A lithium battery RUL prediction method and system based on area characteristics and data drive
Through the combination of triangular area characteristics in DTV curve and BILSTM model, the problems of poor deep learning interpretability and difficulty in internal resistance measurement in RUL prediction of lithium batteries are solved, the prediction accuracy and model interpretability are improved, and the simple and efficient battery is achieved Lifespan assessment.
Patent Information
- Application Number
- CN202311639816.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2043-11-30
AI Technical Summary
Among the existing lithium battery RUL prediction methods, deep learning methods have poor interpretability, limited information on traditional charge and discharge curves, and difficult internal resistance measurement and high cost, resulting in low prediction accuracy.
The triangular area feature extraction method based on DTV is adopted, combined with the BILSTM model and the dot product attention mechanism, and the triangular area in the DTV curve is used as input to train the model to improve the prediction accuracy.
It improves the accuracy and generalization of RUL prediction of lithium batteries, enhances the interpretability of the model, simplifies the calculation process, is suitable for combining with BMS systems, and has online application potential.
Smart Images

Figure CN118050640B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of lithium battery technology, and in particular to a lithium battery RUL prediction method and system based on area characteristics and data drive. Background Art
[0002] Lithium-ion batteries are widely used in automobiles, electronic devices, and spacecraft due to their high energy density, wide operating range, strong temperature adaptability, and long cycle life. As one of the safety indicators for lithium batteries, the remaining life has become a research hotspot in this field in recent years. Reliable and accurate RUL (Remaining Useful Life) prediction is of great significance to ensure the performance, safety, and economy of battery-powered systems.
[0003] In the study of RUL of lithium-ion batteries, the health indicator (HI) is usually used to reflect the degradation state of the battery. The accuracy of battery RUL prediction is affected by the quality of HI extraction. The more accurate the HI extraction, the higher the RUL prediction accuracy. During the degradation process of lithium-ion batteries, many researchers use capacity or internal resistance as indicators to reflect the degradation of lithium-ion batteries. However, in practical applications, the measurement process of lithium-ion battery internal resistance is difficult and the measurement cost is high. Therefore, many studies have overcome the disadvantages of direct HI by extracting indirect HI to characterize the degree of degradation of lithium batteries. It is easy to measure or calculate, and the extraction is relatively simple.
[0004] At present, the estimation methods for battery RUL can be roughly divided into model-based methods, data-driven methods and fusion methods. Since the deep learning method avoids the complex mathematical calculations of feature physical modeling, it has been widely used, but its interpretability is poor and it is highly dependent on the quality of input features. At the same time, the traditional charge and discharge curve contains limited information and low sensitivity. Therefore, it is necessary to use variable features that are strongly related to battery degradation to train and test the model and establish a multi-model fusion estimation method that combines the advantages of different methods.
[0005] Differential Thermal Voltammetry (DTV), as a characteristic signal analysis method, can establish the connection between macroscopic signal characteristics and microscopic battery degradation characteristics. At the microscopic level, battery degradation can be characterized by the type and degree of internal phase change of the positive electrode material; at the macroscopic level, external signal characteristics can be obtained and measured to observe phase change characteristics, that is, changes in battery surface temperature can correspond to the degree of phase change, and changes in voltage can correspond to the type of phase change of the electrode material.
[0006] The DTV method can be used to mine micro-degradation features from macro-signal features, making the deep learning method interpretable. However, in the face of a large number of features that can be extracted by signal analysis technology, selecting the most relevant features is another key. Therefore, the present invention proposes a method for extracting triangular area features based on DTV. At the same time, considering that the traditional LSTM model uses a cyclic method to average the internal hidden features when modeling the sequence without focusing on some target-related features, in order to improve the RUL prediction accuracy of lithium batteries, the present invention combines the advantages of BILSTM and dot product attention, introduces dot product attention in the BiLSTM network, and captures features with high similarity to the target through attention to extract key information. Summary of the invention
[0007] In view of the technical problems existing in the prior art, the present invention provides a lithium battery RUL prediction method and system based on area characteristics and data driving, which can efficiently and accurately know the remaining service life of the lithium battery.
[0008] According to a first aspect of the present invention, the present invention provides a lithium battery RUL prediction method based on area characteristics and data driving, comprising:
[0009] S1. Obtaining voltage data and temperature data charged at a set rate, calculating the ratio of temperature and voltage changes to obtain a DTV curve, and preprocessing the DTV curve using a denoising and smoothing method;
[0010] S2. Select three feature points in the smooth DTV curve, including two peak points and one valley point, and calculate the area S of the triangle connected by the three feature points;
[0011] S3. Establish a BILSTM prediction model based on dot product attention, use the area S as input to train the model, and obtain the maximum remaining capacity prediction value of the battery;
[0012] S4. Calculate the RUL value of the battery using the obtained maximum remaining capacity prediction value.
[0013] Based on the above technical solution, the present invention can also make the following improvements.
