Electric vehicle battery life prediction method based on digital twinning

Abstract

The invention belongs to the technical field of battery life analysis, and provides an electric vehicle battery life prediction method based on digital twinning, which comprises the following steps: when detecting that an electric vehicle is in a collision working condition, triggering physical damage detection on a battery pack, identifying the type of physical damage existing in the battery pack, and determining the battery life of the battery pack; the physical damage types comprise dominant damage and recessive damage; wherein before the battery pack is subjected to physical damage type identification, a state judgment value is output by collecting speed data and vibration acceleration data. When battery pack physical damage detection is triggered under the collision working condition, whether the vehicle is in the vibration working condition or not is judged through the speed and vibration acceleration data, and then a differential detection strategy is adopted, so that the influence of vibration interference on a detection result is reduced, the accuracy of physical damage type recognition is improved, and particularly, hidden damage is captured.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of battery life analysis, and in particular relates to a battery life prediction method for electric vehicles based on digital twinning. BACKGROUND

[0002] With the popularity of electric vehicles, battery life prediction has become a core technology to ensure vehicle safety, economy and user experience.

[0003] However, electric vehicle batteries may encounter extreme conditions such as collisions during use. These scenarios not only directly threaten the physical integrity of the battery, but also the physical damage to the battery pack caused by the collision can be divided into explicit damage and implicit damage. Implicit damage accelerates capacity decay and even triggers catastrophic consequences such as thermal runaway. Traditional battery life prediction models are mainly based on laboratory cycle aging data, and are relatively lacking in capturing the complex effects of extreme events such as collisions on battery life, resulting in significant deviations between predicted results and actual decay trajectories.

[0004] Therefore, the application provides a battery life prediction method for electric vehicles based on digital twinning. SUMMARY

[0005] In order to make up for the deficiencies of the prior art and solve at least one technical problem raised in the background art.

[0006] The technical scheme adopted by the application to solve the technical problems is: a battery life prediction method for electric vehicles based on digital twinning, comprising the following steps:

[0007] When the electric vehicle is detected to be in a collision condition, physical damage detection of the battery pack is triggered, and the type of physical damage existing in the battery pack is identified. The type of physical damage includes explicit damage and implicit damage.

[0008] Before identifying the type of physical damage of the battery pack, a state judgment value is output by collecting speed data and vibration acceleration data, and it is judged whether the electric vehicle is currently in a vibration condition according to the state judgment value. Based on the judgment result, a differentiated physical damage type identification strategy is used to accurately identify the type of physical damage of the battery pack.

[0009] If the detected type of physical damage is explicit damage, the service life of the battery pack is terminated, and if the detected type of physical damage is implicit damage, the damage degree is evaluated.

[0010] If the damage degree evaluation result of the implicit damage is mild implicit damage, then according to the historical data of the battery pack of the electric vehicle, and combining a long short-term memory network model and a random forest regression model, the final life prediction value of the battery pack is fused and output.

[0011] As a further scheme of the present application: the process of judging whether the electric vehicle is currently in a vibration working condition is:

[0012] Based on the speed value and vibration acceleration data, the state judgment value is output. If the state judgment value is greater than or equal to the state judgment limit value, the electric vehicle is in a vibration working condition. If the state judgment value is less than the state judgment limit value, the electric vehicle is in a non-vibration working condition.

[0013] As a further scheme of the present application: the process of obtaining the state judgment value is

[0014] A collection time window is set. The speed value data and vibration acceleration data of the electric vehicle are obtained within the collection time window. The average speed value and the average vibration acceleration within the collection time window are taken;

[0015] The average speed value and the average vibration acceleration are respectively compared with their preset threshold values, and then the ratios are summed to obtain the state judgment value.

[0016] As a further scheme of the present application: the differentiated physical damage type identification strategy includes:

[0017] When in a non-vibration working condition, the explicit damage is determined by comparing the point cloud data deviation of the 3D modeling of the structured light camera with the reference model, and the implicit damage is determined by analyzing the temperature mean deviation and distribution dispersion deviation using infrared thermal imaging;

[0018] When in a vibration working condition, the identification process is set based on the identification process in a non-vibration working condition;

[0019] The explicit damage identification is corrected by quantifying the vibration impact on the point cloud data deviation calculation, and the implicit damage identification is updated by introducing a smoothing coefficient to update the current temperature mean reference value and the temperature standard deviation reference value, to obtain new temperature mean reference value and temperature standard deviation reference value.

