State evaluation and life prediction based on multi-dimensional data application in charging and discharging process of power battery

Abstract

The invention provides state evaluation and service life prediction based on multi-dimensional data application in the charging and discharging process of a power battery, and relates to the technical field of battery management. According to the state evaluation and life prediction based on the multi-dimensional data application in the charging and discharging process of the power battery, the multi-dimensional data of the battery in the charging and discharging process is collected and processed, so that the data quality is ensured; carrying out similarity analysis on the battery charging and discharging curve by utilizing the Pauta criterion, and quantitatively evaluating the health state of the battery cell; and removing abnormal data through a K-means clustering method, clustering the data, and deeply predicting the service life of the battery through at least two layers of long short-term memory (LSTM) network models. The method can effectively improve the accuracy of battery state evaluation, predict the remaining life of the battery, provide a scientific basis for maintenance, replacement and safety management of the battery, and promote cyclic utilization of battery resources.

Description

Technical Field

[0001] The present invention relates to the field of battery management technology, and specifically to a state assessment and life prediction based on the application of multi-dimensional data during the charging and discharging process of a power battery. Background Art

[0002] As the core part of electric vehicles and new energy storage equipment, the cost and technological development of batteries lag behind the response to social needs, which seriously restricts the rapid development of electric vehicles. At the same time, battery safety issues have also attracted widespread attention, especially in many spontaneous combustion or explosion accidents in new energy vehicles and energy storage projects, battery problems are often the main cause.

[0003] As a key link to ensure the safe application of lithium batteries, the battery management system is responsible for the intelligent management and maintenance of each battery unit, preventing overcharging and over-discharging, extending the battery life and monitoring the battery status. If there is a problem with the battery during the battery health status assessment, the existing battery management technology will detect the status of each battery cell, which is very inefficient and has deficiencies in predicting the remaining battery life. Although the model-based fault diagnosis method can monitor the charging and discharging process of the battery system in real time, its accuracy depends on the accuracy of the system mathematical model and may not accurately reflect the actual battery status. Although the data-driven method does not require the establishment of an accurate mathematical model, it has a large amount of calculation and requires a large amount of historical sample data. In addition, it has the problems of insufficient generalization ability and poor accuracy in predicting battery performance degradation. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides a state assessment and life prediction based on the application of multi-dimensional data during the charging and discharging process of the power battery, which solves the problems of low efficiency and poor accuracy of battery life prediction and battery health assessment.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a state assessment and life prediction based on the application of multi-dimensional data during the charging and discharging process of a power battery, comprising:

[0008] a. Collect multi-dimensional data of the power battery during the charging and discharging process; and process the collected data;

[0009] b. Record the charge and discharge curves of each battery group, each series-parallel battery module and each battery cell in turn; perform quantitative analysis on the similarity of the charge and discharge curves using the Laida criterion to evaluate the health status of the battery cells;

[0010] c. Cluster the processed data using a cluster analysis method, obtain the abnormal behavior data of the multi-dimensional data of the battery during the charging and discharging process based on the clustering, remove the data of abnormal behavior, and use the clustered data for the prediction of battery life using a long short-term memory network (LSTM);

[0011] d. Determine whether the battery should be replaced based on the predicted battery life combined with the battery cell health status and abnormal behavior data of multi-dimensional data.

[0012] Preferably, multi-dimensional data of the power battery during the charging and discharging process is collected, and the collected multi-dimensional data includes open circuit voltage, operating voltage, operating current, internal resistance, capacity and temperature; the open circuit voltage is the voltage of the battery under no-load conditions, and it is a basic parameter for evaluating the battery state of charge; the operating voltage and operating current directly reflect the electrical performance of the battery under actual charging and discharging working conditions; the internal resistance, as a measure of the internal resistance of the battery, is used to understand the charging and discharging efficiency and thermal management of the battery; the battery capacity is an indicator of its energy storage capacity, and the battery temperature directly affects its performance and safety; the collected data are cleaned to remove data generated by measurement errors or equipment failures.

