An artificial intelligence-based lithium ion battery electrode potential curve reconstruction, lithium extraction diagnosis and early warning method

Abstract

A lithium ion battery electrode potential curve reconstruction, lithium precipitation diagnosis and early warning method based on artificial intelligence belongs to the technical field of lithium ion battery management. The method uses the charging curve of a three-electrode battery, combines artificial intelligence methods to establish the mapping relationship between the full battery voltage curve and the single electrode potential curve, and the full battery voltage curve of the current time interval and the full battery voltage curve of the next same time interval offline. Further, in the actual operation of the two-electrode battery without reference electrode, the single electrode potential curve can be reconstructed through the full battery voltage curve, and the single electrode potential curve of the future same time period is obtained, the online diagnosis and early warning of battery lithium precipitation are realized, and the early intervention is carried out based on the warning result to prevent lithium precipitation.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery management technology, and in particular relates to an artificial intelligence-based method for reconstructing lithium-ion battery electrode potential curves, diagnosing lithium plating, and providing early warning. Background Technology

[0002] Lithium-ion batteries have been successfully commercialized after decades of effort and are widely used in various portable electronic products. However, lithium-ion batteries commonly experience gradual performance degradation during actual use, one of the main reasons being lithium plating at the negative electrode. The lithium deposited at the negative electrode not only reacts with the electrolyte but also gradually develops into dead lithium, ultimately leading to irreversible loss of active lithium and battery capacity. Furthermore, lithium dendrites formed during plating can penetrate the separator, causing internal short circuits and serious safety issues. Considering the negative impacts of lithium plating, developing reliable detection methods is crucial. Lithium plating is caused by a negative electrode potential below 0V; therefore, a three-electrode battery can be formed by introducing a reference electrode to monitor the negative electrode potential and detect lithium plating. However, for practical battery applications, only the full cell voltage, i.e., the potential difference between the positive and negative electrodes, can be measured, and the negative electrode potential cannot be monitored in real time. Therefore, if the single-electrode potential curve can be reconstructed based on the obtained full-cell voltage through necessary means, and the single-electrode potential curve for the next period can be predicted, thereby enabling the detection and early warning of lithium plating, it will play an important role in improving the safety and reliability of lithium-ion batteries and is of great significance in this field. Summary of the Invention

[0003] The purpose of this invention is to solve the problem of lithium plating detection and early warning in lithium-ion batteries, and to provide an artificial intelligence-based method for reconstructing lithium-ion battery electrode potential curves, diagnosing lithium plating, and providing early warning.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A method for reconstructing electrode potential curves, diagnosing lithium plating, and providing early warning of lithium-ion battery defects based on artificial intelligence, comprising the following steps:

[0006] Step 1: Obtain the complete full-cell voltage and single-electrode potential charging curves of the three-electrode battery under different charging methods as training data;

[0007] Step 2: Divide the acquired voltage data into two datasets: Dataset 1: the full cell voltage during the charging process and its corresponding single electrode potential; Dataset 2: the full cell voltage at a fixed time interval during the charging process and its corresponding full cell voltage at the next equal time interval.

[0008] Step 3: Randomly divide the dataset 1 obtained in Step 2 into a training set, a validation set, and a test set in a ratio of 7:1:2. Input the dataset into the selected artificial intelligence algorithm for training and establish the mapping relationship between the full battery voltage and the single electrode potential.

[0009] Step 4: Randomly divide the dataset 2 obtained in Step 2 into a training set, a validation set, and a test set in a ratio of 7:1:2. Input the selected artificial intelligence algorithm for training to establish a mapping relationship between the full battery voltage of a fixed time interval and the full battery voltage of the next equal time interval.

[0010] Step 5: Apply the trained artificial intelligence algorithm in Steps 3 and 4 online. Input the charging segment data during the actual battery operation into the artificial intelligence algorithm described in Step 3 to reconstruct the single electrode potential curve corresponding to the current full battery voltage curve and determine whether lithium plating has occurred.

