A Method for Constructing Evaluation Indicators for Power Batteries of Pure Electric Vehicles Based on Big Data Analysis

Abstract

This invention discloses a method for constructing evaluation indicators for power batteries of pure electric vehicles based on big data analysis, including the following steps: A1. Marking the vehicles; A2. During the operation and use of the P1th vehicle; A3. The P2th...P... n Repeat steps A2 and A5 for each new car, and remove the first DB. ND The parameters are based on the data before the 1% mark and the data after the 99th mark following a normal distribution. A6. Use the labeled samples as the target values. A7. Use neural network models MM1...MM. n The calculated values ​​at each sampling point are used as the dependent variable. A8. Regression analysis is used. A9. One or more regression equations with the lowest overall error rate are extracted from the multiple regression equations and used as a continuous parameter prediction model. Compared with traditional neural network model calculation methods, this invention significantly reduces the computational load. Compared with single correlation regression prediction methods, it has better accuracy and adaptability, and has broad practical application value.

Description

Technical Field

[0001] This invention relates to the field of power battery evaluation technology, and in particular to a method for constructing evaluation indicators for pure electric vehicle power batteries based on big data analysis. Background Technology

[0002] Currently, there are many testable parameters for the production process of new batteries for pure electric vehicle power batteries. These include parameters that are mandatory by national standards, as well as parameters that are evaluated and used by individual manufacturers.

[0003] However, for consumers, the most important parameter is the battery capacity degradation rate. Figure 1 This refers to the battery capacity retention rate of the Tesla Model S / Model X after 200,000 miles (approximately 320,000 kilometers). Currently, most electric vehicle manufacturers stipulate that the battery pack's capacity is considered to have reached the end of its life when it has degraded to 80%, at which point the battery can be replaced.

[0004] Therefore, for pure electric vehicles of the same type and with the same power battery specifications, the question of when to replace the battery after use can be addressed by constructing a reverse neural network based on the actual operating conditions of multiple pure electric vehicles when they reach a critical state (capacity decay to 80%), the corresponding technical parameters of the power battery, or the driving parameters of the vehicle. This can yield a more accurate evaluation index for the power battery after a period of use. Only when this evaluation index is applicable to most vehicles of this type can it have practical application value and, with good accuracy, reasonably protect consumer rights.

[0005] Existing methods for evaluating power battery performance or predicting lifespan mostly rely on neural network models. These models require extracting continuous parameters for modeling, making them unsuitable for certain ordinal parameters. For example, they cannot model based on road conditions or effectively consider the impact of environmental factors such as ambient temperature and solar radiation intensity on battery and vehicle lifespan. Furthermore, in the current process of constructing multi-parameter neural network models, some parameters are considered manually, and no corresponding scientific analysis has been conducted to determine whether these parameters are correlated with power battery lifespan. Therefore, effective correlation analysis is lacking in the multi-parameter construction process. Summary of the Invention

[0006] This invention provides a method for constructing evaluation indicators for power batteries of pure electric vehicles based on big data analysis.

[0007] The solution of the present invention is:

[0008] The method for constructing evaluation indicators for power batteries of pure electric vehicles based on big data analysis includes the following steps:

[0009] A1. Label several vehicles of the same type and equipped with the same power battery as P1, P2, ..., P1. n The vehicle is equipped with a communication module that uploads data to the server via communication equipment;

[0010] A2. During the operation of vehicle P1, the vehicle's power battery capacity decay rate C% is collected at a certain sampling period T. Multiple C% values ​​collected at multiple sampling periods T are used as calibration samples. At the same time, vehicle parameter index set [V]1 and power battery parameter index set [B]1 are collected synchronously at the sampling period T. The [V]1 includes multiple vehicle-related parameters during vehicle operation, and the [B]1 includes multiple parameters related to the power battery installed in the vehicle. The calibration samples, [V]1, and [B]1 are uploaded to the server.

[0011] A3, Page 2...P n Repeat step A2 for each new car. When the obtained C% value is different, the corresponding multiple vehicle parameter index sets [V]2……[V] are... n And multiple sets of power battery parameter indicators [B]2……[B] n ;

[0012] A4. Regarding [V]1……[V] n The multiple parameters in the data are statistically classified into ordinal parameters (DX) or continuous parameters (DB), and [B]1...[B] n The multiple parameters in the data are statistically classified into ordinal parameters DX or continuous parameters DB.

[0013] For parameter DB, extract n parameters from n vehicles within the same sampling period T, and test the normal distribution of these n parameters. If a parameter does not conform to a normal distribution, discard that sampling point and its n parameter values; if a parameter conforms to a normal distribution, retain it. When the number of retained sampling points exceeds 90% of the total sampling points, define the remaining data in the corresponding parameter DB as DB. ND Parameters are retained;

[0014] A5. Remove the first DB. ND The parameters are arranged according to the data before the first 1% and after the 99th percentile of the normal distribution, retaining the envelope data from the 1% to 99th percentile of the normal distribution; step A5 is repeated in this manner for all remaining databases. ND Parameters are processed;

[0015] A6. Using the calibration sample as the target value, the first DB ND The parameter data is used as input parameters, and a neural network model MM1 is built using a CNN convolutional neural network; and so on, multiple DBs are used. NDThe parameters yield multiple neural network models MM2...MM n ;

[0016] A7. Using neural network models MM1...MM n The calculated values ​​at each sampling point are used as dependent variables, and the calibration samples are used as independent variables. Correlation analysis is used to analyze the correlation between the calibration samples and the calculated values ​​of multiple neural network models. The correlation coefficient γ and the significance test probability ρ% are obtained. One or more neural network models with the significance test probability ρ% < 0.05 and the largest correlation coefficient γ value are extracted to form a continuous model group [MM].

[0017] A8. Using regression analysis, establish a regression equation between the calibration sample and the calculated values ​​of a neural network model in the continuous model group [MM]; the number of regression equations established should be consistent with the number of calculated values ​​of the neural network model in [MM].

