Lithium ion battery RUL prediction method and device based on data driving, and storage medium
The data-driven method using Fourier function, CEEMDAN decomposition, and LSTM model addresses complex degradation patterns and noise issues in lithium-ion batteries, achieving accurate RUL prediction and enhanced battery management.
Patent Information
- Application Number
- CN202411272586.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional battery life prediction methods, particularly for lithium-ion batteries, struggle with complex degradation patterns and nonlinear features, and are adversely affected by noise and improper data preprocessing, leading to inaccurate predictions.
A data-driven method combining Fourier function for smoothing, CEEMDAN decomposition to extract residual and modal components, and LSTM neural network model for accurate prediction of lithium-ion battery Remaining Useful Life (RUL).
The method enhances prediction accuracy by stabilizing data, extracting key degradation trends, and modeling battery capacity changes, thereby improving battery management and extending the effective use cycle.
Smart Images

Figure CN119087268A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of battery life prediction, and specifically relates to a data-driven lithium-ion battery RUL prediction method and device, and a storage medium. Background Art
[0002] In battery management systems, accurate prediction of the remaining useful life (RUL) of batteries is one of the key technologies. Lithium-ion batteries are widely used in various fields, such as electric vehicles and portable electronic devices, and their performance and safety directly affect the reliability and user experience of the equipment. Battery capacity degradation is an inevitable process, which is usually caused by multiple factors such as charge and discharge cycles, temperature changes, etc. In order to effectively manage the use and maintenance of batteries, researchers have developed a variety of prediction methods.
[0003] Traditional methods rely on statistical models and empirical formulas, but these methods often have difficulty dealing with complex degradation patterns and nonlinear features. In recent years, data-driven methods have gradually become an important tool for battery life prediction. LSTM can capture long-term and short-term dependencies in time series data, and improve prediction accuracy and system reliability through in-depth learning of battery degradation patterns. However, the accuracy of the LSTM model is greatly affected by data quality and feature selection. If there is noise in the input data, improper feature selection, or insufficient data preprocessing, it may affect the learning effect of the model and cause the model's prediction performance to deteriorate. Summary of the invention
[0004] In view of the problems in the related art, the present invention proposes a data-driven lithium-ion battery RUL prediction method and device, and a storage medium to overcome the above-mentioned technical problems existing in the existing related art.
[0005] To this end, the specific technical solution adopted by the present invention is as follows: a data-driven lithium-ion battery RUL prediction method, comprising the following steps: Step 1: Select the lithium battery capacity as the life prediction health factor, conduct a discharge characteristic test experiment on the battery to obtain the original data of the battery capacity, and draw a lithium-ion battery capacity degradation curve; Step 2: Establishing a lithium-ion battery RUL prediction equation, and using Fourier function to process the original battery capacity data to obtain a capacity sequence with a relatively stable degradation trend; Step 3: Use the CEEMDAN method to decompose and smooth the capacity sequence to obtain the residual and modal components ; Step 4: As the input value of the LSTM model, the mapping relationship between the early and late capacity is established, and the unknown capacitance sequence is predicted. The number of cycles when the capacity reaches the threshold is obtained through iterative prediction, and the battery RUL is calculated.
[0006] In a further embodiment, the specific process of step 1 is as follows: Step 1-1, placing the lithium-ion battery in a constant temperature box and setting a predetermined temperature; Step 1-2, charging the lithium-ion battery in a constant current mode until the voltage value reaches a first voltage threshold; the lithium-ion battery continues to be charged in a constant voltage mode until the charging current drops to the first current threshold, and charging is terminated; Step 1-3, discharging the lithium-ion battery in a constant current mode until the discharge current reaches a second current threshold and the voltage reaches a second voltage threshold; Step 1-4, repeat steps 1-2 and 1-3 at intervals of 0.5-1 hour until the number of cycles allowed by the battery is reached, end the experiment and draw a capacity degradation curve of the lithium-ion battery.
[0007] In a further embodiment, the process of establishing the RUL prediction equation in step 2 is as follows: First, establish the RUL observation equation: ; In the formula, Indicates the initial cycle of lithium-ion battery charge and discharge cycle. Indicates the cycle at the end of the battery life under actual charging and discharging conditions of lithium-ion batteries. Indicates that the life of lithium-ion batteries is represented by the number of cycles; The RUL prediction equation is established based on the RUL observation equation: ; In the formula, Indicates the cycle at which the lithium-ion battery reaches the end of its life under predicted conditions. Represents the predicted lifetime of lithium-ion batteries.