[0020] As a further scheme of the present application: the process of comparing the point cloud data deviation is:

[0021] For each point in the real-time point cloud, search for the nearest neighbor point in the reference model, calculate the Euclidean distance, and calculate the average deviation of all point cloud corresponding Euclidean distances. Extract the maximum Euclidean distance value and the minimum Euclidean distance value in all point clouds, and calculate the difference to obtain the maximum deviation. The average deviation and the maximum deviation are output as point cloud data deviation. If the point cloud data deviation is greater than the point cloud data deviation standard, it is determined that the shell is abnormal and there is explicit damage.

[0022] As a further scheme of the present application: the process of determining the temperature mean deviation and distribution dispersion deviation using infrared thermal imaging is:

[0023] Collecting thermal images and calculating the mean and standard deviation of the battery pack surface temperature;

[0024] Subtracting the mean temperature from the mean temperature reference value, and then calculating the ratio with the temperature standard deviation reference value to obtain the mean temperature deviation; subtracting the temperature standard deviation from the temperature standard deviation reference value, and then calculating the ratio with the temperature standard deviation reference value to obtain the temperature distribution dispersion deviation;

[0025] If the mean temperature deviation is greater than or equal to 1.5 or the temperature distribution dispersion deviation is greater than or equal to 1.2, it is determined that there is a hidden damage.

[0026] As a further scheme of the present application: the process of correcting the point cloud data deviation calculation is:

[0027] The difference between the state judgment value and the state judgment limit value is calculated, and the product of the difference and 0.5 is calculated; the calculated vibration correction coefficient is the weight coefficient corresponding to the maximum deviation, and the weight corresponding to the average deviation is the difference between 1 and the vibration correction; the point cloud data deviation is weighted calculated by the average deviation and the maximum deviation.

[0028] As a further scheme of the present application: the process of obtaining the new mean temperature reference value and the temperature standard deviation reference value is:

[0029] Collecting thermal images and calculating the mean and standard deviation of the battery pack surface temperature;

[0030] The smoothing coefficient is the weight of the current mean temperature reference value, and 1 minus the smoothing coefficient is the weight of the mean temperature, which is weighted to obtain the new mean temperature reference value; the smoothing coefficient is the weight of the current temperature standard deviation reference value, and 1 minus the smoothing coefficient is the weight of the temperature standard deviation, which is weighted to obtain the new temperature standard deviation reference value.

[0031] As a further scheme of the present application: the process of evaluating the damage degree is:

[0032] Obtaining the mean temperature deviation and the temperature distribution dispersion deviation, and outputting the mean temperature deviation and the temperature distribution dispersion deviation to obtain a damage evaluation coefficient;

[0033] If the damage evaluation coefficient is less than the damage degree judgment threshold, it is determined to be a mild hidden damage.

[0034] As a further scheme of the present application: the process of outputting the final life prediction value is:

[0035] The historical temperature mean deviation and temperature distribution dispersion deviation of several time steps in the past are taken as inputs, and the remaining capacity prediction value after several cycles in the future is taken as output, and a long short-term memory network model is constructed

[0036] The damage evaluation coefficient, the current remaining capacity of the battery and the current cycle number of the battery are taken as inputs, and the remaining life prediction value is taken as output, and the unit of output is cycle number, and a random forest regression model is constructed;

[0037] The remaining capacity prediction value output by the long short-term memory model is converted into an equivalent remaining life prediction value based on the capacity attenuation relationship, as a first remaining life prediction value;

[0038] The output value of the random forest regression model is taken as a second remaining life prediction value;

[0039] The first remaining life prediction value and the second remaining life prediction value are fused to output a third remaining life prediction value as the final remaining life prediction value.

[0040] The beneficial effects of the present application are as follows:

[0041] 1、The present application triggers battery pack physical damage detection under collision working condition, first judges whether the vehicle is in vibration working condition through speed and vibration acceleration data, and then adopts differentiated detection strategy, thereby reducing the influence of vibration interference on detection result, improving the accuracy of physical damage type identification, especially the capture of implicit damage.

