Power battery full life cycle state evaluation method based on big data

Abstract

The invention discloses a power battery full life cycle state evaluation method based on big data, and relates to the technical field of power battery health management, and the method comprises the following steps: 1, collecting data; 2, data preprocessing; step 3, carrying out full life cycle attenuation model training; 4, optimizing the model; 5, performing state evaluation and residual life prediction; and 6, establishing a dynamic updating mechanism. According to the power battery full life cycle state evaluation method based on the big data, the SOH evaluation error is smaller than or equal to 3%, the residual life prediction error is smaller than or equal to 150 cycles (improved by 60% compared with a traditional method), key attenuation factors can be identified, for example, 70% attenuation of a certain battery is from high temperature exposure, personalized maintenance suggestions are provided, and the power battery full life cycle state evaluation method based on the big data is suitable for popularization and application. A user can take targeted measures to prolong the service life of the battery, large-scale management is supported, 100,000 + batteries can be evaluated in parallel, the cloud computing delay is smaller than 10 seconds, and the large-scale application requirement is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power battery health management, and particularly relates to a power battery full life cycle state evaluation method based on big data. BACKGROUND

[0002] With the rapid development of the new energy automobile industry, the full life cycle management of power batteries has become a key problem. The full life cycle (from factory to scrap) state evaluation of power batteries needs to integrate factory parameters, charging and discharging records, fault history and other multi-dimensional data.

[0003] The existing technology has the following problems:

[0004] The existing method has the following defects: 1. It relies on detection data at a single time point (such as the capacity of a certain charge), which cannot reflect the decay trend, and the evaluation error is > 10%; 2. It does not consider environmental factors (such as 20% faster battery decay in high-temperature areas) and usage habits (such as 30% shorter battery life for high fast charging proportion), resulting in a prediction deviation of > 500 cycles; 3. It lacks a dynamic updating mechanism, and the model accuracy cannot be continuously optimized as the number of battery samples increases. SUMMARY

[0005] The present application provides a power battery full life cycle state evaluation method based on big data to solve the problems raised in the background.

[0006] To solve the above technical problems, the technical solution adopted by the present application is:

[0007] A power battery full life cycle state evaluation method based on big data, comprising the following steps, characterized by:

[0008] Step 1: Collect data;

[0009] Step 2: Data preprocessing;

[0010] Step 3: Full life cycle decay model training

[0011] Step 4: Model optimization;

[0012] Step 5: State evaluation and remaining life prediction;

[0013] Step 6: Establish a dynamic updating mechanism.

[0014] Preferably, the step 1 includes factory parameters (capacity, material, production date), real-time charging and discharging data (voltage, current, temperature, SOC, sampling frequency 1Hz), environmental data (average annual temperature and humidity in the area), user behavior data (fast charging proportion, discharge depth), fault records (alarm code, maintenance history);

[0015] Factory basic parameters: In addition to capacity, material (ternary lithium needs to mark the nickel cobalt manganese ratio, such as NCM811; iron lithium phosphate needs to mark the purity of the positive material), production date, add battery cell number, cell size specification (thickness / width / height), initial internal resistance, factory quality inspection report number (correlation with quality inspection data traceability), voltage platform range (such as 3.0-4.2V);

[0016] Real-time charge and discharge data: On the basis of voltage, current, temperature (add cell tab temperature and battery pack environment temperature dual-dimension collection), SOC (State of Charge), supplement charge and discharge cycle stage identification (such as constant current charging, constant voltage charging, constant current discharging), charging start / stop voltage, discharging start / stop voltage, single battery voltage balance degree (to avoid the influence of single difference on overall evaluation), sampling frequency 1Hz remains unchanged, while adding millisecond level peak current / voltage capture (to deal with instantaneous working conditions such as sudden acceleration and sudden deceleration);

[0017] Environmental data: In addition to the average annual temperature and humidity in the area, supplement extreme environmental data (such as annual cumulative time of extreme high temperature ≥50℃, extreme low temperature ≤-20℃), altitude (affecting battery electrolyte activity), environmental pollutant concentration (such as salt concentration in coastal areas to prevent battery shell corrosion), sunshine duration (related to the sunning of vehicle-mounted battery);

[0018] User behavior data: On the basis of fast charging ratio and discharge depth (DOD, Depth of Discharge), refine discharge depth classification (shallow discharge: DOD<30%; medium discharge: 30%≤DOD≤70%; deep discharge: DOD>70%), charging and discharging interval time (such as daily charging frequency, single discharge duration), long-term standing time (such as the number of times the vehicle is idle for more than 7 days), high-power discharge frequency (such as the number of times the discharge power is greater than or equal to 80% of the rated power for more than 10 seconds);

[0019] Fault and maintenance data: In addition to alarm code and maintenance history, supplement working condition information when the fault occurs (such as charging / mid-discharge / idle), fault duration, maintenance parts (such as replacing the cell / BMS module), calibration data after maintenance (such as SOC calibration value, internal resistance retest value), disassembly and detection data of retired batteries (such as electrode structure loss degree, electrolyte deterioration condition).