[0013] Preferably, the charge and discharge curves of each battery pack, each series-parallel battery module, and each battery cell are recorded in sequence, and the performance changes of the battery during the charge and discharge process are captured through these records to provide data support for subsequent analysis; the similarity of these charge and discharge curves is calculated by the Laida criterion to evaluate the consistency of the health status between the battery cells; the similarity measure of the charge and discharge curves of each battery cell is obtained by calculation, and the similarity measure reflects the consistency of the battery cell performance; based on the similarity measure of the charge and discharge curves, the mean and standard deviation are further calculated, and the overall health status of the battery pack is understood based on the mean and standard deviation. ; If the discharge curves in a group of batteries show significant differences, there are cells with poor performance in the battery pack, so the series and parallel battery modules in the battery group are checked in depth, and the specific cell with the status problem is determined by further analysis of the discharge curve; the problem cell is efficiently identified and maintenance measures are taken; if the discharge curves in each group of batteries are similar, it indicates that the health status of each cell in the battery pack is relatively consistent, and there is no obvious performance degradation or failure phenomenon. Therefore, there is no need to make further detailed judgments through the discharge curves recorded in sequence, because the battery pack performs well as a whole, and there is no need for additional detailed analysis.

[0014] Preferably, the Laida criterion obtains a d-order polynomial f by performing polynomial fitting on d+1 point pairs of each battery cell. j(x)(j = 1, 2, …, m); Re - select p points on the fitted polynomial curve by discrete sampling, and represent the curve similarity of the voltage change rate of each battery cell by the distance and mean value of these p points; The specific process is as follows:

[0015] S1. Arbitrarily select a value in the interval [1, i - 1] and denote it as z1;

[0016] S2. Arbitrarily select a value Δz, with the requirement that

[0017] S3. Taking z1 as the reference, select the remaining p - 1 values at intervals of Δz, denoted as (z2, z3, …, zp); If Zp < i - 1 is not satisfied, return to step S2;

[0018] S4. Substitute the selected p values into each d - degree polynomial f j (x) to calculate its value;

[0019] S5. Using the calculation method of Euclidean distance, obtain the curve similarity of the voltage change rate corresponding to each battery cell:

[0020] According to the curve similarity calculated above, calculate the corresponding mean value μ and standard deviation σ for dst(1), dst(2), …, dst(m); If for each battery cell, |dist(j)-μ| > 3σ is satisfied, then the health state of this battery cell is not good, that is, the health state of the battery is not good.

[0021] Preferably, the clustering analysis method is K - means clustering. Cluster the processed data by battery health state and battery capacity, obtain the performance patterns and trends of the battery according to the clustering, and identify abnormal behaviors in the battery data through clustering; K - means clustering provides structural data processing for the long - short - term memory model, and divides the processed and clustered data into datasets for long - short - term memory network model training and prediction; The K - means clustering process is as follows:

[0022] Step 1. Divide the battery health state and battery capacity data into K groups according to the corresponding charge - discharge process and the number of samplings, and randomly select K objects as the initial clustering centers;

[0023] Step 2. Calculate the distance between each object and each clustering center, and assign each object to the clustering center closest to it; The clustering centers and the objects assigned to them represent a cluster. For each assigned sample, the clustering center of the cluster is recalculated according to the existing objects in the cluster;

[0024] Step 3. Repeat step 2 until the object assignment no longer changes.

[0025] Preferably, the long short-term memory network (LSTM) model adopts at least two layers of recurrent neural network structure because the accuracy of one layer of LSTM in medium- and long-term prediction results is greatly reduced. The multi-level network structure can more deeply capture the performance of the battery and the attenuation of the battery life. The LSTM model includes an input gate, a forget gate and an output gate. The coordinated work of these gates enables the network to remember long-term dependencies. In battery performance analysis, LSTM identifies and memorizes data during the battery charge and discharge cycle, while ignoring noise and short-term fluctuations that have little effect on the battery life.