[0018] A9. Select m more vehicles, use the regression equation in step A8 to predict the value of each C% according to the sampling period, and compare the prediction results with the C% in the actual driving process of the m vehicles. The comprehensive error rate of the comparison results is ≤5% and is judged as qualified. The comprehensive error rate is the root mean square value of the error rate at all sampling points.

[0019] One or more regression equations with the lowest overall error rate are extracted from multiple regression equations and used as continuous parameter prediction models.

[0020] As a preferred technical solution, step A7 further includes the following steps:

[0021] A7-1. Using correlation analysis, perform correlation analysis between the data in the first ordinal parameter DX and the calculated values ​​of all mathematical models MM in the continuous model group [MM].

[0022] A7-2. If the correlation level between the ordinal parameter DX and more than 50% of the mathematical models MM in the continuous model group [MM] is level 3 or above, then proceed to the next step.

[0023] A7-3. Calculate the percentage of each ordinal data in the corresponding DX parameter of the vehicle when the vehicle's power battery capacity decay rate C% changes from 100% to 80% under the sampling period T; the percentage Z of the first ordinal data in the first DX parameter. 1-1 %, the proportion of the second ordinal data Z 1-2 %...the proportion of the Zth ordered data Z 1-z %; Calculate the percentage of each ordered data in the first DX parameter of each of the n vehicles; form a Q×n×Z matrix; where Q is the number of samplings;

[0024] A7-4. Using the Q×n×Z matrix as the input layer, and taking the change in battery capacity decay rate C% from 100% to 80% as the target value, a CNN convolutional neural network model is used for training to obtain a multi-layer neural network model.

[0025] A7-5. Repeat step A7-1 for the remaining ordinal parameters DX respectively; then in step A9, select m more vehicles, and use the multilayer neural network model to predict the value of each C% in sequence according to the sampling frequency of the interval, and compare the prediction result with the C% in the actual driving process of the m vehicles. If the comprehensive error rate of the comparison result is ≤5%, it is judged as qualified.

[0026] One or more multi-layer neural network models with the lowest overall error rate are extracted from multiple multi-layer neural network models and used as ordinal parameter prediction models.

[0027] As a preferred technical solution, the vehicle parameter index set [V] includes the vehicle's cumulative mileage L, the vehicle's cumulative usage time Time1, acceleration frequency, braking frequency, and road conditions; the acceleration frequency is the acceleration frequency / a single sampling period; the braking frequency is the braking frequency / a single sampling period.

[0028] The power battery related indicator set [B] includes the number of charge cycles N1, cumulative charging time Time3; open circuit voltage U1, operating voltage U2, internal resistance R1, self-discharge rate ΔC; maximum cumulative working time of a single battery module Time2; the open circuit voltage U1 and operating voltage U2 are the minimum measured values ​​of a single battery module in the power battery pack under normal operating conditions; the internal resistance R1 and self-discharge rate ΔC are the maximum measured values ​​of a single battery module in the power battery pack under normal operating conditions.

[0029] As a preferred technical solution, the cumulative mileage L, cumulative vehicle usage time Time1, acceleration frequency, and braking frequency in the vehicle parameter index set [V] and the cycle charging number N1, cumulative charging time Time3, maximum cumulative working time of a single battery module Time2, open circuit voltage U1, working voltage U2, internal resistance R1, and self-discharge rate ΔC in the power battery related index set [B] are set as continuous parameters DB in step A4; the driving road conditions in the vehicle parameter index set [V] are set as ordinal parameters DX in step A4; at the same time, the ambient temperature parameter Temp-1 under vehicle driving conditions, the ambient temperature parameter Temp-2 under vehicle parking conditions, and the solar radiation intensity SH are also collected, and Temp-1, Temp-2, and SH are set as ordinal parameters DX in step A4; the acceleration frequency in the vehicle parameter index set [V] is the acceleration frequency / single sampling period, and the braking frequency is the braking frequency / single sampling period.

[0030] As a preferred technical solution, the driving conditions in the vehicle parameter index set [V] are divided into four ordered data: urban congestion conditions, general urban conditions, suburban conditions, and highway conditions. In the vehicle parameter index set [V], a braking frequency > 7.73 indicates urban congestion conditions; a braking frequency ≤ 2.8 and < 7.73 indicates general urban conditions; a braking frequency ≤ 1.82 and < 2.8 indicates suburban conditions; and a braking frequency < 1.82 indicates highway conditions. The braking frequency is the average number of braking actions per kilometer.

[0031] The solar radiation intensity SH is divided into 5 ordinal data, which are classified into Class I, Class II, Class III and Class IV regions.

[0032] The first category of areas refers to regions with 3200-3300 hours of daily sunshine per year and an annual radiation dose of 1860-2330 kW·h / m². 2 ;

[0033] The second category of areas refers to regions with 3000-3200 hours of sunshine per year and an annual radiation of 1630-1860 kW·h / m². 2 ;

[0034] The three categories of areas are defined as areas with 2200-3000 hours of sunshine per year and annual radiation ranging from 1390-1630 kW·h / m². 2 ;

[0035] The four categories of areas are defined as areas with 1400-2200 hours of sunshine per year and annual radiation ranging from 1160-1390 kW·h / m². 2 ;

[0036] Category V areas: Annual sunshine hours of approximately 1000-1400 hours, annual radiation of 933-1160 kW·h / m² 2 .

[0037] As a preferred technical solution, the correlation level in step A7-1 is divided into the following four levels.

[0038] The data in the first ordinal parameter DX is used as the first column of data, and the calculated values ​​of all mathematical models MM in the continuous model group [MM] are used as the second column of data.

[0039] Level 1 consists of data in the first and second columns that have absolutely no relation to each other.

[0040] Level 2 shows a vague correlation between the data in the first and second columns;

[0041] Level 3 shows a strong, but not clear, functional relationship between the first and second columns of data.