[0020] Preferably, the step two comprises: removing noise by Kalman filtering, filling missing values by interpolation method, and standardizing data format (unifying unit and sampling frequency).

[0021] Preferably, the step three comprises:

[0022] S1: Extract key features: cycle number, cumulative charge and discharge capacity, high-temperature exposure time (> 40°C cumulative duration), fast charging frequency ratio, maximum discharge depth;

[0023] Decay rate features: capacity decay amount per unit time (e.g., daily decay, monthly decay), cycle decay coefficient (capacity decay rate per 100 cycles);

[0024] Health-related features: battery internal resistance growth rate ((current internal resistance - initial internal resistance) / initial internal resistance), voltage platform drop (difference between current voltage platform and initial voltage platform), charge and discharge efficiency (discharge capacity to charge capacity ratio);

[0025] Environmental impact features: low-temperature exposure time (< 0°C cumulative duration), temperature and humidity comprehensive impact coefficient (weighted coefficient constructed by combining temperature and humidity, such as high-temperature high-humidity environment coefficient higher than high-temperature low-humidity);

[0026] Feature selection: use Pearson correlation coefficient (select features with correlation > 0.7 with capacity decay rate) and recursive feature elimination (RFE) algorithm to eliminate redundant features (e.g., "cumulative charge and discharge capacity" and "cycle number" have strong correlation, keep the former which has more information), finally determine 8-10 core features for model training, improve model training efficiency;

[0027] S2: Model training: use gradient boosting tree (GBDT) algorithm, with "capacity decay rate = (nominal capacity - current capacity) / nominal capacity" as the target variable, input the above features, train the model with 5000+ sets of battery full life cycle data (goodness of fit R 2 > 0.95);

[0028] Algorithm details: in gradient boosting tree (GBDT) algorithm, set the number of decision trees to 200-300 (to avoid overfitting), learning rate to 0.05-0.1, and maximum tree depth to 5-8 layers; use cross-validation (5-fold cross-validation) to divide the training set and validation set (training set accounts for 70%, validation set accounts for 30%), to reduce the risk of model overfitting; when calculating the target variable "capacity decay rate", supplement the detection standard of "current capacity" (e.g., use 0.2C rate charge and discharge test to obtain actual capacity, to ensure calculation accuracy).

[0029] Data size and quality assurance: 5000+ sets of battery full life cycle data cover different brands, different models (e.g., square, cylindrical, soft package), and the proportion of retired batteries is not less than 30% (to ensure the model's ability to assess the end-of-life state); data time span covers 3-5 years (including complete seasonal cycles and battery decay periods), with goodness of fit R 2On the basis of >0.95, the newly added root mean square error (RMSE) <2%, and the mean absolute error (MAE) <1.5% precision indicators comprehensively measure the model fitting effect

[0030] Preferably, the fourth step comprises training sub-models for different battery types (ternary lithium / lithium iron phosphate) and different use scenarios (private cars / taxis) to improve the evaluation pertinence.

[0031] Subdivided by battery type:

[0032] Ternary lithium battery sub-model: for high energy density characteristics, the weights of “high temperature exposure time” and “fast charging frequency ratio” are optimized (ternary lithium has poor high temperature stability, and these features have a greater impact on degradation);

[0033] Lithium iron phosphate battery sub-model: for high safety and long cycle life characteristics, the “low temperature performance degradation coefficient” feature is added (lithium iron phosphate has more obvious low temperature capacity degradation), and the weight of “cycle number” is adjusted (to adapt to its long cycle life of more than 1500 times).

