Lithium ion battery health state estimation method based on MSCNN

Abstract

The invention relates to a lithium ion battery health state estimation method based on an MSCNN, and belongs to the technical field of lithium ion batteries. The method comprises the following steps: performing an aging cycle test on a lithium ion battery monomer, and establishing a working condition test experiment database; collecting voltage, current and temperature data of the battery in each aging cycle, respectively calculating average values, standard deviations, skewness and kurtosis of the voltage, the current and the temperature, taking the average values, the standard deviations, the skewness and the kurtosis as health factors, and screening out the health factors which are highly related to the health state of the battery; constructing an SOH estimation model, and training the model by taking the screened health factors as input; and the model is used for predicting the health state of the lithium ion battery after being trained. The method can effectively extract features from data of different scales, has higher SOH estimation precision compared with a traditional CNN, can perform accurate SOH estimation on data sets of different working conditions, and has good generalization ability.

Description

Technical Field

[0001] The invention belongs to the technical field of lithium ion batteries and relates to a lithium ion battery health state estimation method based on MSCNN. Background Art

[0002] The new power system is an important part of the modern energy system, featuring clean, low-carbon, safe, efficient and rapid deployment of clean energy. Electrochemical energy storage has become an important part of the new power system and one of the fastest-growing energy storage technologies due to its fast response time, high energy density and flexible structure. In this field, lithium-ion batteries are particularly important because of their high energy density, low self-discharge rate, high cycle efficiency, long service life and flexible configuration, making them indispensable in electrochemical energy storage technology.

[0003] Lithium-ion batteries have been widely used in the fields of electrified transportation and energy storage facilities, serving as the main energy storage devices for electric vehicles and energy storage facilities, respectively. Due to the irreversible electrochemical properties of lithium-ion batteries, continuous charge and discharge cycles will cause their output performance, operating conditions, mechanical stress, and chemical reactions to gradually degrade. In order to ensure the long-term safe and stable operation of the battery management system, as well as to perform regular maintenance of important lithium-ion battery energy storage systems, the state of health (SOH) is an important indicator for measuring system safety and stability. It is defined as the ratio of the current maximum capacity to the initial capacity and is a key indicator for assessing battery degradation. When the SOH drops to a certain level, it indicates that the battery no longer meets the requirements for running new electric vehicles, but it can be reused in areas such as lithium-ion battery energy storage stations that have lower health requirements.

[0004] Normally, the battery SOH cannot be measured directly, and the capacity needs to be obtained by integrating the complete discharge curve from the high cut-off voltage to the low cut-off voltage through the ampere-hour integration. However, in practical applications, since the battery is rarely fully discharged, it is difficult to obtain a complete charge and discharge curve, and this method is only applicable to laboratory conditions. As batteries continue to age, safety issues are on the rise. Conducting research on battery health status assessment and predictive operation and maintenance is of great significance to the safe and stable operation of energy storage systems. In recent years, researchers have conducted fruitful research on the SOH of lithium-ion batteries and proposed a variety of methods. Lithium-ion battery health status estimation methods can generally be divided into two categories: model-based methods and data-driven methods.

[0005] The model-based method uses a formula to represent the battery health state and then extrapolates the battery aging trajectory. The model-based method mainly includes electrochemical models, equivalent circuit models and empirical models. The electrochemical model is based on the first principle model, which can describe the electrochemical reaction in the lithium-ion battery. The model parameters have a clear physical meaning, and the internal state of the battery has a clear coupling relationship with the external characteristics. With the rapid development of artificial intelligence technology, data-driven methods have received extensive attention from scholars in the field of lithium-ion battery SOH. Unlike model-based methods, data-driven methods extract features from data sources such as voltage, current, temperature and capacity, and explore the coupling relationship between battery performance attenuation and key features. This method does not require an in-depth understanding of the complex mechanisms and chemical changes during battery operation, nor does it rely on prior knowledge of electrochemistry. By constructing a nonlinear attenuation model of the battery health state and using various algorithms to analyze and process the data, accurate prediction of lithium battery SOH is achieved, and it has wide applicability.

