Lithium battery short circuit early warning method based on deep learning

Abstract

The invention belongs to the technical field of lithium ion battery fault diagnosis, and discloses a lithium battery short circuit early warning method based on deep learning, comprising the following steps: activating a battery through a charge-discharge cycle test; constructing an internal short-circuit fault data set caused by overcharging; performing data cleaning and data normalization operation on the data set for training a deep learning model; carrying out actual short-circuit time labeling on the processed data according to a data label sequence acquired by a sensor; dividing the data set into a training set and a test set; performing regression analysis on the data set by adopting a neural network; and testing the lithium ion battery under the mixed working condition, and finely adjusting the trained deep learning model by using a small amount of data to obtain the detection method suitable for the actual complex working condition. By adopting the method, the time of possible internal short circuit of the single battery can be accurately and quickly predicted, accurate fault information is provided for a battery management system, and the method has safety and reliability.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion battery fault diagnosis, and in particular to a lithium battery short circuit early warning method based on deep learning. Background Art

[0002] In recent years, lithium-ion batteries have become a hot research direction in the field of energy storage applications due to their high energy density. In order to achieve real-time monitoring and precise control of the charge and discharge status of lithium-ion batteries, they are usually equipped with a battery management system (BMS). However, due to the complexity and harshness of its working environment, under overcharge or low-temperature charging conditions, slow lithium dendrite growth may occur inside the battery, leading to serious safety hazards. In particular, in many actual working conditions, lithium-ion batteries need to perform continuous charge and discharge operations. The internal short circuit caused by lithium dendrites will not only significantly reduce the battery performance, but may also pose a threat to the personal safety of users.

[0003] In the prior art, most early warning monitoring methods for lithium short circuits rely on violent disassembly of batteries, X-ray imaging or nuclear magnetic resonance. However, these methods are difficult to implement on batteries in operation, and the corresponding detection conditions are not available in most usage scenarios. At present, the early warning research on non-destructive lithium short circuits in lithium-ion batteries is still insufficient, especially in practical applications, conventional detection methods can only collect battery parameters such as current, voltage and temperature through sensors for status diagnosis. In terms of research methods, existing methods are mostly based on electrochemical models to achieve lithium short circuit warning. However, due to the complexity of electrochemical models and the limitation of computational complexity, it is difficult to achieve effective transplantation on hardware platforms under actual working conditions. In addition, short circuits often occur instantaneously, and existing algorithms are difficult to achieve accurate early warning of lithium short circuits.

[0004] Therefore, there is an urgent need for a lithium battery short circuit warning method based on deep learning to improve the safety and practicality of lithium-ion batteries. Summary of the invention

[0005] The purpose of the present invention is to provide a lithium battery short circuit warning method based on deep learning. By using the voltage, current and temperature characteristics of the battery and limited sensor data, it can accurately and quickly predict the time when an internal short circuit may occur in a battery cell, and provide accurate fault information for the battery management system (BMS), thereby realizing corresponding fault-tolerant control and significantly improving the safety and reliability of the BMS.

[0006] To achieve the above object, the present invention provides a lithium battery short circuit warning method based on deep learning, comprising the following steps:

[0007] Step S1, activating the battery through a charge and discharge cycle test;

[0008] Step S2, constructing an internal short circuit fault data set caused by overcharging;

[0009] Step S3, performing data cleaning and data normalization operations on the internal short circuit fault data set for training a deep learning model;

[0010] Step S4, marking the actual short-circuit time of the processed data according to the data label sequence collected by the sensor;

[0011] Step S5, dividing the data set into a training set and a test set;

[0012] Step S6: using a neural network to perform regression analysis on the data set;

[0013] Step S7: Test the lithium-ion battery under mixed working conditions, and fine-tune the trained deep learning model with a small amount of data to obtain a detection method suitable for actual complex working conditions.

[0014] Preferably, in step S1, the battery is activated through a charge and discharge cycle test according to the operating mode of the lithium-ion battery and its operating environment, so that the state of the lithium-ion battery is close to the actual operating condition.

[0015] Preferably, in step S2, under the actual use condition of the lithium-ion battery, an overcharging experiment is induced to trigger the lithium precipitation process inside the battery, and the current, voltage and temperature characteristics of the battery are collected to construct an internal short circuit fault data set caused by overcharging.