[0034] Subdivided by use scenario:

[0035] Private car scenario sub-model: considering the user's variable use conditions (such as city commuting and long-distance travel alternation), the “condition complexity coefficient” is added (constructed in combination with driving speed fluctuations and parking times), and the feature weight of “long-term static duration” is optimized (private cars are commonly idle);

[0036] Taxi scenario sub-model: for the characteristics of high frequency charging and discharging (2-3 times of charging per day) and long driving mileage (300-500 km per day), the influence of “cumulative charging and discharging amount” and “fast charging ratio” is highlighted, and the training data uses actual operation data of taxis (to ensure that the model adapts to high-intensity use scenarios);

[0037] Energy storage scenario sub-model: to adapt to the working conditions of “long-term floating charge and intermittent discharge” of energy storage batteries, the features of “floating charge duration ratio” and “discharge power fluctuation amplitude” are added, and in addition to the capacity degradation rate, the “charging and discharging efficiency degradation rate” is supplemented (energy storage scenarios have higher requirements for efficiency).

[0038] Preferably, the fifth step comprises:

[0039] S1: Real-time evaluation: input the latest data of the battery to be evaluated into the model, and output the current SOH (health) and capacity degradation rate;

[0040] S2: remaining life prediction: based on the current attenuation rate and user habits, the long short-term memory network (LSTM) is used to predict the future attenuation trend, and the remaining cycle life (error <200 cycles) and the recommended maintenance measures (such as "reduce fast charging, avoid full charging and full discharging") are output. Preferably, the step six includes: updating the model parameters every 3 months with newly collected battery data (actual life of retired batteries) to ensure that the evaluation accuracy improves as the amount of data increases (annual error reduction <5%).

[0041] Compared with the prior art, the present application has the following beneficial effects:

[0042] 1. The present application provides a power battery full life cycle state evaluation method based on big data, with SOH evaluation error ≤3%, and remaining life prediction error ≤150 cycles (60% higher than traditional methods).

[0043] 2. The present application provides a power battery full life cycle state evaluation method based on big data, which can identify key attenuation factors, such as that 70% of the attenuation of a certain battery is caused by high temperature exposure, and provide personalized maintenance suggestions, which helps users to take targeted measures to prolong the life of the battery.

[0044] 3. The present application provides a power battery full life cycle state evaluation method based on big data, which supports large-scale management and can evaluate 100,000+ batteries in parallel, with cloud computing delay <10 seconds, meeting the needs of large-scale applications.

[0045] 4. The present application provides a power battery full life cycle state evaluation method based on big data, which provides a basis for battery step utilization (such as SOH >80% of the battery can be used for energy storage), and improves resource utilization by 20%, with good economic and environmental benefits. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The figure is a schematic diagram of the evaluation process of the present application;

[0047] Figure 2 The figure is a schematic diagram of the full life cycle attenuation curve prediction of the present application. DETAILED DESCRIPTION

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

[0049] Example 1:

[0050] A power battery full life cycle state evaluation method based on big data, comprising the following steps:

[0051] Full life cycle attenuation model training is performed:

[0052] Extract the number of cycles, cumulative charge and discharge capacity, high temperature exposure time, and the proportion of fast charging times from the pretreated data; use the GBDT algorithm, take the capacity attenuation rate as the target variable, train the model using a large amount of full life cycle data, and train a sub-model for private car ternary lithium type;

[0053] Implement state evaluation and remaining life prediction:

[0054] Input the latest data of the battery into the GBDT sub-model (private car ternary lithium model), the model outputs the current SOH 91% (capacity attenuation rate 9%); based on the current attenuation rate and user usage habits, the remaining cycle life is predicted to be about 1200 times (according to the current usage habits) by LSTM; the analysis shows that "high temperature exposure contributes 40% of the attenuation", and it is suggested that "avoid direct sunlight when parking in summer, and use sunshades".

[0055] Establish a dynamic updating mechanism:

[0056] Update the data after 6 months (the number of cycles increases to 920, and the measured SOH is 89%), the model error is 2%, and the model parameters are updated using the new data to improve the prediction accuracy of subsequent predictions.

[0057] Example 2:

[0058] Establish the latest GBDT sub-model, the model outputs the current SOH 50% (capacity attenuation rate 10%), based on the current attenuation rate and user usage habits, the remaining cycle life is predicted to be about 1500 times (according to the current usage habits) by LSTM, the analysis shows that "high temperature and fast charging factors contribute 50% to the battery attenuation rate".

[0059] The working principle of the present application is as follows: first, rely on the mutual cooperation of various data to establish corresponding different sub-models, according to the working environment and driving habits of the driver, and then rely on the results given by the model to analyze which specific behavior of the driver has a greater impact on the battery attenuation, and then analyze and propose subsequent suggestions, analyze the remaining life, and provide personalized maintenance suggestions to meet the demand of large-scale application, which has good economic and environmental benefits.