[0042] 2、The present application directly terminates the service life of the battery pack for the explicit damage, eliminates the risk of thermal runaway caused by structural damage from the safety level, and guarantees the operation safety of the electric vehicle, while the implicit damage is evaluated by the temperature mean deviation and distribution dispersion deviation to realize the quantitative characterization of the implicit aging state of the battery, and provide reliable basis for life prediction and safety decision.

[0043] 3、The present application combines the long short-term memory network model and the random forest regression model in the mild implicit damage scene, combines the current state and historical data of the battery to predict the life, and improves the prediction accuracy, robustness and generalization ability through the complementary advantages of the models, so that the prediction result is closer to the actual attenuation trajectory. BRIEF DESCRIPTION OF DRAWINGS

[0044] The present application will be further described below with reference to the accompanying drawings.

[0045] Figure 1 is a step flow chart of the present application based on a digital twin electric vehicle battery life prediction method;

[0046] Figure 2Is a kind of logic flow chart for judging whether vibration working condition in the battery life prediction method of electric vehicle based on digital twinning of the present application;

[0047] Figure 3 It is the architecture diagram of a kind of battery life prediction system of electric vehicle based on digital twinning of the present application. DETAILED DESCRIPTION

[0048] In order to make the technical means, creative features, purposes and effects realized by the present application easy to understand, the present application will be further described below in conjunction with specific embodiments.

[0049] Example 1

[0050] Please refer to Figure 1 And Figure 2 As shown in the figure, the battery life prediction method based on digital twinning of the electric vehicle of the embodiment of the present application comprises the following steps:

[0051] Step one: triggering physical damage detection of battery pack when detecting that the electric vehicle is in collision working condition;

[0052] In some embodiments, first, define the electric vehicle in collision working condition: real-time monitor the collision signal through acceleration sensor, collision sensor array, when detecting that the impact force exceeds the impact force threshold, determine that the electric vehicle is in collision working condition;

[0053] Wherein, the impact force threshold is summarized and set by those skilled in the art according to industry regulations and the characteristics of electric vehicle, which is a conventional setting;

[0054] When determining that the electric vehicle is in collision working condition, trigger the physical damage detection process of battery pack;

[0055] Before starting the physical damage detection process of battery pack, the driving state of the electric vehicle needs to be determined;

[0056] The driving state of the electric vehicle is determined because: in the process of physical damage type identification, the false image caused by vibration in the process of vehicle driving will significantly reduce the detection accuracy, and then may lead to inaccurate battery life prediction results, so when the vehicle is in different states, different detection and identification strategies need to be used;

[0057] The definition of the driving state of the electric vehicle includes: the electric vehicle is in non-vibration working condition and the electric vehicle is in vibration working condition;

[0058] Set the collection time window, obtain the speed value data and vibration acceleration data of the electric vehicle in the collection time window, take the mean value of speed value and the mean value of vibration acceleration in the collection time window;

[0059] The speed value average is compared with the speed threshold value to obtain a speed average ratio; the vibration acceleration average is compared with the vibration acceleration threshold value to obtain a vibration acceleration average ratio; the above ratio calculation process is to eliminate the influence of dimension on calculation;

[0060] The speed average ratio and the vibration acceleration average ratio are summed to obtain a state judgment value;

[0061] The state judgment value is compared with the state judgment limit value, if the state judgment value is greater than or equal to the state judgment limit value, it means that the electric vehicle is in a vibration working condition, if the state judgment value is less than the state judgment limit value, it means that the electric vehicle is in a non-vibration working condition;

[0062] It can be known by those skilled in the art that the setting of the vehicle speed threshold value and the vibration acceleration threshold value is based on the reference setting of the industry safety standard, referring to ISO15622 "Lane Keeping Assist System (LKAS) Performance Requirements", defining the low-speed working condition as the vehicle speed less than 20km / h, the vehicle dynamics response is controllable in this range, and the collision energy is low; according to SAEJ1030 "Automobile Vibration Test Standard", the human perception threshold (0.1g) and the structure safety threshold (0.5g) are determined; when the speed average ratio is 0 and the vibration acceleration average ratio is 1, it means pure high-speed smooth road, and when the speed average ratio is 1 and the vibration acceleration average ratio is 0, it means pure static strong vibration, both of which are equivalent to the mechanical load of the battery pack, so the theoretical state judgment limit value is 1, and in the actual working condition, the coupling effect (such as the superposition damage of high speed + vibration) needs to be considered, so the limit value needs to be adjusted by the engineering correction coefficient, and the state judgment limit value can be set to 0.8-0.9;