[0026] Preferably, by stacking at least two layers of LSTM structures, the LSTM model can not only learn the short-term patterns of the battery charge and discharge curve, such as the performance changes in a single charge and discharge event, but also can further abstract longer-term and complex performance trends on this basis; the first layer of LSTM structure is responsible for capturing the local features and short-term dependencies in the battery charge and discharge curve, while the second layer of structure integrates the local features in the battery charge and discharge curve to extract higher-level features, which are related to the long-term health status and life decay of the battery; the use of two-layer LSTM structure enhances the LSTM model's processing ability for time series data, enabling it to better understand and predict the evolution of battery performance; the hierarchical feature extraction and learning mechanism provides a richer and more abstract data representation for battery life prediction, thereby improving the accuracy and reliability of the prediction; the output of the LSTM model is converted through one or more fully connected layers to map the extracted high-level features into specific prediction values ​​for the remaining life of the battery, so that the output of the LSTM model not only conforms to the actual change law of battery performance, but also meets the specific needs of battery life prediction in practical applications.

[0027] (III) Beneficial effects

[0028] The present invention provides a state assessment and life prediction based on the application of multi-dimensional data during the charging and discharging process of a power battery, which has the following beneficial effects:

[0029] 1. The present invention uses the Laida criterion to perform similarity analysis on battery charge and discharge curves, and can quantitatively evaluate the health status of the battery cell. Accurate evaluation of the health status of the battery cell helps prevent battery failures, reduce safety risks, avoid spontaneous combustion or explosion accidents of new energy vehicles caused by battery problems, and enhance the safety of battery use.

[0030] 2. The battery remaining life prediction method provided by the present invention can provide a scientific basis for battery maintenance and replacement. Accurate battery life prediction helps to determine the retirement time of the battery, reduce unnecessary battery replacement, improve the economy of battery use, and promote the recycling of resources.

[0031] 3. The present invention can effectively identify batteries with inconsistent performance in a battery pack, keep the health status of each battery cell in the battery pack relatively consistent, and avoid the failure of a single battery cell affecting the performance of the entire battery pack. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the process of the present invention;

[0033] Figure 2 It is a schematic diagram of the battery pack structure distribution of the present invention. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present invention will be described clearly and completely below. 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.

[0035] Collect multi-dimensional data of the power battery during the charging and discharging process, including open circuit voltage, operating voltage, operating current, internal resistance, capacity and temperature; open circuit voltage is the voltage of the battery under no-load conditions, which is a basic parameter for evaluating the battery state of charge; operating voltage and operating current directly reflect the electrical performance of the battery under actual charging and discharging conditions; internal resistance, as a measure of the internal resistance of the battery, is used to understand the battery's charging and discharging efficiency and thermal management; battery capacity is an indicator of its energy storage capacity, while the battery's temperature directly affects its performance and safety; these collected data are cleaned to remove data generated by measurement errors or equipment failures.

[0036] In the state assessment and life prediction based on the multi-dimensional data application in the process of power battery charging and discharging, the charging and discharging curves of each battery pack, each series-parallel battery module and each battery cell of the power battery are recorded in sequence. By recording the charging and discharging curves, the performance changes of the battery in the charging and discharging process are captured to provide data support for subsequent analysis. If the power battery is composed of 6 battery packs, such as Figure 2As shown, each battery pack contains 8 series - parallel battery modules, and each series - parallel battery module contains 72 cells; the similarity of these charge - discharge curves is calculated by the Pauta criterion to evaluate the consistency of the health state among the cells; the similarity degree of the charge - discharge curves of each cell is calculated, and the consistency of the cell performance is judged according to the similarity degree; based on the similarity degree of the charge - discharge curves, the mean value and the standard deviation are further calculated, and the overall health state of the battery pack is understood according to the mean value and the standard deviation; if the discharge curves of one group of batteries show significant differences, there are underperforming cells in the battery pack, so the 8 series - parallel battery modules in this group of batteries are inspected in depth, and the specific cell with a state problem is determined by further analyzing the discharge curves; the problem cells are efficiently identified, and maintenance or replacement of the cells is carried out; if the discharge curves of each group of batteries are similar, it indicates that the health states of the cells in the battery pack are relatively consistent, without obvious performance degradation or failure phenomena, so there is no need to make further detailed judgments through the sequentially recorded discharge curves, because the battery pack as a whole performs well and there is no need for additional detailed analysis.