[0008] In a further embodiment, the Fourier function in step 2 is expressed as: ; In the formula, Indicates the number of charge and discharge cycles. The parameter values can be obtained through fitting.
[0009] In a further embodiment, an index evaluation of the Fourier function is also included, specifically including: Mean absolute error and RMS error As an evaluation index, the formula is: ; ; In the formula, Indicates the length of the time series, Indicates the actual battery capacity. represents the predicted battery capacity, Indicates the number of charge and discharge cycles.
[0010] In a further embodiment, the residual in step 3 and modal components The acquisition process is as follows: Lithium-ion battery capacity decay data Add Gaussian white noise to Get white noise capacity data ,in is the initial amplitude of white noise, Indicates The white noise added is Gaussian distributed. Indicates the number of charge and discharge cycles; Use EMD to add white noise to the battery capacity data To decompose, ; Get the first stage modal components , then Take the average to get the first , which is announced as follows: ; ; Calculate the first residual of the battery capacity data : ; Calculate the second modal component : ; Calculate the kth residual : ; For The mode is decomposed again to obtain the modal components : ; This is repeated until the final residual cannot be decomposed further. : ; In the formula, represents the battery capacity data after k times of EMD decomposition, is the total number of modal components, Reflects the local characteristics of lithium-ion battery capacity data, residual It is the remaining part after the battery capacity data is decomposed, and contains the overall characteristics of the capacity data.
[0011] In a further embodiment, the calculation process of the battery RUL in step 4 is: The residual Enter the forget gate of the LSTM model, which determines the cell state at the previous moment How many are retained to the current moment, and its output matrix The calculation formula is as follows: ; In the formula, the unit state It is dynamically generated by the characteristics of the input data and changes all the time; Refers to the sigmoid activation function. When sigmoid=1, it means to save all , when sigmoid=0, it means completely forgetting ; is the weight matrix of the forget gate; The battery life result output at the last moment; Refers to the residual battery capacity input at the current moment; is the forget gate bias matrix, which is used to adjust the switch state and output of the gating mechanism; The input gate determines the output at the previous moment And the current input How much is saved to the cell state The calculation formula is as follows: ; ; ; In the formula, represents the input gate output matrix, and Represent the weight and bias matrix of the forget gate respectively, For use A new unit state residual constructed by the function, and Respectively The weight and bias matrices of the function; The output gate determines the final output result. , and its calculation formula is as follows: ; ; In the formula, and Represent the weight and bias matrix of the output gate respectively; The output of the LSTM model Substituting into the prediction equation, we get RUL: .
[0012] A device includes: a processor and a memory; the memory stores a computer program, and the processor implements the above-mentioned lithium-ion battery RUL prediction method when executing the computer program.
[0013] A computer storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the above-mentioned lithium-ion battery RUL prediction method.
[0014] Beneficial effects of the present invention: The present invention utilizes the Fourier function smoothing method to ensure the stability of the capacity data and reduce the interference of noise on the prediction results. Secondly, the CEEMDAN method further decomposes the capacity sequence, extracts key degradation trends and modal components, and improves the accuracy of the data and the reliability of the prediction. Finally, by inputting the smoothed residual into the LSTM model, the change of battery capacity can be accurately modeled, the remaining useful life (RUL) of the battery can be accurately predicted, the battery management strategy can be optimized, the risk of failure can be reduced, and the effective use cycle of the battery can be extended, thereby significantly improving the overall reliability and economic benefits of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 1 is a capacity degradation curve of the lithium-ion battery in Example 1.
[0016] Figure 2 It is the Fourier fitting curve diagram of B0005 in Example 1.
[0017] Figure 3 This is the lithium-ion battery life prediction flow chart proposed in Example 1.
[0018] Figure 4 This is a structural diagram of the LSTM model in Example 1.
