A method for predicting capacity attenuation of a new energy vehicle power battery

Abstract

The application discloses a kind of in new energy automobile power battery capacity attenuation prediction method, belong to new energy automobile battery management field.Can effectively compensate local capacity prediction error.Because battery capacity is influenced by calendar time and temperature, capacity attenuation curve exists local rebound or sudden drop.Most capacity prediction models based on neural network can effectively capture the overall capacity attenuation trend of battery, but cannot capture its local change.Capacity prediction residual compensation model proposed in the application can effectively compensate the local capacity prediction error caused by calendar time and temperature difference.

Description

Technical Field

[0001] This invention belongs to the field of new energy vehicle battery management, and specifically relates to a method for predicting the capacity degradation of power batteries in new energy vehicles. Background Technology

[0002] To reduce dependence on fossil fuels and achieve carbon neutrality in transportation systems, lithium-ion batteries are widely used in new energy vehicles. Due to the limited availability of lithium resources, ensuring the efficient use and recycling of lithium-ion batteries is crucial for the sustainable development of new energy vehicles. Furthermore, as a critical and expensive component of new energy vehicles, timely maintenance, residual value assessment, and secondary utilization of lithium-ion battery systems are of paramount importance. Among these, battery capacity degradation prediction, a key and common technology, remains unresolved.

[0003] Battery capacity degradation prediction is mainly divided into model-based and data-driven methods. Model-based methods require establishing a mathematical model to describe the battery capacity degradation process, with typical models such as electrochemical aging models. However, due to the coupling of various side reactions within the battery, the complex degradation process is difficult to calculate and quantify. Furthermore, as the battery ages, changes in the parameters of the electrochemical aging model can cause significant errors in capacity prediction. In recent years, with the generation and accumulation of large amounts of battery data and the penetration of artificial intelligence technology, data-driven methods have been widely used in battery capacity prediction. This method can establish a nonlinear model to map the relationship between battery characteristics and degradation state (battery capacity or remaining lifespan) without any prior knowledge of battery kinetics. For data-driven methods, the key is to develop advanced algorithms and feature extraction methods for specific applications. This invention proposes a data-driven method for predicting the capacity degradation of in-use new energy vehicle power batteries. Summary of the Invention

[0004] For a large number of new energy vehicles in use, the actual capacity of the battery system is difficult to obtain, resulting in insufficient labeled capacity for training data-driven models. Furthermore, existing feature extraction techniques typically depend on operating conditions and lack universality. This invention proposes a novel method for obtaining labeled battery system capacity and a universal feature extraction technique. Using the aforementioned method, labeled capacity and a feature set characterizing battery capacity degradation are extracted from new energy vehicle charging data. Then, advanced data-driven algorithms are employed to establish a battery capacity degradation prediction model based on early-stage data.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for predicting the capacity degradation of power batteries in new energy vehicles includes the following steps:

[0007] S1: Extract tag capacity based on vehicle charging data, specifically including:

[0008] S11: Collect charging information of new energy vehicles using charging equipment, including time, total voltage, current, SOC, maximum voltage of individual cells, minimum voltage of individual cells, maximum temperature of individual cells, and minimum temperature of individual cells; record and store the above charging data; for each charging process, calculate the battery capacity data using the following modified ampere-hour integral formula.

[0009]

[0010] Among them, C a The calculated battery capacity is given by Δt, where Δt is the sampling interval, I is the charging current (negative during charging), and t1 and t2 are the start and end times of charging. For initial SOC of charging, The SOC is terminated for charging;

[0011] S12: Set a fixed time window, use multiple charging processes within the time window to calculate multiple battery capacities, and calculate the median value to obtain the tag capacity; use historical charging information to extract the tag capacity of the vehicle after 2-5 years of driving to obtain the capacity decay trajectory of the power battery; if the obtained capacity decay trajectory does not show a downward trend, it is necessary to adjust the time window or delete data with charging time shorter than the set threshold, and then recalculate the capacity decay trajectory.

[0012] S2: Extract a feature set characterizing battery capacity degradation from charging data; specifically including:

[0013] S21: Fixed time window, calculate the charging data within the time window, including: total voltage, current, SOC, highest single cell voltage, lowest single cell voltage, highest single cell temperature, lowest single cell temperature, single cell voltage difference, single cell temperature difference mean, sum, and standard deviation, and regard them as candidate features;

[0014] S22: Calculate the Pearson correlation coefficient and grey relational degree between different candidate features and battery capacity; wherein, the formula for calculating the Pearson correlation coefficient is:

[0015]

[0016] Where, x i Let z be the i-th feature sequence, and z be the capacity sequence. and ρ are the average values ​​of the i-th feature sequence and the capacity sequence, respectively. xi The correlation coefficient is r; the grey relational degree is r. i The calculation formula is:

[0017]

[0018]

[0019] Where n is the sequence length, k represents the k-th data point, ρ is the resolution coefficient, z(k) represents the capacity sequence of the k-th data point, and x i (k) represents the i-th feature sequence in the k-th data point;

[0020] S23: Using a feature selection strategy based on Pearson correlation coefficient and grey relational degree, select a feature set that is strongly correlated with battery capacity and has low correlation among features. The specific steps are as follows: Select features with Pearson correlation coefficient > 0.5 or grey relational degree > 0.8 as the candidate feature set, and sort them from largest to smallest according to Pearson correlation coefficient. Then, remove features with autocorrelation coefficient or autogrey relational degree > 0.9 from the candidate feature set in descending order. The remaining feature set is the optimal feature set. The autocorrelation coefficient represents the Pearson correlation coefficient between the current feature and the other features in the current candidate feature set; the grey relational degree represents the grey relational degree between the current feature and the other features in the current candidate feature set.