[0037] The Pauta criterion obtains a d - degree polynomial f j (x) (j = 1, 2, …, m) by performing polynomial fitting on d + 1 pairs of points for each cell; p points are re - selected on the fitted polynomial curve by means of discrete sampling, and the curve similarity of the voltage change rate of each cell is represented by the distance and the mean value of these p points; the specific process is as follows:

[0038] S1. Select a value denoted as z1 arbitrarily in the interval [1, i - 1];

[0039] S2. Select a value Δz arbitrarily, requiring

[0040] S3. Based on z1, select the remaining p - 1 values at intervals of Δz, denoted as (z2, z3, …, z p ); if z p < i - 1 is not satisfied, return to step S2;

[0041] S4. Substitute the selected p values into each d - degree polynomial f j (x) to calculate its value;

[0042] S5. Using the calculation method of Euclidean distance, obtain the curve similarity of the voltage change rate corresponding to each cell:

[0043] Based on the similarity of the curves obtained by the above calculation, the corresponding mean μ and standard deviation σ are calculated for dst(1), dst(2), …, dst(m); if |dist(j)-μ|>3σ for each battery cell, the health status of the battery cell is not good, that is, the health status of the battery is not good.

[0044] The clustering analysis method is K-means clustering. The processed data is clustered by battery health status and battery capacity. The battery performance mode and trend are obtained based on clustering, and abnormal behavior in battery data is identified through clustering. K-means clustering provides structural data processing for the long short-term memory model, and divides the processed clustered data into data sets for long short-term memory network model training and prediction. The K-means clustering process is as follows:

[0045] Step 1. Divide the battery health status and battery capacity data into K groups according to the corresponding charging and discharging process and the number of sampling. If K=5, randomly select 5 objects as the initial clustering centers;

[0046] Step 2. Calculate the distance between each object and each cluster center, and assign each object to the cluster center closest to it; the cluster centers and the objects assigned to them represent a cluster. Each time a sample is assigned, the cluster center of the cluster is recalculated based on the existing objects in the cluster.

[0047] Step 3. Repeat step 2 until the object allocation does not change.

[0048] The long short-term memory network (LSTM) model uses at least two layers of recurrent neural network structure because the accuracy of a single-layer LSTM in medium- and long-term prediction results drops dramatically. The multi-level network structure can more deeply capture the battery performance and battery life attenuation. The LSTM model includes an input gate, a forget gate, and an output gate. The coordinated work of these gates enables the network to remember long-term dependencies. In battery performance analysis, LSTM identifies and memorizes data during the battery charge and discharge cycle, while ignoring noise and short-term fluctuations that have little effect on battery life.

[0049] By stacking at least two layers of LSTM structures, the LSTM model can not only learn the short-term patterns of the battery charge and discharge curve, such as the performance changes in a single charge and discharge event, but also further abstract longer-term and complex performance trends on this basis; the first layer of LSTM structure is responsible for capturing the local features and short-term dependencies in the battery charge and discharge curve, while the second layer integrates the local features in the battery charge and discharge curve to extract higher-level features, which are related to the long-term health status and life decay of the battery; the use of two-layer LSTM structure enhances the LSTM model's ability to process time series data, enabling it to better understand and predict the evolution of battery performance; the hierarchical feature extraction and learning mechanism provides a richer and more abstract data representation for battery life prediction, thereby improving the accuracy and reliability of the prediction; the output of the LSTM model is converted through one or more fully connected layers to map the extracted high-level features into specific prediction values ​​for the remaining battery life, so that the output of the LSTM model not only conforms to the actual change law of battery performance, but also meets the specific needs of battery life prediction in practical applications; the battery health status and battery life prediction and the abnormal behavior data obtained in K-means clustering are used to determine whether the battery needs maintenance or replacement.