[0016] Preferably, in step S3, data cleaning and data normalization operations are performed on the internal short circuit fault data set for training the deep learning model. The specific process is as follows:

[0017] Step S31, use the 3 times standard deviation method to clean the data and detect and process the outliers;

[0018] Among them, data points that deviate from the normal range due to sensor failure are removed; the missing data points are filled using the linear interpolation method;

[0019] Step S32: Based on step S31, the maximum and minimum normalization method is used to normalize the internal short circuit fault data set, as shown below:

[0020]

[0021] Among them, X is the value of a state component of the current sample; X min is the minimum value of the state component of all samples; X max is the maximum value of this state component of all samples.

[0022] Preferably, in step S4, the current, voltage and temperature data of the constant current charging condition in the internal short circuit fault data set are extracted, and the time when the internal short circuit is actually about to occur at each sampling data point is calculated as follows:

[0023]

[0024] Among them, T real is the actual time when the short circuit occurs in the current battery state; N is the sampling sequence label in the current battery state; N start is the sampling sequence label at the beginning of charging, and f refers to the sampling frequency.

[0025] Preferably, in step S5, the training set and the test set account for 70% and 30% of the total data respectively; wherein the distribution of the actual short-circuit time in different sets should be the same.

[0026] Preferably, in step S6, a neural network is used to perform regression analysis on the data set, and the specific process is as follows:

[0027] Step S61, the input data of the deep learning neural network is the current, voltage and temperature data of the battery, and the output data is the predicted time of short circuit;

[0028] Step S62, a neural network is constructed using 5 fully connected layers, wherein ReLU is used as the activation function between each fully connected layer, and the dimensions from the input layer to the output layer are 3, 128, 64, 32, 16, and 1 respectively. The network uses Adam as the optimizer, and MSE is used to measure the difference between the predicted failure time and the actual failure time. The learning rate is set to 0.003;

[0029] Step S63: Use the constructed neural network to perform regression analysis on the data set.

[0030] Preferably, in step S7, a mixed working condition test is performed based on step S1, and the specific process is as follows:

[0031] Step S71, first, constant current charging, the charging current is set to 500mA, charging until the lithium ion battery voltage is 4.0V, and standing for 1 minute;

[0032] Step S72, constant current discharge, the discharge current is set to 500mA, and the lithium ion battery voltage is discharged to 3.5V, step S74, stand for 1 minute;

[0033] Step S73, looping step S71 and step S72 three times;

[0034] Step S74, constant current charging, the charging current is set to 500mA, and the lithium-ion battery is charged until the voltage is 4.9V.

[0035] Preferably, data processing is performed on the verification data, and the neural network model is fine-tuned using mixed operating condition test data, wherein 10% of the data is used for fine-tuning training of the model and the remaining 90% of the data is used for model verification.

[0036] Therefore, the present invention adopts the above-mentioned lithium battery short circuit warning method based on deep learning, and the beneficial effects are as follows:

[0037] (1) The present invention makes full use of the data-driven method and has the advantage of not needing to build an accurate model, and can accurately predict the occurrence time of internal short circuit caused by lithium deposition in lithium-ion battery cells.

[0038] (2) Different from traditional detection methods such as violent disassembly, X-ray imaging or nuclear magnetic resonance, the present invention realizes fault warning by monitoring the external characteristics of the battery (such as current, voltage and temperature), which significantly reduces the cost of fault warning.

[0039] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of a lithium battery short circuit warning method based on deep learning of the present invention;

[0041] Figure 2 Schematic diagram of experimental topology in an embodiment of the present invention;

[0042] Figure 3 Schematic diagram of a deep learning model in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0044] like Figure 1 As shown, a lithium battery short circuit warning method based on deep learning includes the following steps:

[0045] Step S1, activating the battery through a charge and discharge cycle test;

[0046] Step S2, constructing an internal short circuit fault data set caused by overcharging;

[0047] Step S3, performing data cleaning and data normalization operations on the internal short circuit fault data set for training a deep learning model;

[0048] Step S4, marking the actual short-circuit time of the processed data according to the data label sequence collected by the sensor;

[0049] Step S5, dividing the data set into a training set and a test set;

[0050] Step S6: using a neural network to perform regression analysis on the data set;

[0051] Step S7: Test the lithium-ion battery under mixed working conditions, and fine-tune the trained deep learning model with a small amount of data to obtain a detection method suitable for actual complex working conditions.