[0063] Step two: identifying the physical damage type of the battery pack through the physical damage detection process of the battery pack, wherein the physical damage type includes explicit damage and implicit damage;

[0064] In some embodiments, the physical damage type of the battery pack is identified based on the driving state of the electric vehicle;

[0065] It should be noted that the following identification process is to obtain the result of whether it exists or not;

[0066] Firstly, the identification process when the electric vehicle is in a non-vibration working condition is as follows:

[0067] For explicit damage identification, a structured light camera deployed at the bottom of the vehicle is used to model the battery pack shell in 3D to obtain real-time point cloud data, which is compared with the factory baseline model; the point cloud data deviation is calculated, if the point cloud data deviation is greater than the point cloud data deviation standard, it is determined that the shell is abnormal, and there is explicit damage, if the point cloud data deviation is less than or equal to the point cloud data deviation standard, there is no explicit damage.

[0068] The process of calculating the electric cloud data deviation is as follows: for each point in the real-time point cloud, the nearest neighbor point in the reference model is searched, the Euclidean distance is calculated, the average deviation is obtained by averaging the Euclidean distances of all point clouds; the maximum Euclidean distance value and the minimum Euclidean distance value in all point clouds are extracted, the difference is calculated, and the maximum deviation is obtained; the average deviation and the maximum deviation are weighted and summed to obtain the point cloud data deviation.

[0069] Wherein, it is known to those skilled in the art that the weights of the average deviation and the maximum deviation are equal, and the influence of the two on the point cloud data deviation is equivalent; and the point cloud data deviation standard is an industry specification known to those skilled in the art;

[0070] The identification of hidden damage is by the method of infrared thermal imaging; infrared thermal imaging converts invisible heat distribution into visual images by capturing the thermal radiation difference (temperature field distribution) of the surface of an object; when the battery pack has hidden damage (such as internal micro-cracks, poor electrode contact, electrolyte leakage), the heat conduction efficiency and electrochemical reaction rate of the damaged area will be abnormal, causing local temperature abnormalities, which can be identified by the temperature distortion of the thermal imaging graph;

[0071] The thermal image data is continuously collected for 5 minutes, and the mean value of the thermal images in the stable stage of the last 3 minutes is taken as the effective data;

[0072] Based on the collected thermal images, the mean value and the standard deviation of the battery pack surface temperature are calculated;

[0073] The difference between the temperature mean value and the temperature mean value reference value is calculated, and then the ratio with the temperature standard deviation reference value is calculated to obtain the temperature mean value deviation;

[0074] The difference between the temperature standard deviation and the temperature standard deviation reference value is calculated, and then the ratio with the temperature standard deviation reference value is calculated to obtain the temperature distribution dispersion deviation;

[0075] If the temperature mean value deviation is greater than or equal to 1.5 or the temperature distribution dispersion deviation is greater than or equal to 1.2, it is determined that there is hidden damage, otherwise, it is determined that there is no hidden damage;

[0076] Wherein, 1.5 corresponds to a temperature anomaly exceeding 1.5 times the standard deviation; under standard conditions (such as standing, low load operation), the temperature distribution dispersion of the battery pack is affected by factors such as environmental heat dissipation and material thermal conductivity, and the fluctuation is usually ≤20% (i.e. the temperature standard deviation is ≤1.2 times the temperature standard deviation reference value);

[0077] Wherein, the infrared thermal imaging data of the battery pack at the time of factory shipment is taken as the reference, or more than 3 data are collected under normal operating conditions (no collision history) to obtain the average value, to determine the temperature mean value reference value and the temperature standard deviation reference value;

[0078] Secondly, the identification process when the electric vehicle is in the vibration working condition is as follows:

[0079] The vibration can cause the surface temperature distribution of the battery pack to fluctuate temporarily, and dynamically updating the reference value can avoid misjudging the temperature change caused by normal vibration as hidden damage.