[0014] Optionally, step S1 includes the following steps:
[0015] S11, charging the lithium battery at a set charging rate, obtaining corresponding voltage data and temperature data, and calculating the ratio of temperature to voltage change to obtain a DTV curve;
[0016] S12. Preprocess the DTV curve using a denoising and smoothing method.
[0017] Optionally, the step S11 includes the following steps:
[0018] S111, charging the lithium battery under constant current and constant voltage conditions, and obtaining voltage data and temperature data by appropriately adjusting the sampling interval;
[0019] S112, calculating the ratio of temperature and voltage change by the following formula to obtain a DTV curve:
[0020]
[0021] Wherein, T(k) represents the temperature value of the kth sampling, V(k) represents the terminal voltage value of the kth sampling, and N represents the sampling interval.
[0022] Optionally, in step S12, a GF denoising and smoothing method is used to pre-process the DTV curve; wherein the GF algorithm is:
[0023]
[0024] Among them, σ represents the standard deviation, x represents the DTV data value before smoothing filtering, and μ represents the average value. Finally, the DTV curve after filtering is obtained.
[0025] Optionally, step S2 includes the following steps:
[0026] S21. According to the obtained smooth DTV curve, two peak points are selected, which are recorded as point A and point B, and a trough point is selected, which is recorded as point C;
[0027] S22. Calculate the area of the triangle formed by the three points according to the following formula, denoted as S ABC ;
[0028] L=(AB+BC+AC) / 2
[0029]
[0030] Among them, L represents the half side length of the triangle, AB represents the distance between point A and point B, and the same applies to BC and AC.
[0031] Optionally, step S3 includes the following steps:
[0032] S31. Build a prediction model, including an input layer, a BILSTM layer, a Dropout layer, a dot product attention layer, and an FC output layer;
[0033] S32, after preprocessing the area S data, divide it into a training set and a test set, use a sliding window to generate training sequence data from the training set, use the corresponding maximum remaining capacity as a training label, and then train the model;
[0034] S33. Use the trained prediction model to predict the lithium battery test data, and use RMSE and MAE to evaluate the prediction results of the model.
[0035] Optionally, the dot product attention layer in step S31 specifically includes the following steps:
[0036] S311. Compare the output matrix Y of the BILSTM network with the final hidden state h n Do the dot product to get vector e, which represents Y and h n The degree of relationship between them is shown in the following formula:
[0037] e i =Dot(Y i ,h n )
[0038] Where Dot(·) represents the dot product, Y i represents the output vectors that make up the matrix Y;
[0039] S312, the obtained e vector is passed through the softmax function to obtain the attention weight α, as shown in the following formula:
[0040] α=softmax(e);
[0041] S313. Perform a dot product of the attention weight and the output matrix Y to obtain the final output value.
[0042] Optionally, the use of RMSE and MAE to evaluate the prediction results of the model includes:
[0043] Output the remaining maximum capacity C of the battery through the prediction model max , and then get the corresponding charge and discharge times The prediction performance of the model is evaluated using RMSE and MAE scores:
[0044]
[0045]
[0046] Among them, n is the length of the sequence, m is the length of the sample generated by the training sequence, and y i Indicates the actual maximum battery capacity value, Indicates the predicted maximum battery capacity value.
[0047] Optionally, the RUL calculation formula of the battery is:
[0048] RUL=n eol -n s
[0049] Among them, n eol Indicates the number of charge and discharge cycles when the remaining capacity reaches the retirement threshold, n s Indicates the number of charge and discharge cycles when prediction starts.
[0050] According to a second aspect of the present invention, a lithium battery RUL prediction system based on area characteristics and data driving is provided, comprising:
[0051] The data acquisition module is used to obtain the voltage data and temperature data of the charging at the set rate, calculate the ratio of the temperature and voltage changes to obtain the DTV curve, and pre-process the DTV curve using a denoising and smoothing method;
[0052] The regional triangle area calculation module is used to select three feature points in the smooth DTV curve, including two peak points and one valley point, and calculate the area S of the triangle connected by the three feature points;
[0053] The data-driven model building module is used to build a BILSTM prediction model based on dot product attention, and the area S is used as the input training model to obtain the maximum remaining capacity prediction value of the battery;
[0054] The battery life prediction module is used to calculate the RUL value of the battery using the obtained maximum remaining capacity prediction value.