[0021] S3: Based on the extracted tag capacity and feature set, a data-driven method is used to predict the capacity degradation of the power battery; specifically including:

[0022] S31: Using a sequence-to-sequence neural network, with the first 20%-50% of the tag capacity and feature data as input and the subsequent tag capacity sequence as output, a power battery capacity decay trajectory prediction model is established; the prediction error of this model is defined as Δy.

[0023] S32: Using the Gaussian process regression algorithm, with the month, highest single cell temperature and lowest single cell temperature as inputs, and the prediction error Δy of the sequence-to-sequence model as output, a compensation model for the residual of power battery capacity prediction is established. This regression model can only perform single-point estimation. In order to predict the future capacity residual, it is necessary to know the future highest single cell temperature and lowest single cell temperature. In actual operation, it is set according to the temperature data of historical years.

[0024] S33: The final capacity decay prediction result is equal to the result of the sequence-to-sequence model plus the result of the Gaussian process regression model.

[0025] The beneficial effects of this invention are as follows:

[0026] This invention proposes a method for extracting the capacity of tags on in-use new energy vehicles, which extracts the capacity decay curve of the power battery from the vehicle's historical charging data.

[0027] This invention proposes a general feature engineering method for obtaining the optimal feature set from battery charging data. It is applicable to data with different charging strategies and different acquisition frequencies, and has good versatility for engineering applications.

[0028] This invention proposes a battery capacity prediction residual compensation model that can effectively compensate for local capacity prediction errors. Because battery capacity is affected by calendar time and temperature, the capacity decay curve exhibits local rises or falls. Most capacity prediction models based on neural networks can effectively capture the overall battery capacity decay trend, but cannot capture its local variations. The capacity prediction residual compensation model proposed in this invention can effectively compensate for local capacity prediction errors caused by differences in calendar time and temperature.

[0029] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0030] Figure 1 A flowchart for predicting the capacity degradation of power batteries;

[0031] Figure 2 Feature selection strategy;

[0032] Figure 3 A method for predicting the capacity degradation of power batteries;

[0033] Figure 4 This is the predicted result for the power battery capacity. Detailed Implementation

[0034] like Figure 1 As shown, the present invention mainly includes three steps.

[0035] Step 1: Battery Tag Capacity Extraction. Use charging equipment to collect charging information from new energy vehicles, including time, total voltage, current, SOC, maximum voltage of individual cells, minimum voltage of individual cells, maximum temperature of individual cells, and minimum temperature of individual cells. Record and store this charging data. For each charging process, calculate the battery capacity data using the following modified ampere-hour integral formula.

[0036]

[0037] Where C a The calculated battery capacity is given by Δt, where Δt is the sampling interval, I is the charging current (negative for charging), and t1 and t2 are the start and end times of charging. For initial SOC of charging, To terminate SOC (State of Charge), a fixed time window (1-3 months) is set. Multiple charging processes within this window are used to calculate multiple battery capacities, and the median value is calculated to obtain the tag capacity. Using historical charging information, the tag capacity over 2-5 years of vehicle operation can be extracted to obtain the battery capacity degradation trajectory. If the obtained capacity degradation trajectory does not show a clear trend, the time window needs to be adjusted or charging processes with short charging times need to be deleted.

[0038] Step Two: Battery Feature Extraction. First, within a fixed time window (1 month or 3 months), calculate the mean, sum, and standard deviation of charging data (total voltage, current, SOC, highest single-cell voltage, lowest single-cell voltage, highest single-cell temperature, lowest single-cell temperature, single-cell voltage difference, single-cell temperature difference) and consider them as candidate features. Then, calculate the Pearson correlation coefficient and grey relational degree between different candidate features and battery capacity. To ensure accuracy, calculate the correlation coefficient and grey relational degree between candidate features and battery capacity for multiple vehicles (e.g., 10 vehicles) and average them separately. The formula for calculating the Pearson correlation coefficient is...

[0039]

[0040] Where x i Let z be the i-th feature sequence, and z be the capacity sequence. and ρ are the average values ​​of the i-th feature sequence and the capacity sequence, respectively. xi The correlation coefficient is given by the formula for calculating the grey relational degree.