[0006] However, existing data-driven methods have some shortcomings. For example, some methods rely too much on a single data source or feature, ignoring the multi-scale information of the battery aging process, resulting in limited estimation accuracy. In addition, some methods may exhibit poor generalization capabilities when faced with data sets under different working conditions, making it difficult to adapt to complex and changeable scenarios in practical applications. Summary of the invention

[0007] In view of this, the purpose of the present invention is to provide a lithium-ion battery health state estimation method based on MSCNN, fully explore the lithium-ion battery aging information implicit in the charging process, use multiple key health factors to fusion prediction, and construct an MSCNN network model suitable for lithium-ion battery health state prediction tasks, so as to obtain higher lithium-ion battery health state estimation accuracy and generalization ability.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] A method for estimating the health status of a lithium-ion battery based on MSCNN, the method comprising:

[0010] S1. Perform aging cycle tests on lithium-ion battery cells under certain environmental conditions and establish a working condition test experimental database;

[0011] S2, dividing each aging cycle process into stages, and collecting the voltage, current and temperature data of the battery in each stage, sampling the collected data, and calculating the average value, standard deviation, skewness and kurtosis of the voltage, current and temperature based on the sampled data;

[0012] S3. The average value, standard deviation, skewness and kurtosis of the voltage, current and temperature calculated during each aging cycle test are used as health factors, and the health factors are correlated with the battery capacity to screen out the health factors with strong correlation with the battery health status;

[0013] S4, constructing a SOH estimation model based on MSCNN, and taking the screened health factors as input to train the model;

[0014] S5. Perform the aging cycle test on any lithium-ion battery, calculate the health factor (at least one of the health factors screened out in step S4) during the cycle, and input the obtained health factor into the SOH estimation model to predict the health status of the lithium-ion battery.

[0015] Furthermore, in step S2, dividing each aging cycle process into stages includes dividing each aging cycle process into a constant current charging stage, a constant voltage charging stage and a constant current discharging stage which are connected in sequence.

[0016] The voltage, current and temperature data of the battery in the constant current charging stage, the constant voltage charging stage and the constant current discharging stage are collected respectively; the collected data are sampled; based on the sampled data, the average values ​​of the voltage, current and temperature, the standard deviations of the voltage, current and temperature, the skewness of the voltage, current and temperature, and the kurtosis of the voltage, current and temperature in the constant current charging stage, the constant voltage charging stage and the constant current discharging stage are calculated respectively.

[0017] Furthermore, in step S3, the health factor also includes the charging time in the constant current stage, the charging time in the constant voltage stage, and the sum of the charging time in the constant current stage and the charging time in the constant voltage stage.

[0018] Furthermore, in step S3, the correlation analysis between the health factor and the battery capacity is performed using the Pearson correlation coefficient and the Spearman correlation coefficient. The correlation coefficients of each health factor and the battery capacity are first calculated respectively, and then the health factors whose Pearson correlation coefficient and the Spearman correlation coefficient are both greater than 0.9 are screened out.

[0019] Furthermore, in step S4, the SOH estimation model includes an input layer, a first convolutional layer to a sixth convolutional layer, a discard layer, a fully connected layer, a regression layer and an output layer which are connected in sequence.

[0020] It also includes a first maximum pooling layer between the first and second convolutional layers, a second maximum pooling layer between the second and third convolutional layers, a third maximum pooling layer between the fourth and fifth convolutional layers, and a fourth maximum pooling layer between the sixth convolutional layer and the dropout layer.

[0021] Furthermore, in step S4, during the training process of the SOH estimation model, the mean square error is used as the loss function, and the loss function is optimized by the Adam optimizer; after the training is completed, the SOH estimation model is tested, and the prediction accuracy of the model is evaluated by RMSE, MAE and MAPE.