[0011] Step 6: Input the charging segment data during the actual battery operation into the artificial intelligence algorithm described in Step 4 to predict the full battery voltage curve for the next equal time period.

[0012] Step 7: Input the full-cell voltage curve predicted in Step 6 into the artificial intelligence algorithm described in Step 3, and output the single-electrode potential curve corresponding to the next time period to provide early warning of lithium plating.

[0013] Step 8: Based on the lithium plating warning results obtained in Step 7, if the warning result indicates that lithium plating will occur, intervene in advance by controlling external environmental conditions to prevent lithium plating from happening; if the warning result indicates that lithium plating will not occur, no intervention is taken.

[0014] Furthermore, in step one, the different charging methods refer to one or more of the following: different charging rates, different temperatures, different operating conditions, or different aging states.

[0015] Furthermore, step two specifically includes: determining a time interval segment length, and progressively sliding the segment length on the charging curve, thereby dividing the charging curve in step one into several data segments of equal length for subsequent training.

[0016] Furthermore, the time interval segment length can be different lengths such as 60 seconds or 120 seconds.

[0017] Furthermore, in steps three and four, the artificial intelligence algorithm is one of a deep neural network, a convolutional neural network, a recurrent neural network, or an attention mechanism.

[0018] The advantages of this invention compared to existing technologies are as follows: This invention considers the importance of lithium plating detection and early warning for battery management systems, as well as the characteristics of three-electrode batteries that can be used to determine lithium plating and their inapplicability in practical scenarios. Compared to other methods, this invention focuses on two issues: electrode potential curve reconstruction and lithium plating diagnosis and early warning. It uses the charging curve of a three-electrode battery, combined with artificial intelligence methods, to offline establish the relationship between the full-cell voltage curve and the single-electrode potential curve, and between the full-cell voltage curve of the current time interval and the full-cell voltage curve of the next equal time interval. In two-electrode batteries without a reference electrode, the single-electrode potential curve can be reconstructed from the full-cell charging curve. Furthermore, by predicting the full-cell voltage curve of the next time interval, the single-electrode potential curve for the next time interval can be obtained. Simultaneously, online diagnosis and early warning of lithium plating are achieved, and timely intervention is implemented based on the warning results to prevent lithium plating and further improve the safety and reliability of lithium-ion batteries. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the method provided by the present invention;

[0020] Figure 2 This is an example diagram of the reconstructed positive electrode potential curve based on the present invention;

[0021] Figure 3 This is an example diagram of the reconstructed negative electrode potential curve based on the present invention;

[0022] Figure 4 This is an example diagram of full-cell voltage curve prediction based on the present invention;

[0023] Figure 5 This is a positive electrode potential curve obtained by reconstructing the full-cell voltage prediction curve based on the present invention.

[0024] Figure 6 This is a negative electrode potential curve obtained by reconstructing the full-cell voltage prediction curve based on the present invention.

[0025] Figure 7 This is a diagram showing the confusion between the test results of lithium-plated batteries and normal batteries based on the present invention. Detailed Implementation

[0026] The above is merely an overview of the technical solution of the present invention. To better understand the technical means of the present invention, the technical solution will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] This invention utilizes the charging curve of a three-electrode battery and combines artificial intelligence methods to establish offline mapping relationships between the full-cell voltage curve and the single-electrode potential curve, as well as between the full-cell voltage curve at the current time interval and the full-cell voltage curve at the next same time interval. Furthermore, in the actual operation of a two-electrode battery without a reference electrode, the single-electrode potential curve can be reconstructed from the full-cell voltage curve, and the single-electrode potential curve for the same time interval in the future can be obtained. Simultaneously, online diagnosis and early warning of lithium plating in the battery can be achieved, and intervention can be carried out based on the warning results to prevent lithium plating from occurring.

