0060] Based on the above, the advantages of the present invention are: the method for extracting battery aging health factors in the time-frequency domain provided by the present invention constructs a time-frequency domain joint learning model to evaluate the battery health status, obtains time-frequency domain original data through charging experiments and EIS experiments, eliminates interference factors through the Pearson correlation coefficient method and battery prior knowledge, and constructs an XGBoost integrated extreme gradient decision tree model to learn the battery aging laws behind the health factors. From the model evaluation results, it can be seen that the model constructed by the joint time-frequency domain health factors can characterize the battery aging characteristics more comprehensively and stably.
[0061] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
Claims
1. A method for rapid assessment of battery health status based on joint features in time and frequency domains, comprising step 1, designing a test process; step 2, acquiring data in time and frequency domains; step 3, extracting joint features; step 4, building an assessment model; and step 5, deploying an application; Features: In the above step 1, according to the parameter information such as battery type and capacity, the details of voltage regulation experiment, static test and EIS test experiment are set, and the battery health status assessment test process is designed; In the above step 2, the battery is charged to 20% SOC through the capacity distribution cabinet to obtain voltage regulation data, that is, time domain data, and the battery is tested with a broadband current through an electrochemical AC fast impedance spectrum analysis device to obtain EIS data, that is, frequency domain data; In the above step 3, the indirect and direct features of voltage, EIS, and DRT are extracted to construct a list of battery preparation health factors, and the impurity elements in the list are removed using the correlation coefficient method, and the list is updated as the time-frequency domain joint features of the model input; In the above step 4, a machine learning regression estimator is used as a learning model, and the constructed time-frequency domain features are used as input to learn the battery aging law, and a residual value estimation model based on the joint features of the time-frequency domain is established; In the above step five, the time-frequency domain joint feature method and capacity estimation algorithm are deployed in the data analysis module of the fast impedance spectrum analyzer so as to use the learned model to quickly evaluate the battery health status.
2. According to claim 1, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In the step 1, the entire evaluation test process consists of three parts: the charging and discharging stage, the static stage, and the EIS test stage. In the charging and discharging stage, the battery is first discharged to ensure that it is at 0% SOC, and then charged to 20% SOC after static. After charging, it is necessary to stand for 1 hour to allow the internal state of the battery to stabilize. After the static state is completed, the EIS test is performed. The entire test process takes 1 hour and 24 minutes relative to a single battery.
3. According to claim 1, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In the step 2, the battery is charged to 20% SOC by the charging and discharging equipment, and the working step information is: constant current discharge 1-standing 2-constant current discharge 3-standing 4-constant current charging 5-standing 6, wherein the standing time of standing 2 is 5 minutes, the standing time of standing 4 is 1 minute, and the standing time of standing 6 is 30 seconds. The current of constant current discharge 1 is 1C, the current of constant current discharge 3 is 0.1C, the charging current is 0.6C, and the pre-charge capacity = rated capacity * 20% SOC. This part of the operation coincides with the battery state of charge required by the gradient battery when it leaves the factory, so it is not necessary to use this part as an additional part. The fast impedance spectrum equipment is set to a frequency range of 10kHz-0.1Hz and uses a set current excitation. The excitation current is set to 5A. The frequency domain data under different aging states are obtained by testing gradient batteries with different aging degrees.
4. According to claim 1, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In the step three, the training samples are grouped by the DBSCAN clustering algorithm, and the clustering results with less than 3 samples in the group are eliminated. The samples with less than the threshold in the cluster group are regarded as abnormal samples. The samples after elimination are subjected to time-frequency domain feature extraction to characterize the health factor of the battery. The voltage regulation charging and discharging step in step two is used as the time domain feature extraction method. The complex impedance information in the frequency range of 10kHz-0.1Hz is obtained by the frequency sweep method. The frequency logarithmic distribution is taken, and 10 frequency points are taken for each logarithmic scale, for a total of 51 groups. The final combined features in the time-frequency domain are 300-0.1Hz real impedance, 3.2-3.35V charging time, voltage mean, voltage variance, and IC capacity increment peak.
5. According to claim 4, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In step 3, the time-frequency domain feature selection method uses the Pearson correlation coefficient method to obtain the positive and negative correlation between each feature and the capacity, and uses the features and capacity to visualize the correlation degree, and eliminates the features with low correlation. The Pearson correlation coefficient calculation formula is: In the formula and They represent the sample means of the two respectively.
6. According to claim 5, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In the step three, the time-frequency domain features are eliminated and the range is reduced based on the prior knowledge of the battery equivalent circuit ECM and the correlation analysis results obtained by using the Pearson correlation coefficient method. The prepared health factor feature list before elimination consists of the real and imaginary impedance of the electrochemical impedance spectrum, the γ in the DRT feature calculated using the relaxation time distribution method, the charging time of the voltage regulation curve, the voltage mean, the voltage variance, and the IC incremental capacity peak. The correlation score of each health factor and the capacity is calculated according to the Pearson method. The correlation score is [-1, 1], a negative number is negatively correlated with the capacity, and a positive number is positively correlated with the capacity, so as to further screen out better features.
7. According to claim 1, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In the step 4, after constructing a feature list by using a combination of multiple features, the data is divided into a training set and a validation set with a division ratio of 7:
3. When the data is small, the leave-one-out method and K-fold cross-validation method can be used for division, and an extreme gradient regression tree XGBoost model is constructed to learn the battery aging law and output the existing capacity residual value of the battery. The verification estimation results are used through a variety of model evaluation indicators to indicate the degree of model learning. The evaluation indicators used here are: model accuracy r2, relative root mean square percentage error RRMSE, mean absolute percentage error MAPE, and maximum relative error.
8. According to claim 7, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In step 4, the calculation formula of the model accuracy determination coefficient is: The relative RMS percentage error is calculated as: The mean absolute percentage error is calculated as: The maximum relative error calculation formula is:
9. According to claim 7, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In the step 4, in the battery capacity estimation algorithm part, the xgboost regression model is trained by serially connecting multiple weak learners (i.e., regression trees). Each learner will further correct the learning based on the learning residual results of the previous learner, and gradually improve the estimation ability of the model. Each round of iteration builds a new regression tree model by optimizing the objective function. During the construction process, the greedy optimization method is used to minimize the objective function, and regularization is used to avoid overfitting.
10. According to claim 1, a method for rapid assessment of battery health status based on joint features in time and frequency domains, Features: In the step five, the online application process consists of sampling, data processing, algorithm estimation, and result visualization output. The corresponding hardware and software implementation consists of an electrochemical impedance spectroscopy excitation sampling module, a Raspberry Pi: data storage, transmission, algorithm deployment, and a visualization result host computer. The best model is selected and saved as a pkl format file through the pickle library function. The model is deployed on the edge device Raspberry Pi, the model is called by building a FastAPI interface, and the algorithm is tested online to evaluate the battery capacity. A large amount of battery information is collected, and the model is continuously retrained and optimized.
Citation Information
Cited By
On-site battery health assessment method based on multi-impedance measurement
CN119270115A
Battery state prediction method and system, computer equipment and storage medium
CN119395562A
Method and device for quickly estimating internal temperature of battery
CN119493012A
Automobile emergency starting power supply monitoring method and system based on intelligent sensing
CN120028710A
Lithium ion battery health state diagnosis method based on equivalent circuit feature screening
CN120064998A