[0022] The beneficial effects of the present invention are:

[0023] (1) The SOH estimation model based on MSCNN constructed in the present invention can effectively extract features from data of different scales and has higher SOH estimation accuracy than traditional CNN.

[0024] (2) The SOH estimation model proposed in the present invention can also perform accurate SOH estimation for data sets under different working conditions and has good generalization ability.

[0025] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0027] Figure 1 A schematic diagram of a flow chart of a method for estimating the health status of a lithium-ion battery provided by an embodiment of the present invention;

[0028] Figure 2 A schematic diagram of a lithium-ion battery charging and discharging strategy proposed in one embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram of data isochronous sampling;

[0030] Figure 4 A schematic diagram of the MSCNN structure proposed in one embodiment of the present invention;

[0031] Figure 5 to Figure 7 Schematic diagram of data distribution of training set and test set in different embodiments of the present invention. DETAILED DESCRIPTION

[0032] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0033] An embodiment of the present invention provides a method for estimating the health status of a lithium-ion battery based on MSCNN, such as Figure 1 As shown, the method specifically includes:

[0034] 1. Carry out aging cycle test on the lithium-ion battery cells to be tested under certain environmental conditions, and establish a working condition test experiment database (including current, voltage, temperature, and capacity data).

[0035] Among them, the aging cycle test of lithium-ion batteries includes the nominal capacity stage and the aging cycle stage.

[0036] (1) The nominal capacity stage is as follows: first, the initial capacity of the battery is measured, then it is charged at a constant current of 1A to a cut-off voltage of 4.2V, and then it is transitioned to constant voltage charging until the charging current drops to 0.04A; after a 5-minute rest, it begins to discharge at a constant current of 0.4A until it reaches a lower cut-off voltage of 2.5V.

[0037] (2) The aging cycle stage is as follows: charging starts at a constant current of 4 A until the upper cut-off voltage of 4.2 V is reached, and then charging continues at a fixed voltage until the current drops to 0.1 A; after a 5-minute rest, discharging starts at a constant current of 2 A until the lower cut-off voltage of 2.5 V is reached, and then the battery is idle for another 5 minutes to complete an aging cycle.

[0038] The above aging cycle stage is repeated until the battery capacity decreases to less than 80% of the initial capacity.

[0039] In this embodiment, the selected lithium-ion batteries are manufactured by Lishen (Tianjin, China), with two batches of batteries. The first batch of 8 batteries is marked as A1~A8, and the second batch of 8 batteries is marked as B1~B8. The nominal capacity of the lithium-ion batteries is 2000mAh, the nominal voltage is 3.6V, the charging cut-off voltage is 4.2V, and the discharging cut-off voltage is 2.5V.

[0040] For the batteries in group A, data from the aging cycle test was collected through the operations described in (1) and (2) above. For the batteries in group B, during the aging cycle stage, the constant current charging current was 6A, and the other operations remained unchanged.

[0041] 2. During the aging cycle stage of each cycle, calculate the average values ​​of voltage, current and temperature during constant current charging, constant voltage charging and constant current discharging, as well as the standard deviation, skewness and kurtosis of voltage, current and temperature.

[0042] (1) During the constant current charging process, the voltage, current and temperature data are divided into 128 sampling points at equal time intervals. Let x1(i), y1(i), z1(i) be the values ​​of voltage, current and temperature at each sampling point, respectively, i = 1, 2, ..., 128. Then the average voltage value during the constant current charging process is:

[0043]

[0044] The average current is:

[0045]

[0046] The average temperature is:

[0047]

[0048] Where n = 128. Similarly, during constant voltage charging, the average values ​​of voltage, current and temperature are:

[0049]

[0050] Among them, x2(i), y2(i), z2(i) are the values ​​of voltage, current and temperature at each sampling point in the constant voltage charging process. Similarly, the average values ​​of voltage, current and temperature in the constant current discharge process are:

[0051]

[0052] Among them, x3(i), y3(i), and z3(i) are the values ​​of voltage, current, and temperature at each sampling point in the constant voltage charging process data collection.