[0028] Example 1:

[0029] An artificial intelligence-based method for reconstructing lithium-ion battery electrode potential curves and providing early warning of lithium plating, as shown in the attached figure. Figure 1 As shown, it specifically includes the following parts:

[0030] Step 1: Obtain the complete full-cell voltage and single-electrode potential charging curves of the three-electrode battery under different charging methods as training data. Specifically, this includes charging schemes with different charging rates, temperatures, operating conditions, and aging states to obtain the complete full-cell voltage and single-electrode potential charging curves of the three-electrode battery, including signals such as the battery charging rate (e.g., 2C), temperature (e.g., 10℃), and other special environmental conditions under a given charging scheme.

[0031] Step 2: Divide the acquired voltage data into two datasets: Dataset 1: the full cell voltage during charging and its corresponding single electrode potential; Dataset 2: the full cell voltage at a fixed time interval during charging and its corresponding full cell voltage at the next equal time interval. Specifically, this includes: determining a time interval segment length, sliding the segment length progressively on the charging curve, thereby dividing the charging curve in Step 1 into data segments of a certain length, such as 120s, for subsequent training.

[0032] Step 3: Use artificial intelligence algorithms to establish the mapping relationship between the full cell voltage and the single electrode potential. Specifically, this includes: selecting a deep neural network algorithm, dividing the dataset 1 obtained in Step 2 into a training set, a validation set, and a test set in a ratio of 7:1:2, inputting the selected deep neural network algorithm for training, and obtaining the reconstructed single electrode potential curve.

[0033] Step 4: Use artificial intelligence algorithms to establish a mapping relationship between the full battery voltage at a fixed time interval and the full battery voltage at the next equal time interval. Specifically, this includes: selecting a deep neural network algorithm, dividing the dataset 2 obtained in Step 2 into a training set, a validation set, and a test set in a 7:1:2 ratio, inputting the selected deep neural network algorithm for training, and obtaining the predicted full battery voltage curve for the same future time period.

[0034] Step 5: In actual battery application, the charging segment data during actual battery operation is input into the artificial intelligence algorithm described in Step 3 to reconstruct the single-electrode potential curve corresponding to the current full battery voltage curve. Based on the current single-electrode potential value, it is determined whether lithium plating has occurred. Specifically, this includes: during actual battery operation, the collected charging segment data during operation is input into the deep neural network algorithm trained in Step 3 to obtain the single-electrode potential curve corresponding to the current full battery voltage curve. In this embodiment, the charging curve of a three-electrode battery charged at 0.1C charging rate at 25℃ is used as an example, and the single-electrode potential curve is reconstructed using a deep neural network algorithm. (Appendix) Figure 2 and 3 The diagram shows a comparison between the electrode potential curve reconstructed based on the present invention and the actual curve.

[0035] Step Six: During actual battery use, input the charging segments from the actual battery operation into the artificial intelligence algorithm described in Step Four to predict the full battery voltage curve for the next equal time period. Specifically, this includes: during actual battery operation, inputting the collected charging segments from the operation process into the deep neural network algorithm trained in Step Four to obtain the full battery voltage curve for the next time period corresponding to the current full battery voltage curve. In this embodiment, the charging curve of a three-electrode battery charged at 0.1C charging rate at 25℃ is used as an example, and the deep neural network algorithm is used to predict the full battery voltage curve for the next time period. (Appendix) Figure 4 An example of full-cell voltage curve prediction based on the present invention is shown.

[0036] Step 7: Input the full-cell voltage curve predicted in Step 6 into the artificial intelligence algorithm described in Step 3, and output the single-electrode potential curve corresponding to the next time period to provide early warning of lithium plating. Specifically, this includes: inputting the full-cell voltage curve predicted in Step 6 into the deep neural network algorithm trained in Step 3 to obtain the single-electrode potential curve corresponding to the full-cell voltage curve for the same time period in the future, and realizing early warning of lithium plating based on the single-electrode potential value at the future time. (Appendix) Figure 5 and 6 An example of an electrode potential curve reconstructed from the full-cell voltage prediction curve based on the present invention is shown. The aforementioned lithium plating detection and early warning method classifies samples with an anode potential greater than 0 as normal batteries and samples with an anode potential less than 0 as lithium plating batteries. To verify the effectiveness of the proposed method, it was validated using measured data. With a data length of 120 seconds as a judgment interval, in a total of 1,158,023 samples, the accuracy rates for classifying normal batteries and lithium plating batteries were 98.8% and 84%, respectively, achieving an average accuracy of 97.4% across the entire test set. (Appendix) Figure 7 The accuracy of lithium plating detection based on the present invention is shown.