[0060] The above shows and describes the basic principles and main features of the present application and the advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the full life-cycle state of a power battery based on big data, comprising the following steps, characterized in that: Step 1: Collect data; Step 2: Data preprocessing; Step 3: Train the full lifecycle decay model Step 4: Model optimization; Step 5: Conduct a condition assessment and predict remaining useful life; Step Six: Establish a dynamic update mechanism.

2. The method for assessing the full life cycle status of a power battery based on big data as described in claim 1, characterized in that: Step one includes: Factory parameters (capacity, materials, production date), real-time charging and discharging data (voltage, current, temperature, SOC, sampling frequency 1Hz), environmental data (average annual temperature and humidity of the area), user behavior data (fast charging ratio, depth of discharge), and fault records (alarm codes, repair history). Basic factory parameters: In addition to capacity, materials (ternary lithium batteries need to indicate the nickel-cobalt-manganese ratio, such as NCM811; lithium iron phosphate batteries need to indicate the purity of the cathode material), and production date, the following are added: battery cell number, cell size specifications (thickness / width / height), initial internal resistance, factory quality inspection report number (related to quality inspection data traceability), and voltage platform range (such as 3.0-4.2V). Real-time charge and discharge data: In addition to voltage, current, temperature (with the addition of dual-dimensional acquisition of cell tab temperature and battery pack ambient temperature), and SOC (State of Charge), the data includes charge and discharge cycle stage identifiers (such as constant current charging, constant voltage charging, and constant current discharging), charging start / stop voltage, discharging start / stop voltage, and individual cell voltage balance (to avoid affecting the overall evaluation due to individual cell differences). The sampling frequency remains unchanged at 1Hz, and millisecond-level peak current / voltage capture has been added (to handle instantaneous conditions such as rapid acceleration and deceleration). Environmental data: In addition to the average annual temperature and humidity of the region, supplement with extreme environmental data (such as the cumulative annual duration of extreme high temperature ≥50℃ and extreme low temperature ≤-20℃), altitude (affecting the activity of battery electrolyte), concentration of environmental pollutants (such as salt concentration in coastal areas to prevent battery casing corrosion), and sunshine duration (related to the sun exposure of vehicle batteries). User behavior data: Based on fast charging ratio and depth of discharge (DOD), further refine the depth of discharge classification (shallow discharge: DOD < 30%; medium discharge: 30% ≤ DOD ≤ 70%; deep discharge: DOD > 70%), charge and discharge interval time (such as average number of daily charges, duration of single discharge), long-term idle time (such as the number of times the vehicle has been idle for more than 7 days), and high power discharge frequency (such as the number of times the discharge power is ≥ 80% of the rated power for more than 10 seconds). Fault and maintenance data: In addition to alarm codes and maintenance history, supplement the operating conditions when the fault occurred (e.g., charging / discharging / resting), fault duration, repaired parts (e.g., replacing cells / battery management system BMS module), calibration data after maintenance (e.g., SOC calibration value, internal resistance retest value), and disassembly and testing data of retired batteries (e.g., degree of electrode structure wear, electrolyte deterioration).

3. The method for assessing the full life cycle status of a power battery based on big data according to claim 1, characterized in that: Step two includes: Noise is removed using Kalman filtering, missing values ​​are filled using interpolation, and the data format is standardized (unified units and sampling frequency).