[0042] Level 4 shows a clear functional relationship between the first and second columns of data.

[0043] As a preferred technical solution, the correlation analysis method is Spearman analysis or covariance analysis.

[0044] As a preferred technical solution, the sampling period T in step A2 is set to either a certain time interval (Time) or a certain distance interval (Trip).

[0045] The sampling period T is set according to a certain mileage Trip, where the sampling period T is one of the driving mileages of 0.5 million kilometers, 1 million kilometers, 2 million kilometers, and 2.5 million kilometers.

[0046] The sampling period T is set according to a certain time, which is one of 3 months, 6 months, 8 months and 1 year.

[0047] As a preferred technical solution, the communication device in step A1 is an Internet communication device or a mobile communication device.

[0048] The method for constructing evaluation indicators for pure electric vehicle power batteries based on big data analysis, as described above, includes the following steps: A1. Several vehicles of the same type and equipped with the same power battery are labeled as P1, P2...P... n The vehicle is equipped with a communication module that uploads data to the server via communication equipment;

[0049] A2. During the operation of vehicle P1, the vehicle's power battery capacity decay rate C% is collected at a certain sampling period T. Multiple C% values ​​collected at multiple sampling periods T are used as calibration samples. At the same time, vehicle parameter index set [V]1 and power battery parameter index set [B]1 are collected synchronously at the sampling period T. The [V]1 includes multiple vehicle-related parameters during vehicle operation, and the [B]1 includes multiple parameters related to the power battery installed in the vehicle. The calibration samples, [V]1, and [B]1 are uploaded to the server.

[0050] A3, Page 2...P n Repeat step A2 for each new car. When the obtained C% value is different, the corresponding multiple vehicle parameter index sets [V]2……[V] are... n And multiple sets of power battery parameter indicators [B]2……[B] n ;

[0051] A4. Regarding [V]1……[V] n The multiple parameters in the data are statistically classified into ordinal parameters (DX) or continuous parameters (DB), and [B]1...[B]n The multiple parameters in the data are statistically classified into ordinal parameters DX or continuous parameters DB.

[0052] For parameter DB, extract n parameters from n vehicles within the same sampling period T, and test the normal distribution of these n parameters. If a parameter does not conform to a normal distribution, discard that sampling point and its n parameter values; if a parameter conforms to a normal distribution, retain it. When the number of retained sampling points exceeds 90% of the total sampling points, define the remaining data in the corresponding parameter DB as DB. ND Parameters are retained;

[0053] A5. Remove the first DB. ND The parameters are arranged according to the data before the first 1% and after the 99th percentile of the normal distribution, retaining the envelope data from the 1% to 99th percentile of the normal distribution; step A5 is repeated in this manner for all remaining databases. ND Parameters are processed;

[0054] A6. Using the calibration sample as the target value, the first DB ND The parameter data is used as input parameters, and a neural network model MM1 is built using a CNN convolutional neural network; and so on, multiple DBs are used. ND The parameters yield multiple neural network models MM2...MM n ;

[0055] A7. Using neural network models MM1...MM n The calculated values ​​at each sampling point are used as dependent variables, and the calibration samples are used as independent variables. Correlation analysis is used to analyze the correlation between the calibration samples and the calculated values ​​of multiple neural network models. The correlation coefficient γ and the significance test probability ρ% are obtained. One or more neural network models with the significance test probability ρ% < 0.05 and the largest correlation coefficient γ value are extracted to form a continuous model group [MM].

[0056] A8. Using regression analysis, establish a regression equation between the calibration sample and the calculated values ​​of a neural network model in the continuous model group [MM]; the number of regression equations established should be consistent with the number of calculated values ​​of the neural network model in [MM].

[0057] A9. Select m more vehicles, use the regression equation in step A8 to predict the value of each C% according to the sampling period, and compare the prediction results with the C% in the actual driving process of the m vehicles. The comprehensive error rate of the comparison results is ≤5% and is judged as qualified. The comprehensive error rate is the root mean square value of the error rate at all sampling points.

[0058] One or more regression equations with the lowest overall error rate are extracted from multiple regression equations and used as continuous parameter prediction models.

[0059] The advantages of this invention are as follows: First, the data collected from each vehicle under continuous parameters is statistically analyzed using a normal distribution to make the data statistically significant. Then, a CNN convolutional neural network model is constructed from the collected data from multiple vehicles. This neural network model is then used to output a series of calculated values, transforming the collected data from multiple vehicles into a series of statistically significant calculated values ​​with good predictive accuracy. Next, regression analysis is used in statistics to analyze the correlation between the vehicle's power battery capacity decay rate C% and the calculated values ​​of each neural network model. The regression equation with significant correlation is extracted as a predictive model for power battery performance or lifespan. Compared with traditional neural network model calculation methods, the computational load is greatly reduced. Compared with single correlation regression prediction methods, the accuracy and adaptability are better, and it has broad practical application value. Attached Figure Description

[0060] Figure 1 This diagram illustrates the battery capacity retention rate of the Tesla Model S / Model X after 200,000 miles (approximately 320,000 kilometers).

[0061] Figure 2 This is a schematic diagram of the method for constructing evaluation indicators for the power battery of a pure electric vehicle in Example 1;

[0062] Figure 3 This is a schematic diagram of the method for constructing evaluation indicators for the power battery of a pure electric vehicle in Example 2. Detailed Implementation

[0063] To overcome the above deficiencies, this invention provides a method for constructing evaluation indicators for pure electric vehicle power batteries based on big data analysis to solve the problems mentioned in the background art.

[0064] A1. Label several vehicles of the same type and equipped with the same power battery as P1, P2, ..., P1. n The vehicle is equipped with a communication module that uploads data to the server via communication equipment;

[0065] A2. During the operation of vehicle P1, the vehicle's power battery capacity decay rate C% is collected at a certain sampling period T. Multiple C% values ​​collected at multiple sampling periods T are used as calibration samples. At the same time, vehicle parameter index set [V]1 and power battery parameter index set [B]1 are collected synchronously at the sampling period T. The [V]1 includes multiple vehicle-related parameters during vehicle operation, and the [B]1 includes multiple parameters related to the power battery installed in the vehicle. The calibration samples, [V]1, and [B]1 are uploaded to the server.