[0037] Step 8: Based on the lithium plating warning results obtained in Step 7, when the warning result indicates that lithium plating will occur, intervene in advance by controlling external environmental conditions to prevent lithium plating from occurring; when the warning result indicates that lithium plating will not occur, no intervention is taken. Specifically, based on the lithium plating warning results obtained in Step 7, when the warning result indicates that lithium plating will occur, intervene in advance by controlling external environmental conditions such as increasing temperature, reducing charging rate, or using pulse charging to prevent lithium plating from occurring; when the warning result indicates that lithium plating will not occur, no intervention is taken.

[0038] It should be understood that the sequence number of each step in the embodiments of the present invention does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for reconstructing electrode potential curves, diagnosing lithium plating, and providing early warning of lithium-ion battery electrochemical plating based on artificial intelligence, characterized in that: The method steps are as follows: Step 1: Obtain the complete full-cell voltage and single-electrode potential charging curves of the three-electrode battery under different charging methods as training data; Step 2: Divide the acquired voltage data into two datasets, namely Dataset 1: the full battery voltage during the charging process and its corresponding single electrode potential; Dataset 2: Full cell voltage at a fixed time interval during charging and its corresponding full cell voltage at the next equal time interval; Step 2 specifically includes: determining a time interval segment length, sliding the segment length progressively on the charging curve, thereby dividing the charging curve in Step 1 into several data segments of equal length for subsequent training; Step 3: Randomly divide the dataset 1 obtained in Step 2 into a training set, a validation set, and a test set in a ratio of 7:1:

2. Input the dataset into the selected artificial intelligence algorithm for training and establish the mapping relationship between the full cell voltage and the single electrode potential. Step 4: Randomly divide the dataset 2 obtained in Step 2 into a training set, a validation set, and a test set in a ratio of 7:1:

2. Input the selected artificial intelligence algorithm for training to establish a mapping relationship between the full battery voltage of a fixed time interval and the full battery voltage of the next equal time interval. Step 5: Apply the trained artificial intelligence algorithm in Steps 3 and 4 online. Input the charging segment data during the actual battery operation into the artificial intelligence algorithm described in Step 3 to reconstruct the single electrode potential curve corresponding to the current full battery voltage curve and determine whether lithium plating has occurred. Step 6: Input the charging segment data during the actual battery operation into the artificial intelligence algorithm described in Step 4 to predict the full battery voltage curve for the next equal time period. Step 7: Input the full-cell voltage curve predicted in Step 6 into the artificial intelligence algorithm described in Step 3, and output the single-electrode potential curve corresponding to the next time period to provide early warning of lithium plating. Step 8: Based on the lithium plating warning results obtained in Step 7, if the warning result indicates that lithium plating will occur, intervene in advance by controlling external environmental conditions to prevent lithium plating from happening; if the warning result indicates that lithium plating will not occur, no intervention is taken.

2. The method for reconstructing lithium-ion battery electrode potential curves, diagnosing lithium plating, and providing early warning based on artificial intelligence, as described in claim 1, is characterized in that: In step one, the different charging methods refer to one or more of the following: different charging rates, different temperatures, or different aging conditions.

3. The method for reconstructing lithium-ion battery electrode potential curves, diagnosing lithium plating, and providing early warning based on artificial intelligence, as described in claim 1, is characterized in that: The time interval segment is 60 seconds or 120 seconds in length.

4. The method for reconstructing lithium-ion battery electrode potential curves, diagnosing lithium plating, and providing early warning based on artificial intelligence, as described in claim 1, is characterized in that: In steps three and four, the artificial intelligence algorithm is a deep neural network, a convolutional neural network, a recurrent neural network, or an attention mechanism.

Images