[0019] Figure 5 It is a comparison chart of the prediction results in Example 1. DETAILED DESCRIPTION
[0020] The present invention relates to the field of battery life prediction, and aims to achieve accurate prediction of lithium-ion battery life (RUL). The Fourier function can remove high-frequency noise in the data to make the data smoother and more reliable. The original experimental data is smoothed to obtain relatively stable capacity data. The CEEMDAN decomposition method is used to decompose the processed capacity sequence to extract the residual that can reflect the main trend of battery capacity degradation. and modal components reflecting local characteristics , with the residual that can reflect the main trend of battery capacity degradation The LSTM model is established as input data to establish the mapping relationship between the battery capacity in the early stage and the later stage, so as to predict the unknown battery capacity sequence. The remaining useful life (RUL) of the battery is calculated by iteratively predicting the number of cycles when the capacity reaches the preset threshold. This method combines the Fourier function, CEEMDAN decomposition and LSTM neural network model to achieve data preprocessing of lithium-ion battery capacity and accurate prediction of service life. This method provides a technical path for accurately evaluating the life of lithium-ion batteries, which helps to improve the efficiency and reliability of battery management systems.
[0021] Example 1 This embodiment provides a data-driven lithium-ion battery RUL prediction method, comprising the following steps: Step 1: The most intuitive manifestation of battery life decay is the decay of battery capacity. Lithium battery capacity is selected as a life prediction health factor. The battery is subjected to a discharge characteristic test experiment to obtain the original data of battery capacity, and a lithium-ion battery capacity degradation curve is drawn; Step 2: Establishing a lithium-ion battery RUL prediction equation, and using Fourier function to process the original battery capacity data to obtain a capacity sequence with a relatively stable degradation trend; Step 3: Use the CEEMDAN method to decompose and smooth the capacity sequence to obtain the residual and modal components ; Step 4: As the input value of the LSTM model, the mapping relationship between the early and late capacity is established, and the unknown capacitance sequence is predicted. The number of cycles when the capacity reaches the threshold is obtained through iterative prediction, and the battery RUL is calculated.
[0022] Furthermore, the specific process of step one is as follows: Step 1-1, place the lithium-ion battery in a constant temperature box and set a predetermined temperature; for example, select 18650 lithium-ion batteries numbered B0005 and B0006, place the battery in a constant temperature box, and set the temperature to 24°C.
[0023] Step 1-2, charging the lithium-ion battery in constant current mode until the voltage value reaches a first voltage threshold; the lithium-ion battery continues to be charged in constant voltage mode until the charging current drops to the first current threshold, and the charging is terminated; preferably, the constant current in this embodiment is 1.5A, the first voltage threshold is 4.2V, and the first current threshold is 20mA.
[0024] Step 1-3, discharging the lithium-ion battery in a constant current mode until the discharge current reaches a second current threshold and the voltage reaches a second voltage threshold; preferably, the second current threshold is 2A and the second voltage threshold is 2.5V.
[0025] Step 1-4: Repeat steps 1-2 and 1-3 at intervals of 0.5-1 hour until the number of cycles allowed by the battery is reached, end the experiment and draw a lithium-ion battery capacity degradation curve, such as Figure 1 shown.
[0026] The process of establishing the RUL prediction equation in step 2 is as follows: First, establish the RUL observation equation: ; In the formula, Indicates the initial cycle of lithium-ion battery charge and discharge cycle. Indicates the cycle at the end of the battery life under actual charging and discharging conditions of lithium-ion batteries. Indicates that the life of lithium-ion batteries is represented by the number of cycles; The RUL prediction equation is established based on the RUL observation equation: ; In the formula, Indicates the cycle at which the lithium-ion battery reaches the end of its life under predicted conditions. Represents the predicted lifetime of lithium-ion batteries.
[0027] The Fourier function in step 2 is expressed as: ; In the formula, Indicates the number of charge and discharge cycles. The parameter values can be obtained through fitting.
[0028] In order to further verify the Fourier function, the indicator evaluation of the Fourier function is also included, including: Mean absolute error and RMS error As an evaluation index, the formula is: ; ; In the formula, Indicates the length of the time series, Indicates the actual battery capacity. represents the predicted battery capacity, Indicates the number of charge and discharge cycles.
[0029] Combination Figure 1 For example, after calculation, the fourth-order Fourier can accurately describe the decreasing trend of battery capacity. Through the fitting toolbox, we can get The values of the parameters and the degree of fit of the B0005 and B0006 battery curves are shown in Table 1. The two curves correspond to The values are 0.02453 and 0.0318 respectively, indicating that the curve fitting effect is very good.