[0050] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A state assessment and life prediction based on the application of multi-dimensional data during the charging and discharging process of a power battery, characterized in that; The following steps are involved: a. Collect multi-dimensional data of the power battery during the charging and discharging process; and process the collected data; b. Record the charge and discharge curves of each battery group, each series-parallel battery module and each battery cell in turn; perform quantitative analysis on the similarity of the charge and discharge curves using the Laida criterion to evaluate the health status of the battery cells; c. Cluster the processed data using a cluster analysis method, obtain the abnormal behavior data of the multi-dimensional data of the battery during the charging and discharging process based on the clustering, remove the data of abnormal behavior, and use the clustered data for the prediction of battery life using a long short-term memory network (LSTM); d. Determine whether the battery should be replaced based on the prediction of battery life combined with the abnormal behavior data of the battery cell and multi-dimensional data.

2. The state assessment and life prediction based on multi-dimensional data application during the charging and discharging process of a power battery according to claim 1, characterized in that: Collect multi-dimensional data of power batteries during the charging and discharging process, and perform data cleaning and outlier removal; The multi-dimensional data includes open circuit voltage, operating voltage, operating current, internal resistance, capacity, and temperature.

3. The state assessment and life prediction based on multi-dimensional data application during the charging and discharging process of a power battery according to claim 1, characterized in that: Record the charge and discharge curves of each battery pack, the charge and discharge curves of each series-parallel battery module, and the charge and discharge curves of each battery cell in turn; perform similarity calculation on the recorded charge and discharge curves using the Laida criterion, calculate the corresponding mean and standard deviation based on the obtained similarity, and judge the battery health status based on the similarity, mean, and standard deviation; if there are differences in the discharge curves in a group of batteries, check the series-parallel battery modules in the group of batteries to determine which battery cell has a status problem; if the discharge curves in each group of batteries are similar, there is no need to deeply test each battery cell.

4. The state assessment and life prediction based on multi-dimensional data application during the charging and discharging process of a power battery according to claim 3, characterized in that: The Laida criterion obtains a d-order polynomial f by performing polynomial fitting on d+1 point pairs of each cell. j (x)(j=1,2,…,m); reselect p points on the fitted polynomial curve by discrete sampling, and use the distance and mean of these p points to represent the curve similarity of the voltage change rate of each battery cell; the specific process is: S1. Select any value in the interval [1,i-1] and record it as z1; S2. Choose a value Δz, S3. Based on z1, select the remaining p - 1 values at intervals of Δz, denoted as (z2, z3, …, z p ); If z p < i - 1 is not satisfied, return to step S2; S4. Substitute the selected p values ​​into each d-order polynomial f in turn j (x) calculate its value; S5. Using the Euclidean distance calculation method, the curve similarity of the voltage change rate corresponding to each battery cell is obtained: Based on the similarity of the curves obtained by the above calculation, the corresponding mean μ and standard deviation σ are calculated for dst(1), dst(2), …, dst(m); if |dist(j)-μ|>3σ for each battery cell, the health status of the battery cell is not good, that is, the health status of the battery is not good.

5. The state assessment and life prediction based on multi-dimensional data application during the charging and discharging process of a power battery according to claim 1, characterized in that: The cluster analysis method is K-means clustering, which clusters the processed data by battery health status and battery capacity, obtains battery performance patterns and trends based on clustering, and identifies abnormal behaviors in battery data through clustering; K-means clustering provides structured data processing for the LSTM model, dividing the processed clustered data into data sets for LSTM network model training and prediction.

6. The state assessment and life prediction based on multi-dimensional data application during the charging and discharging process of a power battery according to claim 5, characterized in that: The long short-term memory network (LSTM) model is improved on the basis of a one-layer network structure, and adopts at least two layers of recurrent neural network structure; the multi-layer network structure captures the battery performance and battery life attenuation more deeply.

7. The state assessment and life prediction based on multi-dimensional data application during the charging and discharging process of a power battery according to claim 6, characterized in that: The two-layer recurrent neural network structure can extract higher-level features from the previous layer according to each layer of LSTM structure. The first layer can learn the short-term pattern of the battery charging and discharging curve, and the second layer further abstracts the longer-term and complex battery performance and battery life prediction on this basis; the output of the two-layer LSTM model converts the feature vector of the LSTM model into a predicted value of the remaining battery life through one or more fully connected layers, so that the LSTM model outputs a prediction result that meets the actual application needs.

Images