[0066] A3, Page 2...P nRepeat step A2 for each new car. When the obtained C% value is different, the corresponding multiple vehicle parameter index sets [V]2……[V] are... n And multiple sets of power battery parameter indicators [B]2……[B] n ;

[0067] A4. Regarding [V]1……[V] n The multiple parameters in the data are statistically classified into ordinal parameters (DX) or continuous parameters (DB), and [B]1...[B] n The multiple parameters in the data are statistically classified into ordinal parameters DX or continuous parameters DB.

[0068] For parameter DB, extract n parameters from n vehicles within the same sampling period T, and test the normal distribution of these n parameters. If a parameter does not conform to a normal distribution, discard that sampling point and its n parameter values; if a parameter conforms to a normal distribution, retain it. When the number of retained sampling points exceeds 90% of the total sampling points, define the remaining data in the corresponding parameter DB as DB. ND Parameters are retained;

[0069] A5. Remove the first DB. ND The parameters are arranged according to the data before the first 1% and after the 99th percentile of the normal distribution, retaining the envelope data from the 1% to 99th percentile of the normal distribution; step A5 is repeated in this manner for all remaining databases. ND Parameters are processed;

[0070] A6. Using the calibration sample as the target value, the first DB ND The parameter data is used as input parameters, and a neural network model MM1 is built using a CNN convolutional neural network; and so on, multiple DBs are used. ND The parameters yield multiple neural network models MM2...MM n ;

[0071] A7. Using neural network models MM1...MM n The calculated values ​​at each sampling point are used as dependent variables, and the calibration samples are used as independent variables. Correlation analysis is used to analyze the correlation between the calibration samples and the calculated values ​​of multiple neural network models. The correlation coefficient γ and the significance test probability ρ% are obtained. One or more neural network models with the significance test probability ρ% < 0.05 and the largest correlation coefficient γ value are extracted to form a continuous model group [MM].

[0072] A8. Using regression analysis, establish a regression equation between the calibration sample and the calculated values ​​of a neural network model in the continuous model group [MM]; the number of regression equations established should be consistent with the number of calculated values ​​of the neural network model in [MM].

[0073] A9. Select m more vehicles, use the regression equation in step A8 to predict the value of each C% according to the sampling period, and compare the prediction results with the C% in the actual driving process of the m vehicles. The comprehensive error rate of the comparison results is ≤5% and is judged as qualified. The comprehensive error rate is the root mean square value of the error rate at all sampling points.

[0074] One or more regression equations with the lowest overall error rate are extracted from multiple regression equations and used as continuous parameter prediction models.

[0075] As a preferred technical solution, step A7 further includes the following steps:

[0076] A7-1. Using correlation analysis, perform correlation analysis between the data in the first ordinal parameter DX and the calculated values ​​of all mathematical models MM in the continuous model group [MM].

[0077] A7-2. If the correlation level between the ordinal parameter DX and more than 50% of the mathematical models MM in the continuous model group [MM] is level 3 or above, then proceed to the next step.

[0078] A7-3. Calculate the percentage of each ordinal data in the corresponding DX parameter of the vehicle when the vehicle's power battery capacity decay rate C% changes from 100% to 80% under the sampling period T; the percentage Z of the first ordinal data in the first DX parameter. 1-1 %, the proportion of the second ordinal data Z 1-2 %...the proportion of the Zth ordered data Z 1-z %; Calculate the percentage of each ordered data in the first DX parameter of each of the n vehicles; form a Q×n×Z matrix; where Q is the number of samplings;

[0079] A7-4. Using the Q×n×Z matrix as the input layer, and taking the change in battery capacity decay rate C% from 100% to 80% as the target value, a CNN convolutional neural network model is used for training to obtain a multi-layer neural network model.

[0080] A7-5. Repeat step A7-1 for the remaining ordinal parameters DX respectively; then in step A9, select m more vehicles, and use the multilayer neural network model to predict the value of each C% in sequence according to the sampling frequency of the interval, and compare the prediction result with the C% in the actual driving process of the m vehicles. If the comprehensive error rate of the comparison result is ≤5%, it is judged as qualified.

[0081] One or more multi-layer neural network models with the lowest overall error rate are extracted from multiple multi-layer neural network models and used as ordinal parameter prediction models.

[0082] As a preferred technical solution, the vehicle parameter index set [V] includes the vehicle's cumulative mileage L, the vehicle's cumulative usage time Time1, acceleration frequency, braking frequency, and road conditions; the acceleration frequency is the acceleration frequency / a single sampling period; the braking frequency is the braking frequency / a single sampling period.

[0083] The power battery related indicator set [B] includes the number of charge cycles N1, cumulative charging time Time3; open circuit voltage U1, operating voltage U2, internal resistance R1, self-discharge rate ΔC; maximum cumulative working time of a single battery module Time2; the open circuit voltage U1 and operating voltage U2 are the minimum measured values ​​of a single battery module in the power battery pack under normal operating conditions; the internal resistance R1 and self-discharge rate ΔC are the maximum measured values ​​of a single battery module in the power battery pack under normal operating conditions.

[0084] As a preferred technical solution, the cumulative mileage L, cumulative vehicle usage time Time1, acceleration frequency, and braking frequency in the vehicle parameter index set [V] and the cycle charging number N1, cumulative charging time Time3, maximum cumulative working time of a single battery module Time2, open circuit voltage U1, working voltage U2, internal resistance R1, and self-discharge rate ΔC in the power battery related index set [B] are set as continuous parameters DB in step A4; the driving road conditions in the vehicle parameter index set [V] are set as ordinal parameters DX in step A4; at the same time, the ambient temperature parameter Temp-1 under vehicle driving conditions, the ambient temperature parameter Temp-2 under vehicle parking conditions, and the solar radiation intensity SH are also collected, and Temp-1, Temp-2, and SH are set as ordinal parameters DX in step A4; the acceleration frequency in the vehicle parameter index set [V] is the acceleration frequency / single sampling period, and the braking frequency is the braking frequency / single sampling period.