[0030] Table 1 shows the fitting parameters of B0005 and B0006 batteries The residual in step 3 and modal components The acquisition process is as follows: Lithium-ion battery capacity decay data Add Gaussian white noise to Get white noise capacity data ,in is the initial amplitude of white noise, Indicates The white noise added is Gaussian distributed. Indicates the number of charge and discharge cycles; Use EMD to add white noise to the battery capacity data To decompose, ; Get the first stage modal components , then Take the average to get the first , which is announced as follows: ; ; Calculate the first residual of the battery capacity data : ; Calculate the second modal component : ; Calculate the kth residual : ; For The mode is decomposed again to obtain the modal components : ; This is repeated until the final residual cannot be decomposed further. : ; In the formula, represents the battery capacity data after k times of EMD decomposition, is the total number of modal components, Reflects the local characteristics of lithium-ion battery capacity data, residual It is the remaining part after the battery capacity data is decomposed, and contains the overall characteristics of the capacity data.
[0031] Modal Components reflects the local variation characteristics of the battery capacity, and the residual represents the main degradation trend of the battery, so the residual is selected It is more accurate as the input of the LSTM model. The capacity prediction value of each cycle after the starting point is obtained through iterative prediction. When the capacity reaches the threshold, the cycle ends and the threshold is recorded. And output, and Substitute the above RUL prediction equation into the battery RUL and finally calculate the battery RUL. The calculation process of the battery RUL in step 4 is: The residual Enter the forget gate of the LSTM model, which determines the cell state at the previous moment How many are retained to the current moment, and its output matrix The calculation formula is as follows: ; In the formula, the unit state It is dynamically generated by the characteristics of the input data and changes all the time; Refers to the sigmoid activation function. When sigmoid=1, it means to save all , when sigmoid=0, it means completely forgetting ; is the weight matrix of the forget gate; The battery life result output at the last moment; Refers to the residual battery capacity input at the current moment; is the forget gate bias matrix, which is used to adjust the switch state and output of the gating mechanism; The input gate determines the output at the previous moment And the current input How much is saved to the cell state The calculation formula is as follows: ; ; ; In the formula, represents the input gate output matrix, and Represent the weight and bias matrix of the forget gate respectively, For use A new unit state residual constructed by the function, and Respectively The weight and bias matrices of the function; The output gate determines the final output result. , and its calculation formula is as follows: ; ; In the formula, and Represent the weight and bias matrix of the output gate respectively; The output of the LSTM model Substituting into the prediction equation, we get RUL: .
[0032] In summary, the present invention proposes a data-driven RUL prediction method for lithium-ion batteries for accurate prediction of the life of lithium-ion batteries. This method combines Fourier function, CEEMDAN decomposition method and LSTM neural network model on the basis of fully considering the working principle of lithium-ion batteries, and has higher accuracy than the traditional LSTM model. Among them, for the battery life prediction of B0005 battery, the accuracy is improved by 14.8% compared with the basic LSTM neural network model, and for B0006 battery, the accuracy is improved by 16.7% compared with the basic LSTM neural network model, which shows that the lithium-ion battery life prediction method proposed in the present invention has higher prediction accuracy.
[0033] The above embodiments only illustrate the data-driven RUL prediction method for lithium-ion batteries using B0005 and B0006 lithium-ion batteries as examples, but it is understood that those skilled in the art may make any changes and modifications thereto without departing from the spirit and scope of the present invention.
Claims
1. A data-driven RUL prediction method for lithium-ion batteries. It is characterized in that The following steps are involved: Step 1: Select the lithium battery capacity as the life prediction health factor, conduct a discharge characteristic test experiment on the battery to obtain the original data of the battery capacity, and draw a lithium-ion battery capacity degradation curve; Step 2: Establishing a lithium-ion battery RUL prediction equation, and using Fourier function to process the original battery capacity data to obtain a capacity sequence with a relatively stable degradation trend; Step 3: Use the CEEMDAN method to decompose and smooth the capacity sequence to obtain the residual and modal components ; Step 4: As the input value of the LSTM model, the mapping relationship between the early and late capacity is established, and the unknown capacitance sequence is predicted. The number of cycles when the capacity reaches the threshold is obtained through iterative prediction, and the battery RUL is calculated.