[0052] Example

[0053] Step S1: Activate the battery through charge and discharge cycle test according to the working mode and operating environment of the lithium-ion battery, so that the state of the lithium-ion battery is close to the actual working condition.

[0054] Step S2: Under the actual use conditions of the lithium-ion battery, an overcharging experiment is induced to trigger the lithium deposition process inside the battery, and the current, voltage and temperature characteristics of the battery are collected to construct an internal short circuit fault data set caused by overcharging.

[0055] Step S3: performing data cleaning and data normalization operations on the internal short circuit fault data set for training a deep learning model.

[0056] Step S31, use the 3 times standard deviation method to clean the data, detect and process the outliers; remove the data points that deviate from the normal range due to sensor failure; use the linear interpolation method to fill in the missing data points.

[0057] Step S32: Based on step S31, the maximum and minimum normalization method is used to normalize the internal short circuit fault data set, as shown below:

[0058]

[0059] Among them, X is the value of a state component of the current sample; X min is the minimum value of the state component of all samples; X max is the maximum value of this state component of all samples.

[0060] Step S4: annotate the actual short-circuit time of the data processed in step S3 according to the order of data labels collected by the sensor.

[0061] The current, voltage, and temperature data of the constant current charging condition in the internal short circuit fault data set are extracted, and the time when the internal short circuit is actually about to occur at each sampling data point is calculated, as shown below:

[0062]

[0063] Among them, Treal is the actual time when the short circuit occurs in the current battery state; N is the sampling sequence label in the current battery state; N start is the sampling sequence label at the beginning of charging, and f refers to the sampling frequency.

[0064] Step S5, dividing the data set into a training set and a test set, requiring the training set and the test set to account for 70% and 30% of the total data respectively; wherein the distribution of the actual short-circuit time in different sets should be the same.

[0065] Step S6: Use a neural network to perform regression analysis on the data set.

[0066] Step S61: The input data of the deep learning neural network is the current, voltage, and temperature data of the battery, and the output data is the predicted time of short circuit.

[0067] Step S62, a neural network is built using 5 fully connected layers, where ReLU is used as the activation function between each fully connected layer, and the dimensions from the input layer to the output layer are 3, 128, 64, 32, 16, and 1 respectively. The network uses Adam as the optimizer, and MSE is selected to measure the difference between the predicted failure time and the actual failure time. The learning rate is set to 0.003.

[0068] Step S63: Use the constructed neural network to perform regression analysis on the data set, such as Figure 3 shown.

[0069] Step S7, testing the lithium-ion battery under mixed working conditions, and fine-tuning the deep learning model trained in step S3 with a small amount of data, so as to verify the robustness of the short-circuit warning method and obtain a detection method suitable for actual complex working conditions.

[0070] like Figure 2 As shown, a battery experimental platform is built, and the mixed working condition test is carried out on the basis of step S1. The specific process is as follows:

[0071] Step S71, first, constant current charging, the charging current is set to 500mA, charging until the lithium-ion battery voltage is 4.0V, and standing for 1 minute.

[0072] Step S72, constant current discharge, the discharge current is set to 500mA, and the lithium-ion battery voltage is discharged to 3.5V. Step S74, stand for 1 minute.

[0073] Step S73: loop step S71 and step S72 three times.

[0074] Step S74, constant current charging, the charging current is set to 500mA, and the lithium-ion battery is charged until the voltage is 4.9V.

[0075] The verification data is subjected to the data processing step of step S3, and the neural network model is fine-tuned using the mixed working condition test data, wherein 10% of the data is used for fine-tuning training of the model, and the remaining 90% of the data is used for model verification.