[0085] As a preferred technical solution, the driving conditions in the vehicle parameter index set [V] are divided into four ordered data: urban congestion conditions, general urban conditions, suburban conditions, and highway conditions. In the vehicle parameter index set [V], a braking frequency > 7.73 indicates urban congestion conditions; a braking frequency ≤ 2.8 and < 7.73 indicates general urban conditions; a braking frequency ≤ 1.82 and < 2.8 indicates suburban conditions; and a braking frequency < 1.82 indicates highway conditions. The braking frequency is the average number of braking actions per kilometer.

[0086] The solar radiation intensity SH is divided into 5 ordinal data, which are classified into Class I, Class II, Class III and Class IV regions.

[0087] The first category of areas refers to regions with 3200-3300 hours of daily sunshine per year and an annual radiation dose of 1860-2330 kW·h / m². 2;

[0088] The second category of areas refers to regions with 3000-3200 hours of sunshine per year and an annual radiation of 1630-1860 kW·h / m². 2 ;

[0089] The three categories of areas are defined as areas with 2200-3000 hours of sunshine per year and annual radiation ranging from 1390-1630 kW·h / m². 2 ;

[0090] The four categories of areas are defined as areas with 1400-2200 hours of sunshine per year and annual radiation ranging from 1160-1390 kW·h / m². 2 ;

[0091] Category V areas: Annual sunshine hours of approximately 1000-1400 hours, annual radiation of 933-1160 kW·h / m² 2 .

[0092] As a preferred technical solution, the correlation level in step A7-1 is divided into the following four levels.

[0093] The data in the first ordinal parameter DX is used as the first column of data, and the calculated values ​​of all mathematical models MM in the continuous model group [MM] are used as the second column of data.

[0094] Level 1 consists of data in the first and second columns that have absolutely no relation to each other.

[0095] Level 2 shows a vague correlation between the data in the first and second columns;

[0096] Level 3 shows a strong, but not clear, functional relationship between the first and second columns of data.

[0097] Level 4 shows a clear functional relationship between the first and second columns of data.

[0098] The correlation analysis method is Spearman analysis or analysis of covariance.

[0099] The sampling period T in step A2 is set to either a certain time interval (Time) or a certain distance interval (Trip).

[0100] The sampling period T is set according to a certain mileage Trip, where the sampling period T is one of the driving mileages of 0.5 million kilometers, 1 million kilometers, 2 million kilometers, and 2.5 million kilometers.

[0101] The sampling period T is set according to a certain time, which is one of 3 months, 6 months, 8 months and 1 year.

[0102] The communication device in step A1 is an internet communication device or a mobile communication device.

[0103] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below with reference to specific embodiments.

[0104] Example 1

[0105] like Figure 1 The method for constructing evaluation indicators for pure electric vehicle power batteries based on big data analysis, as shown, is characterized by the following steps performed sequentially:

[0106] A1. Label multiple vehicles of the same type and equipped with the same power battery as P1, P2, ..., P1 respectively. n The vehicle is equipped with a communication module that uploads data to a server via the Internet, Bluetooth, or mobile communication devices.

[0107] A2. During the operation of vehicle P1, the vehicle's power battery capacity decay rate C% is collected at a certain sampling period T. Multiple C% values ​​collected at multiple sampling periods T are used as calibration samples. Simultaneously, vehicle parameter index set [V]1 and power battery parameter index set [B]1 are collected at the same sampling period T. [V]1 includes multiple vehicle-related parameters during vehicle operation, and [B]1 includes multiple parameters related to the power battery installed in the vehicle. The sampling period T is set according to a certain time or a certain mileage trip. The calibration samples, [V]1, and [B]1 are uploaded to the server.

[0108] Because the variation in C% varies significantly among different manufacturers and types of power batteries, it is advisable to collect data on the actual lifespan of some vehicle power batteries before categorizing them.

[0109] For some power batteries with a short service life, such as actual service life of less than 200,000 kilometers or less than 3 years, their C% changes faster, that is, the battery itself has a short service life. The sampling period T can be 2k-5k kilometers, or 15-30 days according to the time interval.

[0110] For power batteries with a long service life, such as an actual service life of more than 400,000 kilometers or more than 7 years, the sampling period can be 0.5, 10,000, 20,000 or 25,000 kilometers of driving mileage; the sampling period T can also be 3 months, 6 months or 8 months or 1 year.

[0111] The vehicle parameter index set [V] includes at least the vehicle's cumulative mileage L, vehicle's cumulative usage time Time1, acceleration frequency (times / single sampling period), braking frequency (times / single sampling period), and road conditions. For example, the acceleration frequency is collected every 5,000 kilometers. For pure electric vehicles, frequent acceleration will cause the power battery to be in a short-term high-current discharge state frequently. Relevant data shows that this condition has an inverse correlation with the service life of the power battery, that is, the higher the acceleration frequency, the shorter the battery service life; similarly, the higher the braking frequency, the shorter the battery service life.

[0112] In addition, relevant data shows that traffic congestion during vehicle operation can also affect the lifespan of the power battery; therefore, this method also collects relevant data on driving conditions as ordinal data.

[0113] The power battery related index set [B] includes the number of charge cycles N1, cumulative charging time Time3; open circuit voltage U1, operating voltage U2, internal resistance R1, self-discharge rate ΔC; maximum cumulative working time of a single battery module Time2; the open circuit voltage U1 and operating voltage U2 are the minimum measured values ​​of a single battery module in the power battery pack under normal operating conditions; the internal resistance R1 and self-discharge rate ΔC are the maximum measured values ​​of a single battery module in the power battery pack under normal operating conditions.