4. The method for assessing the full life cycle status of a power battery based on big data according to claim 1, characterized in that: Step three includes: S1: Extract key features: number of cycles, cumulative charge and discharge capacity, high temperature exposure time (cumulative duration > 40℃), percentage of fast charge cycles, and maximum depth of discharge; Attenuation rate characteristics: capacity decay per unit time (e.g., daily decay, monthly decay), cycle decay coefficient (capacity decay rate per 100 cycles). Health-related characteristics: battery internal resistance growth rate ((current internal resistance - initial internal resistance) / initial internal resistance), voltage plateau drop (difference between current voltage plateau and initial voltage plateau), charge / discharge efficiency (ratio of discharge to charge). Environmental impact characteristics: Low temperature exposure time (cumulative duration < 0℃), temperature and humidity comprehensive impact coefficient (a weighted coefficient constructed by combining temperature and humidity, such as the high temperature and high humidity environment coefficient is higher than the high temperature and low humidity environment coefficient); Feature selection: Pearson correlation coefficient (to select features with a correlation > 0.7 with capacity decay rate) and recursive feature elimination (RFE) algorithm are used to remove redundant features (such as "cumulative charge and discharge" and "number of cycles" have a strong correlation, retaining the former which has more information) and finally determine 8-10 core features for model training to improve model training efficiency. S2: Model Training: The Gradient Boosting Tree (GBDT) algorithm is used, with "Capacity decay rate = (Nominal capacity - Current capacity) / Nominal capacity" as the target variable. The above feature is input, and the model is trained using full lifecycle data from over 5000 battery packs (goodness of fit R²). 2 >0.95); Algorithm details: In the Gradient Boosting Tree (GBDT) algorithm, the number of decision trees is set to 200-300 (to avoid overfitting), the learning rate is 0.05-0.1, and the maximum tree depth is 5-8 layers; cross-validation (5-fold cross-validation) is used to divide the training set and validation set (70% training set and 30% validation set) to reduce the risk of model overfitting; when calculating the target variable "capacity decay rate", a detection standard for "current capacity" is added (such as using a 0.2C rate charge and discharge test to obtain the actual capacity to ensure calculation accuracy). Data Scale and Quality Assurance: Data from over 5000 battery lifecycle datasets covers different brands and models (e.g., prismatic, cylindrical, pouch cells), with retired battery data accounting for at least 30% (ensuring the model's ability to assess end-of-life conditions); the data spans 3-5 years (including complete seasonal cycles and battery degradation cycles), with a goodness-of-fit R-value. 2 Based on a value of >0.95, new accuracy indicators have been added: root mean square error (RMSE) <2% and mean absolute error (MAE) <1.5% to comprehensively measure the model's fitting performance.

5. The method for assessing the full life cycle status of a power battery based on big data according to claim 1, characterized in that: Step four includes: Train sub-models for different battery types (ternary lithium / lithium iron phosphate) and different usage scenarios (private cars / taxi) to improve the targeting of evaluation; Breakdown by battery type: Ternary lithium battery sub-model: For high energy density characteristics, the focus is on optimizing the weights of "high temperature exposure time" and "fast charging cycle ratio" (ternary lithium batteries have poor high temperature stability, and these characteristics have a greater impact on degradation). Lithium iron phosphate battery sub-model: In order to achieve high safety and long cycle life, a new feature of "low temperature performance degradation coefficient" has been added (the capacity degradation of lithium iron phosphate is more obvious at low temperatures), and the weight of "number of cycles" has been adjusted (to adapt to its long cycle life of more than 1,500 cycles). Segmented by usage scenario: Private car scenario sub-model: Considering the changing user conditions (such as alternating urban commuting and long-distance travel), a new "condition complexity coefficient" (constructed by combining driving speed fluctuations and number of stops) is added, and the feature weight of "long-term idle time" is optimized (private cars are often idle). Taxi scenario sub-model: Targeting the characteristics of high-frequency charging and discharging (2-3 times per day) and long driving mileage (300-500km per day), the model focuses on strengthening the impact of "cumulative charging and discharging amount" and "fast charging ratio", and the training data uses actual taxi operation data (to ensure that the model is adapted to high-intensity usage scenarios). Energy storage scenario sub-model: Adapted to the working conditions of "long-term float charging and intermittent discharge" of energy storage batteries, the new features of "float charging time ratio" and "discharge power fluctuation range" are added. In addition to the capacity decay rate, the target variable is supplemented with "charge and discharge efficiency decay rate" (energy storage scenarios have high efficiency requirements).

6. The method for assessing the full life cycle status of a power battery based on big data according to claim 1, characterized in that: Step five includes: S1: Real-time evaluation: Input the latest data of the battery to be evaluated into the model and output the current SOH (health status) and capacity decay rate; S2: Remaining Lifetime Prediction: Based on the current degradation rate and user habits, a Long Short-Term Memory (LSTM) network is used to predict the future degradation trend, outputting the remaining cycle life (error < 200 cycles) and suggested maintenance measures (such as "reduce fast charging and avoid full charging and discharging").

7. The method for assessing the full life cycle status of a power battery based on big data according to claim 1, characterized in that: Step six includes: The model parameters are updated every 3 months with newly collected battery data (the actual lifespan of retired batteries) to ensure that the assessment accuracy improves as the amount of data increases (annual error reduction of <5%).

Images