[0053] (2) Calculate the standard deviation of voltage, current and temperature during constant current charging, constant voltage charging and constant current discharging. Here we only take the constant current charging process as an example, and will not repeat the constant voltage charging and constant current discharging processes.

[0054] According to the data collected during the constant current charging process, the standard deviations of voltage, current and temperature can be calculated as follows:

[0055]

[0056]

[0057]

[0058] Among them, σ1, σ′1, and σ″1 are the standard deviations of voltage, current, and temperature during constant current charging, respectively.

[0059] (3) Calculate the deviation of voltage, current and temperature during constant current charging, constant voltage charging and constant current discharging. Here, only the constant current charging process is used as an example, and the constant voltage charging and constant current discharging processes are not repeated.

[0060] According to the data collected during the constant current charging process, the deviations of voltage, current and temperature can be calculated as follows:

[0061]

[0062]

[0063]

[0064] Among them, skewness1, skewness′1, and skewness″1 are the skewness of voltage, current, and temperature during constant current charging, respectively.

[0065] (4) Calculate the peaks of voltage, current and temperature during constant current charging, constant voltage charging and constant current discharging. Similarly, only the constant current charging process is used as an example here, and the constant voltage charging and constant current discharging processes are not repeated.

[0066] According to the data collected during the constant current charging process, the peaks of voltage, current and temperature can be calculated as follows:

[0067]

[0068]

[0069]

[0070] Among them, kurtosis1, kurtosis′1, and kurtosis″1 are the peaks of voltage, current, and temperature during constant current charging, respectively.

[0071] 3. Calculate the Pearson correlation coefficient and Spearman correlation coefficient between the obtained health factor and the health status of the battery, and select the health factor with a strong correlation with the health status of the battery.

[0072] The health factor includes the average values ​​of voltage, current, and temperature during the aging cycle calculated in Part 2. There are 36 types of standard deviation (σ1~σ3, σ′1~σ′3, σ″1~σ″3), skewness (skewness1~skewness3, skewness′1~skewness′3, skewness″1~skewness″3) and kurtosis (kurtosis1~kurtosis3, kurtosis1′~kurtosis3′, kurtosis1″~kurtosis3″); in addition, the health factors also include constant current charging time, constant voltage charging time and the complete charging time of the battery (that is, constant current charging time plus constant voltage charging time).

[0073] The above 39 health factors are correlated with the battery capacity to evaluate whether the correlation between the health factors and the capacity meets the requirements. The Pearson and Spearman correlation coefficients can be used for evaluation. The Pearson correlation coefficient is calculated by the following formula:

[0074]

[0075] In the formula, X i is the value of the health factor selected in each cycle, is the average value of the selected health factors; Y i is the capacity of the battery at each cycle, and Y is the average capacity.

[0076] The Spearman correlation coefficient was calculated by the following formula:

[0077]

[0078] After calculating the Pearson and Spearman correlation coefficients of each health factor respectively, the health factor with both the Pearson and Spearman correlation coefficients greater than 0.9 is taken as the health factor that is correlated with the battery capacity.

[0079] 4. Construct a SOH estimation model for estimating the health status of lithium batteries, determine its model parameters, and train it to optimize the model parameters.

[0080] The SOH estimation model is built based on MSCNN, which extracts the multi-scale features of the health factors of lithium-ion batteries by using multiple convolution kernels of different scales, integrates richer local information, and fully mines the information hidden in the health factors.

[0081] In this embodiment, the constructed SOH estimation model is as follows: Figure 4As shown in the figure, in this model, two consecutive convolution layers 1 and 2 are first used. Convolution layer 1 is used to extract features. Convolution layer 1 is followed by max pooling layer 1 for downsampling to reduce the dimension of input features while retaining important feature information. The convolution kernel size of convolution layer 1 is 7*1, and the convolution kernel size of pooling layer 1 is 2*1. Convolution layer 2 processes the pooled features, further extracts and aggregates multi-scale information while reducing the data dimension. This processing method enables the network to capture features of different scales. The convolution kernel size of convolution layer 2 is 5*1, and the convolution kernel size of pooling layer 2 is 2*1. Convolution layer 2 is followed by max pooling layer 2.