[0114] A3, Page 2...P n Repeat step A2 for each new car. When the obtained C% value is different, the corresponding multiple vehicle parameter index sets [V]2……[V] are... n And multiple sets of power battery parameter indicators [B]2……[B] n ;

[0115] A4. Regarding [V]1……[V] n The multiple parameters in the data are statistically classified into ordinal parameters (DX) or continuous parameters (DB), and [B]1...[B] n The multiple parameters in the data are statistically classified into ordinal parameters DX or continuous parameters DB.

[0116] The cumulative mileage L, cumulative vehicle usage time Time1, acceleration frequency (times / single sampling period), braking frequency (times / single sampling period), number of charge cycles N1, cumulative charging time Time3, and maximum cumulative working time of the single battery module Time2 are all included. Open circuit voltage U1, operating voltage U2, internal resistance R1, and self-discharge rate ΔC are set as continuous parameters DB in step A4. The continuous parameters DX are statistically significant ratios, for example, using a single battery module open circuit voltage of 48V as a baseline, with subsequent measurements at 48.1V, 48.1V, 48.0V, and 39.9V.

[0117] The driving conditions, ambient temperature parameter Temp-1 when the vehicle is in motion, ambient temperature parameter Temp-2 when the vehicle is parked, and solar radiation intensity SH are set as ordinal parameters DX in step A4.

[0118] Ordinal parameters DX are ordinal data in statistics. For example, the ambient temperature parameter Temp-1 when the vehicle is in motion can be divided into 5 or 7 levels according to the human body's thermal comfort perception based on the temperature range. The ambient temperature parameter Temp-2 when the vehicle is parked can be divided into 3 or 5 levels according to the size of the self-discharge rate ΔC of the power battery at different temperatures.

[0119] In one recommended classification method, the driving conditions are divided into four ordered data: urban congestion conditions, urban general conditions, suburban conditions, and highway conditions. In the vehicle parameter index set [V], when the braking frequency is >7.73, it is urban congestion conditions (average number of braking times per kilometer); when the braking frequency is 2.8 ≤ and <7.73, it is urban general conditions; when the braking frequency is 1.82 ≤ and <2.8, it is suburban conditions; and when the braking frequency is <1.82, it is highway conditions.

[0120] In one recommended classification method, the solar radiation intensity SH is divided into 5 ordinal data:

[0121] Category I areas: Annual sunshine duration of 3200-3300 hours, annual radiation of 1860-2330 kW·h / m² 2 ;

[0122] Category II areas: Annual sunshine hours of 3000-3200 hours, annual radiation of 1630-1860 kW·h / m² 2 ;

[0123] Category III areas: Annual sunshine hours of 2200-3000 hours, annual radiation of 1390-1630 kW·h / m² 2 ;

[0124] Category IV areas: Annual sunshine hours of 1400-2200 hours, annual radiation of 1160-1390 kW·h / m² 2 ;

[0125] Category V areas: Annual sunshine hours of approximately 1000-1400 hours, annual radiation of 933-1160 kW·h / m² 2 ;

[0126] For the DB parameter, since it is a continuous parameter, its value should basically conform to a normal distribution for multiple vehicles. If the data at a certain sampling point is abnormal, the data should be removed to ensure the accuracy of subsequent model construction.

[0127] Extract n parameters from n vehicles within the same sampling period T, and test the normality of these n parameters. This can be done directly using statistical software such as SPSS. If the parameters do not conform to a normal distribution, discard the sampling point and all n parameter values ​​associated with that point. If they conform to a normal distribution, retain them. When the number of retained sampling points exceeds 90% of the total number of sampling points, define the remaining data in the corresponding DB parameter as DB. ND The parameters are then preserved; at this point, the DB... ND The data in the parameters conform to a normal distribution;

[0128] A5. Remove the first DB. ND The parameters include data up to 1% and data after 99% of the normal distribution, meaning the database is retained. ND For data in the parameters that fall within the normal distribution range of 1%-99%, the satisfaction level is 98%. When the data volume is large, the satisfaction level can be set to 95% or 90%. Repeat step A5 in this manner for all remaining databases. ND Parameters are processed;

[0129] A6. Using the calibration sample as the target value, the calibration sample at this time retains the C% value of all sampling points. The first DB... ND Parameter data is used as input parameters, DB ND Some sampling point parameters are missing; a neural network model MM1 is built using a CNN convolutional neural network; based on the neural network model MM1, input DB. ND When parameters are input, the change value of the vehicle's power battery capacity decay rate C% can be obtained, thus realizing the prediction function. Similarly, the change value of C% can be input in reverse to calculate the calculated value of the MM1 model at each sampling point. The calculated value data of each sampling point calculated by the MM1 model in reverse is complete and statistically significant, so it is used in subsequent correlation calculations. This step simplifies the data values ​​of multiple vehicles into a single column of statistically significant data through the construction of a neural network model. When performing correlation analysis with the calibration sample in the future, there will be two columns of data, which can be conveniently used with multiple correlation analysis methods.

[0130] And so on, multiple DBs ND The parameters yield multiple neural network models MM2...MM n ;

[0131] A7. Using neural network models MM1...MM nThe calculated values ​​at each sampling point were used as the dependent variable, and the calibration samples were used as the independent variables; correlation analysis was used to compare the calibration samples with multiple neural network models MM1...MM. n Correlation analysis is performed on the calculated values ​​of each model; the correlation coefficient γ and significance test probability ρ% are obtained, and one or more neural network models with significance test probability ρ% < 0.05 and the largest correlation coefficient γ value are extracted to form a continuous model group [MM].

[0132] A8. Using regression analysis, establish a regression equation between the calibration sample and the calculated values ​​of a neural network model in the continuous model group [MM]; the number of regression equations established should be consistent with the number of calculated values ​​of the neural network model in [MM].