[0080] The identification process in the vibration working condition is set on the basis of the identification process in the non-vibration working condition, and needs to be optimized for vibration interference.

[0081] For the identification of the explicit damage, the point cloud data deviation calculation is corrected by quantifying the vibration influence.

[0082] A vibration correction coefficient is introduced, and the calculation process of the vibration correction coefficient is as follows: difference calculation is performed on the state judgment value and the state judgment limit value, and product calculation is performed on the difference and 0.5; the vibration correction coefficient obtained by calculation is the weight coefficient corresponding to the maximum deviation.

[0083] The point cloud data deviation is weighted calculation of the average deviation and the maximum deviation, and the weight corresponding to the average deviation is the difference between 1 and the vibration correction.

[0084] The stronger the vibration, the greater the influence of noise on the maximum deviation, so the weight corresponding to the maximum deviation is reduced when calculating the point cloud data deviation.

[0085] Then, it is judged whether there is linear damage by the point cloud data deviation.

[0086] For the identification of the hidden damage, the battery pack surface temperature data is collected in real time, a smoothing coefficient is introduced to update the current temperature mean reference value and the temperature standard deviation reference value, and new temperature mean reference value and temperature standard deviation reference value are obtained.

[0087] For the new temperature mean reference value: the smoothing coefficient is the weight of the current temperature mean reference value, and 1 minus the smoothing coefficient is the weight of the temperature mean, which are weighted to obtain the new temperature mean reference value; the smoothing coefficient is generally 0.9.

[0088] The influence of historical data and real-time data is balanced by the smoothing coefficient, so that the temperature mean reference value can dynamically adapt to the temperature change under the vehicle running state and the vibration working condition.

[0089] For the new temperature standard deviation reference value: the smoothing coefficient is the weight of the current temperature standard deviation reference value, and 1 minus the smoothing coefficient is the weight of the temperature standard deviation, which are weighted to obtain the new temperature standard deviation reference value.

[0090] The temperature standard deviation reference value is dynamically updated to reflect the real-time changes of the temperature distribution of the battery pack under the vibration working condition, and the misjudgment caused by short-term fluctuations is reduced;

[0091] Step three: based on the detected physical damage type, if it is an explicit damage, the service life of the battery pack is terminated, if it is an implicit damage, the damage degree is evaluated, according to the evaluation result of the damage degree of the implicit damage, combined with the historical data of the battery pack of the electric vehicle, a battery life prediction model is constructed to predict the battery life;

[0092] The reason for choosing to terminate the service life of the battery pack for the explicit damage is that the explicit damage has a direct and irreversible damage to the physical structure, electrical safety and environmental protection capability of the battery pack, and the thermal runaway risk caused by it is difficult to completely eliminate through maintenance. From the safety priority, maintenance feasibility and industry standard, terminating the service life of the explicit damage battery pack is a measure to ensure the safe operation of the electric vehicle;

[0093] In some embodiments, the damage degree of the implicit damage is evaluated to convert the implicit damage into a quantifiable index, which provides a basis for subsequent life prediction and safety decision. The implicit damage is a key inducement of battery performance degradation, and evaluating its degree can reflect the implicit aging state of the battery in real time;

[0094] Specifically, the process of evaluating the damage degree of the implicit damage is as follows:

[0095] Obtain the temperature mean deviation and the temperature distribution dispersion deviation, and output the temperature mean deviation and the temperature distribution dispersion deviation to obtain a damage evaluation coefficient;

[0096] The comprehensive output can be a weighted sum method. Those skilled in the art can know from experience that the temperature mean deviation reflects the overall thermal state, and the weight is higher. Therefore, the weight of the temperature mean deviation can be allocated as 0.6, and the weight of the temperature distribution dispersion deviation can be allocated as 0.4;

[0097] Based on the preset damage degree judgment threshold, if the damage evaluation coefficient is greater than or equal to the damage degree judgment threshold, it is determined that it is a severe implicit damage, and it is suggested to evaluate whether the battery pack needs to terminate the service life in advance;

[0098] The evaluation process includes but is not limited to: evaluating the safety of the battery for continued use;

[0099] If the damage evaluation coefficient is less than the damage degree judgment threshold, it is determined that it is a mild implicit damage, and a battery life prediction process is triggered;

[0100] Specifically, the process of the battery life prediction process includes constructing a battery life prediction model, and outputting a predicted value of the remaining life based on the battery life prediction model.