[0082] After the max pooling layer 2, two consecutive convolutional layers 3 and 4 are used to enhance the feature extraction capability, first extracting low-level features and then learning more complex features. This hierarchical feature extraction improves the model's robustness to spatial changes, while reducing the risk of overfitting and improving computational efficiency through parameter sharing. The convolution kernel size of convolutional layer 3 is 4*1, the convolution kernel size of convolutional layer 4 is 4*1, and convolutional layer 4 is followed by the max pooling layer 3, whose convolution kernel size is 2*1.

[0083] The stacking of two consecutive convolutional layers after the max pooling layer 3 enables the model to extract more complex features and patterns, helping the model capture high-level features in the input data. The convolution kernel size of convolutional layer 5 is 4*1, and the convolution kernel size of convolutional layer 6 is 4*1. Convolutional layer 6 is followed by the max pooling layer 4, whose convolution kernel size is 2*1. This pooling layer is used to further reduce the feature dimension and integrate the input feature information into global information to facilitate subsequent regression processing.

[0084] By extracting and combining features layer by layer, the MSCNN network allows the model to learn richer and more effective feature representations, thereby better understanding the intrinsic relationships of the data and better integrating these features into the final prediction.

[0085] The health factors with Pearson and Spearman correlation coefficients greater than 0.9 were normalized and used as the input of the SOH estimation model. The health status was used as the model output to train the model. During the training process, the mean square error loss function was used as the evaluation index of the training. The loss function was optimized by the Adam algorithm, and the SOH estimation model was finally obtained.

[0086] 5. After the training is completed, the SOH estimation model is tested. After the test passes, the sampling data of voltage, current and temperature of the lithium-ion battery in the aging cycle stage is collected and the health factor is calculated. The health factor is input into the SOH estimation model to estimate the current health status of the lithium-ion battery.

[0087] Among them, during the SOH estimation model testing process, the prediction accuracy of the model is evaluated by the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Among them:

[0088]

[0089]

[0090]

[0091] In the formula, y i is the actual capacity of the battery, is the predicted capacity of the battery, and ω is the number of cycles.

[0092] The present invention verifies the estimation accuracy of the SOH estimation model for the health status of the battery under the same working condition through another embodiment. In this embodiment, the aging data of the lithium-ion battery selected is from the public battery data set of Xi'an Jiaotong University. The division of the training set and the test set in this embodiment is as follows Figure 5 This embodiment uses group A batteries to verify the prediction accuracy of the SOH estimation model. The three indicators of the prediction results are shown in Table 1 below:

[0093] Table 1

[0094]

[0095] As can be seen from Table 1, the MAE, MAPE and RMSE of the SOH estimation model proposed in the present invention for the SOH prediction of group A batteries are all lower than 0.75%, 0.42% and 1.04%, respectively, which verifies the accuracy of the SOH estimation model proposed in the present invention.

[0096] The present invention verifies the estimation accuracy of the battery health state of the SOH estimation model for the same working condition when the training set is not sufficient through another embodiment. In this embodiment, the training set and the test set are divided as follows: Figure 6 As shown. In this embodiment, the No. 4 battery data in group A is selected as the test set, and the number of batteries in the training set is increased successively. The following Table 2 shows the estimation accuracy of the battery health status by the SOH estimation model when the training set size is different:

[0097] Table 2

[0098]

[0099] It can be seen from Table 2 that when the data of three batteries are used as the training set, the SOH estimation model can obtain a more accurate battery SOH prediction value.