[0133] A9. Select m more vehicles, use the regression equation in step A8 to predict the value of each C% according to the sampling period, and compare the prediction results with the C% in the actual driving process of the m vehicles. The comprehensive error rate of the comparison results is ≤5% and is judged as qualified. The comprehensive error rate is the root mean square value of the error rate at all sampling points.

[0134] One or more regression equations with the lowest overall error rate are extracted from multiple regression equations and used as continuous parameter prediction models.

[0135] In the above model construction process, the data collected from each vehicle under continuous parameters are first statistically analyzed using a normal distribution to make the data statistically significant. Then, a CNN convolutional neural network model is constructed from the collected data from multiple vehicles. This neural network model is then used to output a series of calculated values, transforming the collected data from multiple vehicles into a series of statistically significant calculated values ​​with good predictive accuracy. Next, regression analysis is used in statistics to analyze the correlation between the vehicle's power battery capacity decay rate C% and the calculated values ​​of each neural network model. The regression equation with significant correlation is extracted as a predictive model for power battery performance or lifespan. Compared with traditional neural network model calculation methods, the computational load is greatly reduced, and compared with single correlation regression prediction methods, the accuracy and adaptability are better, and it has broad practical application value.

[0136] Example 2

[0137] like Figure 2 As shown, in the process of constructing traditional power battery evaluation indicators, interval data or ratio data, that is, continuous variable data, are often used for calculation. However, in reality, parameters of certain ordinal data types can also have a certain impact on the performance of power batteries. If continuous variable methods are used for calculation, the amount of calculation is very large. Therefore, this type of data is generally simplified into multiple sequences, that is, grade classification, such as external environmental factors and vehicle factors such as ambient temperature, road conditions, and sunlight intensity.

[0138] In this scheme, both continuous parameter DB and sequential parameter DX are collected. Using both parameters together to evaluate the power battery can better reflect the actual operating conditions of the power battery.

[0139] In the calculation of the continuous parameter DB, it has been correlated with the standard sample through correlation analysis; therefore, after performing correlation analysis between the ordinal parameter DX and the continuous model group [MM] that has already undergone correlation analysis, the power battery is evaluated together.

[0140] Step A7 further includes the following steps:

[0141] A7-1. Using correlation analysis, perform correlation analysis between the data in the first ordinal parameter DX and the calculated values ​​of all mathematical models MM in the continuous model group [MM].

[0142] Correlation analysis can be performed using Spearman analysis, analysis of covariance, or other methods in SPSS software that can be used to perform correlation analysis on ordinal and ratio data with high quantification.

[0143] In the results of the correlation analysis, the correlation level mentioned in step A7-1 is divided into the following four levels:

[0144] The data in the first ordinal parameter DX is used as the first column of data, and the calculated values ​​of all mathematical models MM in the continuous model group [MM] are used as the second column of data.

[0145] Level 1 consists of data in the first and second columns that have absolutely no relation to each other.

[0146] Level 2 shows a vague correlation between the data in the first and second columns;

[0147] Level 3 shows a strong, but not clear, functional relationship between the first and second columns of data.

[0148] Level 4 shows a clear functional relationship between the first and second columns of data.

[0149] A7-2. If the correlation level between the ordinal parameter DX and more than 50% of the mathematical models MM in the continuous model group [MM] is level 3 (strong correlation) or above, it indicates that the ordinal parameter DX has a significant correlation with the power battery capacity decay rate C%.

[0150] Further calculations are performed on the ordinal parameter DX:

[0151] A7-3. Calculate the percentage of each ordinal data in the corresponding DX parameter of the vehicle when the vehicle's power battery capacity decay rate C% changes from 100% to 80% under the sampling period T; the percentage Z of the first ordinal data in the first DX parameter. 1-1 %, the proportion of the second ordinal data Z 1-2 %...the proportion of the Zth ordered data Z 1-z %; for example, the vehicle's driving conditions are recorded once every 5000km, and Table 1 shows the percentage of each driving condition at 3 sampling points; the Zth one is represented by a certain meaning, which is a common mathematical representation method.

[0152] Table 1. Data Collection Form for a Vehicle's Driving Conditions Every 5000km

[0153]

[0154] Calculate the proportion of each ordered data in the first DX parameter of each of the n vehicles; form a Q×n×Z matrix; where Q is the number of sampling points;

[0155] A7-4. Using the Q×n×Z matrix as the input layer, and taking the change in battery capacity decay rate C% from 100% to 80% as the target value, a CNN convolutional neural network model is used for training to obtain a multi-layer neural network model.

[0156] A7-5. Repeat step A7-1 for the remaining ordinal parameters DX respectively; then in step A9, select m more vehicles, predict the value of each C% of the multilayer neural network model according to the sampling period, and compare the prediction results with the C% of the actual driving process of the m vehicles. If the comprehensive error rate of the comparison results is ≤5%, it is judged as qualified.

[0157] One or more multi-layer neural network models with the lowest overall error rate are extracted from multiple multi-layer neural network models and used as ordinal parameter prediction models.

[0158] In Example 2, in addition to outputting a continuous parameter prediction model, a sequential parameter prediction model is also output simultaneously, which greatly expands the applicability of this power battery index evaluation method and further improves its accuracy. Moreover, for a certain type of vehicle with a large number of vehicles in use and a relatively simple usage environment, the sequential parameter prediction model can be used directly for prediction, which greatly simplifies the prediction calculation process.