[0076] Therefore, the present invention adopts the above-mentioned lithium battery short-circuit warning method based on deep learning. Through the voltage, current and temperature characteristics of the battery, and using limited sensor data, the advantage of the data-driven method that does not require the construction of an accurate model is fully utilized. It can accurately and quickly predict the time when an internal short circuit may occur in the battery cell, and provide accurate fault information for the battery management system (BMS), thereby realizing corresponding fault-tolerant control and significantly improving the safety and reliability of the BMS. Unlike traditional detection methods such as violent disassembly, X-ray imaging or nuclear magnetic resonance, the present invention realizes fault warning by monitoring the external characteristics of the battery (such as current, voltage and temperature), which significantly reduces the cost of fault warning.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A lithium battery short circuit warning method based on deep learning, characterized in that: The following steps are involved: Step S1, activating the battery through a charge and discharge cycle test; Step S2, constructing an internal short circuit fault data set caused by overcharging; Step S3, performing data cleaning and data normalization operations on the internal short circuit fault data set for training a deep learning model; Step S4, marking the actual short-circuit time of the processed data according to the data label sequence collected by the sensor; Step S5, dividing the data set into a training set and a test set; Step S6: using a neural network to perform regression analysis on the data set; Step S7: Test the lithium-ion battery under mixed working conditions, and fine-tune the trained deep learning model with a small amount of data to obtain a detection method suitable for actual complex working conditions.

2. A lithium battery short circuit warning method based on deep learning according to claim 1, characterized in that: In step S1, according to the operating mode and operating environment of the lithium-ion battery, the battery is activated through a charge and discharge cycle test to make the lithium-ion battery state close to the actual operating condition.

3. A lithium battery short circuit warning method based on deep learning according to claim 1, characterized in that: In step S2, under the actual use condition of the lithium-ion battery, an overcharging experiment is induced to trigger the lithium precipitation process inside the battery, and the current, voltage and temperature characteristics of the battery are collected to construct an internal short circuit fault data set caused by overcharging.

4. A lithium battery short circuit warning method based on deep learning according to claim 1, characterized in that: In step S3, the internal short circuit fault data set is cleaned and normalized for training the deep learning model. The specific process is as follows: Step S31, use the 3 times standard deviation method to clean the data and detect and process the outliers; Among them, data points that deviate from the normal range due to sensor failure are removed; the missing data points are filled using the linear interpolation method; Step S32: Based on step S31, the maximum and minimum normalization method is used to normalize the internal short circuit fault data set, as shown below: Among them, X is the value of a state component of the current sample; X min is the minimum value of the state component of all samples; X max is the maximum value of this state component of all samples.

5. A lithium battery short circuit warning method based on deep learning according to claim 1, characterized in that: In step S4, the current, voltage and temperature data of the constant current charging condition in the internal short circuit fault data set are extracted, and the time when the internal short circuit is actually about to occur at each sampling data point is calculated as follows: Among them, T real is the actual time when the short circuit occurs in the current battery state; N is the sampling sequence label in the current battery state; N start is the sampling sequence label at the beginning of charging, and f refers to the sampling frequency.

6. A lithium battery short circuit warning method based on deep learning according to claim 1, characterized in that: In step S5, the training set and the test set account for 70% and 30% of the total data respectively; wherein the distribution of the actual short-circuit time in different sets should be the same.

7. A lithium battery short circuit warning method based on deep learning according to claim 1, characterized in that: In step S6, a neural network is used to perform regression analysis on the data set. The specific process is as follows: Step S61, the input data of the deep learning neural network is the current, voltage and temperature data of the battery, and the output data is the predicted time of short circuit occurrence; Step S62, a neural network is constructed using 5 fully connected layers, wherein ReLU is used as the activation function between each fully connected layer, and the dimensions from the input layer to the output layer are 3, 128, 64, 32, 16, and 1 respectively. The network uses Adam as the optimizer, and MSE is used to measure the difference between the predicted failure time and the actual failure time. The learning rate is set to 0.003; Step S63: Use the constructed neural network to perform regression analysis on the data set.

8. A lithium battery short circuit warning method based on deep learning according to claim 1, characterized in that: In step S7, a mixed working condition test is performed based on step S1, and the specific process is as follows: Step S71, first, constant current charging, the charging current is set to 500mA, charging until the lithium ion battery voltage is 4.0V, and standing for 1 minute; Step S72, constant current discharge, the discharge current is set to 500mA, and the lithium ion battery voltage is discharged to 3.5V, step S74, stand for 1 minute; Step S73, looping step S71 and step S72 three times; Step S74, constant current charging, the charging current is set to 500mA, and the lithium-ion battery is charged until the voltage is 4.9V.

9. A lithium battery short circuit warning method based on deep learning according to claim 8, characterized in that: The verification data was processed and the neural network model was fine-tuned using mixed working condition test data, where 10% of the data was used for fine-tuning training of the model and the remaining 90% of the data was used for model verification.

Images