[0100] The present invention verifies the estimation accuracy of the battery health state of the SOH estimation model when the training set includes battery data of different working conditions through another embodiment. In this embodiment, the training set and the test set are divided as follows: Figure 7 As shown. This embodiment uses the data of batteries No. 1 to No. 3 in the batteries of group A and group B as the training set, and the data of batteries No. 4 to No. 8 in the batteries of group A as the test set. Table 3 shows the SOH prediction accuracy of batteries No. 4 to No. 8 in the batteries of group A under different working conditions, wherein the working condition refers to the charging rate during charging. Table 1 shows the prediction results of the SOH estimation model for the test set batteries at the same charging rate for the training set batteries, while Table 3 shows the prediction results of the SOH estimation model for the test set batteries at different charging rates for the training set batteries.

[0101] Table 3

[0102]

[0103] It can be seen from Table 3 that the prediction accuracy of the SOH estimation model proposed in the present invention under cross-operating conditions is lower than that under the same operating conditions.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A lithium-ion battery health status estimation method based on MSCNN, characterized in that: The method comprises: performing an aging cycle test on a lithium-ion battery cell under certain environmental conditions and establishing a working condition test experimental database; Each aging cycle process is divided into stages, and the voltage, current and temperature data of the battery in each stage are collected, the collected data are sampled, and the mean, standard deviation, skewness and kurtosis of the voltage, current and temperature are calculated based on the sampled data; The average value, standard deviation, skewness and kurtosis of voltage, current and temperature calculated during each aging cycle test are used as health factors. The health factors are correlated with the battery capacity to screen out health factors with strong correlation with the battery health status. Construct a SOH estimation model based on MSCNN and use the screened health factors as input to train the model; The aging cycle test is performed on any lithium-ion battery, and at least one of the screened health factors is obtained during the cycle. The obtained health factor is input into the SOH estimation model to realize the prediction of the health status of the lithium-ion battery.

2. The method according to claim 1, characterized in that The step of dividing each aging cycle into stages includes dividing each aging cycle into a constant current charging stage, a constant voltage charging stage and a constant current discharging stage which are connected in sequence.

3. The method according to claim 1 or 2, characterized in that: During each aging cycle, the voltage, current and temperature data of the battery are collected in the constant current charging stage, constant voltage charging stage and constant current discharging stage respectively; Sampling the collected data; Based on the sampling data, the average values ​​of voltage, current and temperature, the standard deviations of voltage, current and temperature, the skewness of voltage, current and temperature, and the kurtosis of voltage, current and temperature are calculated in the constant current charging stage, the constant voltage charging stage and the constant current discharging stage respectively; The mean, standard deviation, skewness and kurtosis during each aging cycle were used as health factors.

4. The method according to claim 3, characterized in that The health factor also includes the charging time in the constant current stage, the charging time in the constant voltage stage, and the sum of the charging time in the constant current stage and the charging time in the constant voltage stage.

5. The method according to claim 4, characterized in that The correlation analysis between health factors and battery capacity was performed using the Pearson correlation coefficient and the Spearman correlation coefficient. The correlation coefficients of each health factor and battery capacity were calculated respectively, and then the health factors whose Pearson correlation coefficient and the Spearman correlation coefficient were both greater than 0.9 were screened out.

6. The method according to claim 1, characterized in that The SOH estimation model includes an input layer, a first convolutional layer to a sixth convolutional layer, a discard layer, a fully connected layer, a regression layer and an output layer connected in sequence; It also includes a first maximum pooling layer between the first and second convolutional layers, a second maximum pooling layer between the second and third convolutional layers, a third maximum pooling layer between the fourth and fifth convolutional layers, and a fourth maximum pooling layer between the sixth convolutional layer and the dropout layer.

7. The method according to any one of claims 1, 5 or 6, characterized in that: During the training process of the SOH estimation model, the mean square error is used as the loss function, and the loss function is optimized by the Adam optimizer; After training, the SOH estimation model was tested, and the prediction accuracy of the model was evaluated using RMSE, MAE, and MAPE.

Images