[0159] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing evaluation indexes of pure electric vehicle power batteries based on big data analysis, characterized in that, It comprises the following steps: A1. Mark several vehicles of the same type, equipped with the same power battery, as P1, P2, …, P n , which are equipped with a communication module for uploading data to a server through a communication device; A2, in the process of driving, the capacity attenuation rate C% of the vehicle power battery is collected at a certain sampling period T, and multiple C% collected at multiple sampling periods T are used as calibration samples; meanwhile, the vehicle parameter index set [V]1 and the power battery parameter index set [B]1 are collected synchronously at the sampling period T; the [V]1 comprises multiple vehicle parameters during driving, and the [B]1 comprises multiple parameters related to the power battery carried by the vehicle; the calibration samples, [V]1 and [B]1 are uploaded to the server; A3, P2 … P n The vehicle repeats step A2 to obtain a plurality of vehicle parameter indicator sets [V]2 … [V] n , and a plurality of power battery parameter indicator sets [B]2 … [B] n ; A4, to [V]1...[V] n with [B]1...[B] n statistically classifying the plurality of parameters into ordinal parameters DX or continuous parameters DB, respectively; For the parameter DB, n parameters of n vehicles at the same sampling period T are extracted, and the normal distribution of the n parameters is verified; if the normal distribution does not appear, the sampling point and the n parameter values at the sampling point are removed; If the distribution is normal, the sampling point is reserved; if the number of the reserved sampling points accounts for more than 90% of the total sampling points, the remaining data in the corresponding parameter DB is defined as DB ND parameters and reserved; A5, eliminate the first DB ND Parameters after the normal distribution of 1% before and after the data, 99% envelope data in the normal distribution of 1% to 99%; repeat A5 step, for the remaining all DB ND Parameters are processed; A6. Using the calibration sample as the target value, the first DB ND The parameter data is used as input parameters, and a neural network model MM1 is built using a CNN convolutional neural network; and so on, multiple DBs are used. ND The parameters yield multiple neural network models MM2...MM n ; A7. using the neural network model MM1...MM n The calculated values at each sampling point are taken as the dependent variable, and the calibration sample is taken as the independent variable; correlation analysis is used to analyze the correlation between the calibration sample and the calculated values of the plurality of neural network models; the correlation coefficient γ and the significance test probability ρ% are obtained, the one or more neural network models with the largest correlation coefficient γ value and the significance test probability ρ% < 0.05 are extracted, and a continuous model group [MM] is formed. A8, using regression analysis method, a regression equation of the calibration samples and the calculation value of a certain neural network model in the continuous type model group [MM] is established; the number of the regression equation established is consistent with the number of the calculation value of the neural network model in the [MM]; A9, another m vehicles are selected, the regression equation in step A8 is used to predict the value of each C% at the sampling period, and the prediction result is compared with the C% in the actual driving process of the m vehicles; if the comprehensive error rate of the comparison result is ≤5%, it is determined to be qualified; the comprehensive error rate is the root mean square value of the error rate at all sampling points; One or more regression equations with the lowest comprehensive error rate in the multiple regression equations are extracted as the continuous type parameter prediction model. 2.The method of claim 1, wherein the method is characterized in that: The vehicle parameter index set [V] comprises vehicle cumulative driving mileage L, vehicle cumulative use time Time1, acceleration frequency, braking frequency and driving road conditions; the acceleration frequency is acceleration frequency / single sampling period; the braking frequency is braking frequency / single sampling period; The power battery parameter index set [B] comprises cycle charging times N1, cumulative charging time Time3, open circuit voltage U1, working voltage U2, internal resistance R1 and self-discharge rate ΔC; the maximum cumulative working time Time2 of the single battery module; the open circuit voltage U1 and the working voltage U2 are the minimum values measured by the single battery module in the normal working state of the power battery group; the internal resistance R1 and the self-discharge rate ΔC are the maximum values measured by the single battery module in the normal working state of the power battery group. 3.The method of claim 2, wherein the method further comprises: determining a plurality of evaluation indexes of the power battery based on the big data analysis. The cumulative mileage L, the vehicle cumulative use time Time1, the acceleration frequency, the braking frequency in the vehicle parameter index set [V] and the cycle charging number N1, the cumulative charging time Time3, the maximum cumulative working time Time2 of the single battery module, the open circuit voltage U1, the working voltage U2, the internal resistance R1, the self-discharge rate ΔC in the power battery parameter index set [B] are set as continuous parameters DB in step A4; The driving road conditions and the environmental temperature parameters Temp-1, Temp-2 under the vehicle driving state, the solar radiation intensity SH in the vehicle parameter index set [V] are set as ordered parameters DX in step A4; The acceleration frequency is the acceleration frequency per single sampling period; The braking frequency is the braking frequency per single sampling period. 4.The method of claim 3, wherein the method is characterized in that: The driving road conditions in the vehicle parameter index set [V] are divided into four ordered data, which are urban congestion working condition, urban general working condition, suburban working condition and high-speed working condition, and the braking frequency > 7.73 is the urban congestion working condition; 2.8≤braking frequency is <7.73 is the urban general working condition; 1.82≤braking frequency <2.8 is the suburban working condition; braking frequency <1.82 is the high-speed working condition; The braking frequency is the average braking frequency per kilometer; The solar radiation intensity SH is divided into five ordered data, which are one-class area, two-class area, three-class area, four-class area and five-class area; The one type of region is that the annual sunshine duration is 3200-3300 hours, and the annual radiation is 1860-2330 kW·h / m 2 ; The second type of region is a region with 3000-3200 hours of annual sunshine and 1630-1860 kW·h / m of annual radiation 2 ; The three types of regions are regions with annual sunshine duration of 2200-3000 hours and annual radiation of 1390-1630 kW·h / m 2 ; The four types of regions are the annual sunshine hours of 1400-2200 hours, the annual radiation of 1160-1390 kW·h / m 2 ; Five categories of regions: the annual sunshine hours is 1000-1400 hours, the annual radiation is 933-1160 kW·h / m 2 . 5.The method of claim 1, wherein the method further comprises: determining a plurality of evaluation indexes of the power battery based on the big data analysis. The sampling period T in step A2 is set according to one of the certain time Time or the certain mileage Trip; In the sampling period T according to the certain mileage Trip, the sampling period T is one of 50,000 kilometers, 100,000 kilometers, 200,000 kilometers and 250,000 kilometers of driving mileage; In the sampling period T according to the certain time Time, the sampling period T is one of 3 months, 6 months, 8 months and 1 year.

Images