2. A data-driven lithium-ion battery RUL prediction method according to claim 1, It is characterized in that The specific process of step one is as follows: Step 1-1, placing the lithium-ion battery in a constant temperature box and setting a predetermined temperature; Step 1-2, charging the lithium-ion battery in a constant current mode until the voltage value reaches a first voltage threshold; the lithium-ion battery continues to be charged in a constant voltage mode until the charging current drops to the first current threshold, and charging is terminated; Step 1-3, discharging the lithium-ion battery in a constant current mode until the discharge current reaches a second current threshold and the voltage reaches a second voltage threshold; Step 1-4, repeat steps 1-2 and 1-3 at intervals of 0.5-1 hour until the number of cycles allowed by the battery is reached, end the experiment and draw a capacity degradation curve of the lithium-ion battery.
3. A data-driven lithium-ion battery RUL prediction method according to claim 1, It is characterized in that The establishment process of the RUL prediction equation in step 2 is as follows: First, establish the RUL observation equation: ; In the formula, Indicates the initial cycle of lithium-ion battery charge and discharge cycle. Indicates the cycle at the end of the battery life under actual charging and discharging conditions of lithium-ion batteries. Indicates that the life of lithium-ion batteries is represented by the number of cycles; The RUL prediction equation is established based on the RUL observation equation: ; In the formula, Indicates the cycle at which the lithium-ion battery reaches the end of its life under predicted conditions. Represents the predicted lifetime of lithium-ion batteries.
4. A data-driven lithium-ion battery RUL prediction method according to claim 1, It is characterized in that The Fourier function in step 2 is expressed as: ; In the formula, Indicates the number of charge and discharge cycles. The parameter values can be obtained through fitting.
5. A data-driven lithium-ion battery RUL prediction method according to claim 4, It is characterized in that It also includes the evaluation of the Fourier function indicators, including: Mean absolute error and RMS error As an evaluation index, the formula is: ; ; In the formula, Indicates the length of the time series, Indicates the actual battery capacity. represents the predicted battery capacity, Indicates the number of charge and discharge cycles.
6. A data-driven lithium-ion battery RUL prediction method according to claim 1, It is characterized in that The residual in step 3 and modal components The acquisition process is as follows: Lithium-ion battery capacity decay data Add Gaussian white noise to Get white noise capacity data ,in is the initial amplitude of white noise, Indicates The white noise added is Gaussian distributed. Indicates the number of charge and discharge cycles; Use EMD to add white noise to the battery capacity data To decompose, ; Get the first stage modal components , then Take the average to get the first , which is announced as follows: ; ; Calculate the first residual of the battery capacity data : ; Calculate the second modal component : ; Calculate the kth residual : ; For The mode is decomposed again to obtain the modal components : ; This is repeated until the final residual cannot be decomposed further. : ; In the formula, represents the battery capacity data after k times of EMD decomposition, is the total number of modal components, Reflects the local characteristics of lithium-ion battery capacity data, residual It is the remaining part after the battery capacity data is decomposed, and contains the overall characteristics of the capacity data.
7. A data-driven lithium-ion battery RUL prediction method according to claim 1, It is characterized in that The calculation process of the battery RUL in step 4 is: The residual Enter the forget gate of the LSTM model, which determines the cell state at the previous moment How many are retained to the current moment, and its output matrix The calculation formula is as follows: ; In the formula, the unit state It is dynamically generated by the characteristics of the input data and changes all the time; Refers to the sigmoid activation function. When sigmoid=1, it means to save all , when sigmoid=0, it means completely forgetting ; is the weight matrix of the forget gate; The battery life result output at the last moment; Refers to the residual battery capacity input at the current moment; is the forget gate bias matrix, which is used to adjust the switch state and output of the gating mechanism; The input gate determines the output at the previous moment And the current input How much is saved to the cell state The calculation formula is as follows: ; ; ; In the formula, represents the input gate output matrix, and Represent the weight and bias matrix of the forget gate respectively, For use A new unit state residual constructed by the function, and Respectively The weight and bias matrices of the function; The output gate determines the final output result. , and its calculation formula is as follows: ; ; In the formula, and Represent the weight and bias matrix of the output gate respectively; The output of the LSTM model Substituting into the prediction equation, we get RUL: 。 8. A device, include: Processor and memory; The memory stores a computer program, and when the processor executes the computer program, the lithium-ion battery RUL prediction method according to any one of claims 1 to 7 is implemented.
9. A computer storage medium, It is characterized in that The computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the lithium-ion battery RUL prediction method according to any one of claims 1 to 7.
Citation Information
Cited By
Method and system for predicting residual life of high-power lithium battery pack
CN120722215A