[0101] The input features include the current state and historical data;

[0102] The current state includes the damage assessment coefficient, the current remaining capacity of the battery, and the current cycle number of the battery;

[0103] The historical data includes a historical mean temperature deviation sequence and a temperature distribution dispersion deviation sequence, and a historical cycle number and capacity attenuation relationship, wherein the capacity attenuation relationship is that the capacity decreases by N percentage points per M cycles;

[0104] For example, the capacity attenuation relationship can be set to decrease by 0.5% per 100 cycles;

[0105] A fusion method combining a long short-term memory network model and a random forest regression model is selected; the fusion method is used to complement each other to improve the accuracy, robustness and generalization ability of the battery life prediction;

[0106] As known by those skilled in the art, when constructing the long short-term memory network model, the historical mean temperature deviation and the temperature distribution dispersion deviation of the past several time steps are taken as inputs, and the remaining capacity prediction value after several cycles in the future is taken as output;

[0107] When constructing the random forest regression model, the damage assessment coefficient, the current remaining capacity of the battery, and the current cycle number of the battery are taken as inputs, and the remaining life prediction value is taken as output, with the unit being cycle number;

[0108] The remaining capacity prediction value output by the long short-term memory model is converted into an equivalent remaining life prediction value based on the capacity attenuation relationship, as a first remaining life prediction value;

[0109] The output value of the random forest regression model is taken as a second remaining life prediction value;

[0110] The first remaining life prediction value and the second remaining life prediction value are weighted and fused to output a third remaining life prediction value as the final remaining life prediction value;

[0111] Optionally, battery usage management is performed based on the final life prediction value, including but not limited to: developing graded maintenance measures based on the remaining cycle number;

[0112] This embodiment has at least the following effects: when the battery pack physical damage detection is triggered under the collision working condition, it is first determined whether the vehicle is in a vibration working condition through the speed and vibration acceleration data, and then a differentiated detection strategy is adopted, thereby reducing the influence of vibration interference on the detection result and improving the accuracy of identifying the physical damage type (explicit damage and implicit damage), especially the capture of implicit damage;

[0113] Under non-vibration working conditions, the structural light camera 3D modeling and infrared thermal imaging can accurately identify the dominant damage (point cloud data deviation comparison) and the hidden damage (temperature field anomaly analysis); under vibration working conditions, by introducing a vibration correction coefficient to optimize the point cloud data deviation calculation, and using a smoothing coefficient to dynamically update the temperature reference value, the normal fluctuations caused by vibration are reduced to avoid misjudgment as damage, ensuring the reliability of hidden damage identification, making up for the defects of insufficient detection accuracy under complex working conditions, providing more accurate damage data support for subsequent battery life prediction and safety decision-making, and fundamentally improving the adaptability and robustness of the battery life prediction system to extreme working conditions;

[0114] The dominant damage directly terminates the service life of the battery pack, and the hidden damage is evaluated by the temperature mean deviation and distribution dispersion deviation to quantify the aging state of the battery, providing a reliable basis for life prediction and safety decision-making;

[0115] In the case of mild hidden damage, the long short-term memory network model and the random forest regression model are fused to predict the life of the battery based on the current state and historical data of the battery, and the advantages of the models are complementary to improve the accuracy, robustness and generalization ability of the prediction, so that the prediction result is closer to the actual attenuation trajectory;

[0116] Based on the final life prediction value, a graded maintenance measure is developed to realize fine management of battery use, which can reduce resource waste caused by premature replacement and timely find potential safety hazards, improve the economy and safety of battery use, and the dynamic updating mechanism of the reference value makes it better adapt to different working conditions, enhancing the practicality and environmental adaptability of the method;