[0041]

[0042]

[0043] Where n is the sequence length, k represents the k-th data point, and ρ = 0.5 is the resolution coefficient. Finally, feature selection strategies are used, such as... Figure 2 As shown, a feature selection strategy is used to select a feature set that is strongly correlated with battery capacity and has low correlation between features based on Pearson correlation coefficient and grey relational degree. The specific steps are as follows: Select features with correlation coefficient > 0.5 or grey relational degree > 0.8 as the candidate feature set, and sort them from largest to smallest according to correlation coefficient. Then, gradually remove features with autocorrelation coefficient (correlation coefficient between features) or autogrey relational degree (grey relational degree between features) > 0.9 from the candidate feature set. The remaining feature set is the best feature set.

[0044] Step 3: Battery Capacity Degradation Prediction. Based on the extracted tag capacity and feature set, a prediction model is built using data-driven methods, such as... Figure 3As shown. First, a sequence-to-sequence neural network is used, with early (first 20%-50%) tag capacity and feature data as input and the future tag capacity sequence as output, to establish a power battery capacity degradation trajectory prediction model. The prediction error of this model is defined as Δy. Then, a Gaussian process regression algorithm is used, with month, highest single-cell temperature, and lowest single-cell temperature as input, and the prediction error Δy of the sequence-to-sequence model as output, to establish a compensation model for the power battery capacity prediction residual. This regression model can only perform single-point estimation. To predict future capacity residuals, the future highest and lowest single-cell temperatures need to be known. In practice, this can be set based on historical year temperature data. The final capacity degradation prediction result is equal to the result of the sequence-to-sequence model plus the result of the Gaussian process regression model, as shown. Figure 4 As shown.

Claims

1. A method for predicting the capacity degradation of power batteries in new energy vehicles, comprising the following steps: S1: Extract tag capacity based on vehicle charging data, specifically including: S11: Collect charging information of new energy vehicles using charging equipment, including time, total voltage, current, SOC, maximum voltage of individual cells, minimum voltage of individual cells, maximum temperature of individual cells, and minimum temperature of individual cells; record and store the above charging data; for each charging process, calculate the battery capacity data using the following modified ampere-hour integral formula. （1）； Among them, C a The calculated battery capacity is given by Δt, where Δt is the sampling interval, I is the charging current (negative during charging), and t1 and t2 are the start and end times of charging. For initial SOC of charging, The SOC is terminated for charging; S12: Set a fixed time window, use multiple charging processes within the time window to calculate multiple battery capacities, and calculate the median value to obtain the tag capacity; use historical charging information to extract the tag capacity of the vehicle after 2-5 years of driving to obtain the capacity decay trajectory of the power battery; if the obtained capacity decay trajectory does not show a downward trend, it is necessary to adjust the time window or delete data with charging time shorter than the set threshold, and then recalculate the capacity decay trajectory. S2: Extract a feature set characterizing battery capacity degradation from charging data; specifically including: S21: Fixed time window, calculate the charging data within the time window, including: total voltage, current, SOC, maximum voltage of a single cell, minimum voltage of a single cell, maximum temperature of a single cell, minimum temperature of a single cell, voltage difference of a single cell, temperature difference of a single cell, mean, sum, and standard deviation of a single cell, and treat them as candidate features; S22: Calculate the Pearson correlation coefficient and grey relational degree between different candidate features and battery capacity; wherein, the formula for calculating the Pearson correlation coefficient is: （2）； Where, x i For the i-th feature sequence, It is a capacity sequence. and These are the average values ​​of the i-th feature sequence and the capacity sequence, respectively. Correlation coefficient; grey relational degree The calculation formula is: （3）； （4）； Where n is the sequence length, k represents the k-th data point, and ρ is the resolution coefficient. This represents the capacity sequence of the k-th data point. This represents the i-th feature sequence in the k-th data point; S23: Using a feature selection strategy based on Pearson correlation coefficient and grey relational degree, select a feature set that is strongly correlated with battery capacity and has low correlation among features. The specific steps are as follows: Select features with Pearson correlation coefficient > 0.5 or grey relational degree > 0.8 as the candidate feature set, and sort them from largest to smallest according to Pearson correlation coefficient. Then, remove features with autocorrelation coefficient or autogrey relational degree > 0.9 from the candidate feature set in descending order. The remaining feature set is the optimal feature set. The autocorrelation coefficient represents the Pearson correlation coefficient between the current feature and the other features in the current candidate feature set; the grey relational degree represents the grey relational degree between the current feature and the other features in the current candidate feature set. S3: Based on the extracted tag capacity and feature set, a data-driven method is used to predict the capacity degradation of the power battery; specifically including: S31: Using a sequence-to-sequence neural network, with the first 20%-50% of the tag capacity and feature data as input and the subsequent tag capacity sequence as output, a power battery capacity decay trajectory prediction model is established; the prediction error of this model is defined as Δy. S32: Using the Gaussian process regression algorithm, with the month, the highest temperature of a single cell, and the lowest temperature of a single cell as inputs, and the prediction error Δy of the sequence-to-sequence model as output, a compensation model for the residual of power battery capacity prediction is established. This compensation model can only perform single-point estimation. In order to predict the future capacity residual, it is necessary to know the future highest temperature of a single cell and the lowest temperature of a single cell. In actual operation, it is set according to the temperature data of historical years. S33: The final capacity decay prediction result is equal to the result of the sequence-to-sequence model plus the result of the Gaussian process regression model.

Images