[0117] Embodiment 2

[0118] Based on the same inventive concept as the battery life prediction method for electric vehicles based on digital twinning in the foregoing embodiments, as shown in Figure 3 The application provides a battery life prediction system for electric vehicles based on digital twinning, wherein the system specifically comprises:

[0119] The damage detection triggering module triggers the physical damage detection of the battery pack when detecting that the electric vehicle is in a collision working condition;

[0120] When it is detected that the electric vehicle is in a collision working condition, the physical damage detection of the battery pack is triggered, and before that, the speed value and vibration acceleration data are collected to calculate the speed mean ratio, vibration acceleration mean ratio and state judgment value, so as to judge whether the vehicle is in a vibration working condition, thereby providing a basis for the subsequent detection strategy;

[0121] The damage type module: through the physical damage detection process of the battery pack, the physical damage type existing in the battery pack, wherein the physical damage type includes the explicit damage and the implicit damage;

[0122] The battery pack physical damage type is identified based on the vehicle driving state, under the non-vibration working condition, the explicit damage is determined by comparing the point cloud data deviation of the 3D modeling of the structural light camera with the reference model, the implicit damage uses the infrared thermal imaging to analyze the temperature mean deviation and the distribution dispersion deviation, under the vibration working condition, the vibration correction coefficient is introduced into the point cloud data deviation calculation of the explicit damage identification, and the implicit damage identification is updated dynamically through the smoothing coefficient;

[0123] The life prediction module: based on the detected physical damage type, if it is the explicit damage, the service life of the battery pack is terminated, if it is the implicit damage, the damage degree is evaluated, according to the damage degree evaluation result of the implicit damage, the battery life prediction model is constructed by combining the historical data of the battery pack of the electric vehicle, and the battery life is predicted;

[0124] If the detection is the explicit damage, the service life of the battery pack is terminated, if it is the implicit damage, the damage evaluation coefficient is calculated to evaluate the damage degree, when the implicit damage is slight, the current state of the battery (damage evaluation coefficient, residual capacity, cycle number) and the historical data (temperature deviation sequence, cycle and capacity attenuation relationship) are combined, the long short-term memory network model and the random forest regression model are fused, the final prediction result is obtained by weighting and fusing the residual life prediction values output by the two.

[0125] The above shows and describes the basic principles, main features and advantages of the present application. It should be understood by those skilled in the art that the present application is not limited to the above embodiments, the above embodiments and the description in the specification are only to illustrate the principles of the present application, and various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the battery life of an electric vehicle based on digital twins, characterized in that: Includes the following steps: When a collision is detected in an electric vehicle, physical damage detection of the battery pack is triggered to identify the types of physical damage present in the battery pack, including overt and covert damage. Before identifying the physical damage type of the battery pack, a state judgment value is output by collecting velocity data and vibration acceleration data. Based on the state judgment value, it is determined whether the electric vehicle is currently in a vibration condition. Based on the judgment result, a differentiated physical damage type identification strategy is implemented. The physical damage type identification strategy is used to accurately identify the physical damage type of the battery pack. If the detected physical damage type is overt damage, the battery pack's service life will be terminated; if the detected physical damage type is covert damage, the degree of damage will be assessed. If the assessment result of the latent damage is mild latent damage, then based on the historical data of the electric vehicle's battery pack, and combined with the long short-term memory network model and the random forest regression model, the final life prediction value of the battery pack is output.

2. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 1, characterized in that: The process of determining whether an electric vehicle is currently in a vibration condition is as follows: Based on the velocity and vibration acceleration data, a state judgment value is output. If the state judgment value is greater than or equal to the state judgment limit, the electric vehicle is in a vibration condition. If the state judgment value is less than the state judgment limit, the electric vehicle is in a non-vibration condition.

3. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 2, characterized in that: The process of obtaining the state judgment value is as follows: Set a data acquisition time window, acquire the speed and vibration acceleration data of the electric vehicle within the acquisition time window, and take the average speed and vibration acceleration values ​​within the acquisition time window; The average velocity value and the average vibration acceleration value are compared with their preset thresholds, and then the ratios are summed to obtain the state judgment value.

4. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 1, characterized in that: Differentiated physical damage type identification strategies include: When in non-vibration conditions, visible damage is determined by comparing point cloud data deviations between 3D modeling with a benchmark model using a structured light camera, while latent damage is determined by analyzing temperature mean deviation and distribution dispersion deviation using infrared thermal imaging. When the system is in a vibration condition, the identification process is set up based on the identification process under a non-vibration condition. For visible damage identification, the deviation calculation of point cloud data is corrected by quantifying the impact of vibration. For latent damage identification, the surface temperature data of the battery pack is collected in real time, and a smoothing coefficient is introduced to update the current temperature mean benchmark value and temperature standard deviation benchmark value, so as to obtain new temperature mean benchmark value and temperature standard deviation benchmark value.

5. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 4, characterized in that: The process for determining the deviation of the compared point cloud data is as follows: For each point in the real-time point cloud, search for the nearest neighbor in the benchmark model, calculate the Euclidean distance, and calculate the average deviation by averaging the Euclidean distances of all point clouds. Extract the maximum and minimum Euclidean distance values ​​from all point clouds, calculate the difference, and obtain the maximum deviation. Output the point cloud data deviation by comparing the average deviation with the maximum deviation. If the point cloud data deviation is greater than the point cloud data deviation standard, it is determined that the shell is abnormal and there is obvious damage.

6. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 4, characterized in that: The process of using infrared thermal imaging to analyze and determine the mean temperature deviation and distribution dispersion deviation is as follows: Acquire thermal images and calculate the mean and standard deviation of the battery pack surface temperature; The temperature mean deviation is calculated by subtracting the temperature mean from the baseline temperature mean and then comparing it with the temperature standard deviation baseline. The temperature distribution dispersion deviation is calculated by subtracting the temperature standard deviation from the baseline temperature standard deviation and then comparing it with the temperature standard deviation baseline. If the mean temperature deviation is greater than or equal to 1.5 or the temperature distribution dispersion deviation is greater than or equal to 1.2, then latent damage is determined to exist.

7. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 4, characterized in that: The process of correcting the deviation calculation of point cloud data is as follows: The difference between the state judgment value and the state judgment limit is calculated, and the difference is multiplied by 0.

5. The calculated vibration correction coefficient is the weight coefficient corresponding to the maximum deviation. The weight corresponding to the average deviation is the difference between 1 and the vibration correction. The point cloud data deviation is calculated by weighting the average deviation and the maximum deviation.

8. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 4, characterized in that: The process for obtaining the new temperature mean baseline value and temperature standard deviation baseline value is as follows: Acquire thermal images and calculate the mean and standard deviation of the battery pack surface temperature; The current temperature mean baseline value is weighted by subtracting the weight of the smoothing coefficient from the current temperature mean value by 1, and then weighted to obtain a new temperature mean baseline value. The current temperature standard deviation baseline value is weighted by subtracting the weight of the smoothing coefficient from the current temperature standard deviation value by 1, and then weighted to obtain a new temperature standard deviation baseline value.

9. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 1, characterized in that: The process for assessing the degree of damage is as follows: The mean temperature deviation and the temperature distribution dispersion deviation are obtained, and the mean temperature deviation and the temperature distribution dispersion deviation are combined and output to obtain the damage assessment coefficient. If the damage assessment coefficient is less than the damage severity judgment threshold, it is judged as mild latent damage.

10. The method for predicting the battery life of an electric vehicle based on digital twins according to claim 9, characterized in that: The process of outputting the final lifetime prediction value is as follows: Using the historical temperature mean deviation and temperature distribution dispersion deviation over several past time steps as input, and the predicted remaining capacity after several future iterations as output, a Long Short-Term Memory (LSTM) network model is constructed. Using damage assessment coefficients, current remaining battery capacity, and current battery cycle count as inputs, and remaining life prediction as output, with the output unit being the number of cycles, a random forest regression model is constructed. The remaining capacity prediction value output by the Long Short-Term Memory model is converted into an equivalent remaining lifetime prediction value based on the capacity decay relationship, and used as the first remaining lifetime prediction value. The output value of the random forest regression model is used as the second remaining lifetime prediction value; The first remaining lifetime prediction value is merged with the second remaining lifetime prediction value to output the third remaining lifetime prediction value, which is used as the final remaining lifetime prediction value.

Images