[0055] Technical effects and advantages of the present invention:
[0056] The present invention provides a lithium battery RUL prediction method and system based on area features and data-driven. On the basis of the existing DTV method, the concept of regional triangular area is introduced, and the regional area S is extracted as a health factor, so as to establish the connection between microphase change and macro signal in the battery degradation process, and improve the interpretability of the model. At the same time, by introducing dot product attention in the BILSTM network, the model's ability to capture strongly correlated features is enhanced, and the prediction accuracy and generalization are improved, so as to detect and evaluate the RUL of the battery. The method of the present invention is simple in process, does not require a large amount of calculation, can be combined with BMS, and has the potential for online application. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 It is a flow chart of a lithium battery RUL prediction method based on area characteristics and data driving provided in an embodiment of the present invention;
[0058] Figure 2 is a DTV curve graph obtained by using a GF filtering algorithm provided by an embodiment of the present invention;
[0059] Figure 3 is a schematic diagram of a BILSTM model based on dot product attention provided by an embodiment of the present invention;
[0060] Figure 4 Schematic diagram of the dot product attention mechanism provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0061] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0062] This embodiment tests lithium-ion batteries in different health states in simulated electric vehicles. The Oxford University public lithium battery data set is used, where the rated capacity of the battery is 0.74Ah, the retirement threshold is defined as 80% of the rated capacity, and 1C and 1 / 20C charge and discharge cycles are used. The RUL prediction method of lithium batteries based on area characteristics and data-driven is used to model the RUL of lithium-ion batteries.
[0063] It is understandable that, based on the defects in the background technology, the embodiment of the present invention proposes a lithium battery RUL prediction method based on area characteristics and data drive, as shown in the attached Figure 1 As shown, the method comprises the following steps:
[0064] S1. Obtaining voltage data and temperature data charged at a set rate, calculating the ratio of temperature and voltage changes to obtain a DTV curve, and preprocessing the DTV curve using a denoising and smoothing method;
[0065] Specifically, in this embodiment, step S1 specifically includes the following steps:
[0066] S11, charging the lithium battery at a set charging rate, obtaining corresponding voltage data and temperature data, and calculating the ratio of temperature to voltage change to obtain a DTV curve;
[0067] S12. Preprocess the DTV curve using a denoising and smoothing method.
[0068] Furthermore, the step S11 specifically includes the following steps:
[0069] S111, charging the lithium battery under constant current and constant voltage conditions, and obtaining voltage data and temperature data by appropriately adjusting the sampling interval;
[0070] S112. Calculate the ratio of temperature and voltage change by the following formula to obtain a DTV curve:
[0071]
[0072] In the above formula, T(k) represents the temperature value of the kth sampling, V(k) represents the terminal voltage value of the kth sampling, and N represents the sampling interval, which can be set as needed.
[0073] Meanwhile, in step S12, in order to obtain a smooth curve, the present invention uses a GF denoising and smoothing method to pre-process the DTV curve to obtain a filtered DTV curve. The GF algorithm is:
[0074]
[0075] Among them, σ represents the standard deviation, x represents the DTV data value before smoothing filtering, and μ represents the mean value.
[0076] S2. Select three feature points in the smooth DTV curve, including two peak points and one valley point, and calculate the area S of the triangle connected by the three feature points;
[0077] Specifically, attached Figure 2 The DTV curve obtained by using the GF filtering algorithm is shown in the figure. Three feature points are selected on the filtered DTV curve, two peak points are selected, recorded as point A and point B, and the trough point is selected as point C; connect the three points ABC and calculate the area of the triangle S ABC ;
[0078] Furthermore, the area S of the triangle formed by the three points is calculated according to the following formula: ABC :
[0079] L=(AB+BC+AC) / 2 (3)
[0080]
[0081] Where L represents the half side length of the triangle, AB represents the distance between point A and point B, and the same goes for BC and AC.
[0082] S3. Establish a BILSTM prediction model based on dot product attention, use the area S as input to train the model, and obtain the maximum remaining capacity prediction value of the battery;
[0083] Figure 3 This is a schematic diagram of the BILSTM model based on dot product attention, such as Figure 3 As shown in Figure 2, the BILSTM prediction model includes an input layer, a BILSTM network, a Dropout network, a dot product attention network, and an FC output layer; the dot product attention network is shown in Figure 2. Figure 4As shown in the figure, the dot product attention network includes the BILSTM model output layer, the softmax normalization layer and the dot product calculation layer; the output matrix Y of the BILSTM network is combined with the final hidden state h n Do the dot product to get vector e, which represents Y and h n The degree of relationship between them is shown in the following formula:
[0084] e i =Dot(Y i ,h n ) (5)
[0085] Where Dot(·) represents the dot product, Y i represents the output vectors that make up the matrix Y;
[0086] The obtained e vector is passed through the softmax function to obtain the attention weight α, as shown in the following formula:
[0087] α=softmax(e) (6)
[0088] Do a dot product of the attention weight and the output matrix Y to get the final output value.
[0089] After normalizing the data, the area of the triangle S ABC As the input of the model, the battery capacity Q is used as the output, and a BILSTM prediction model based on dot product attention is constructed. During the training process, a sliding window is used to generate training data, and the window size is The generated data is:
[0090]
[0091] Where T represents the prediction starting point, and the corresponding training label is:
[0092]
[0093] Where Q represents the corresponding battery capacity. During the training process, the present invention uses the Bayesian optimization algorithm to find the optimal parameters of the model. Bayesian optimization is a global optimization method based on a probability model. The core idea of Bayesian optimization is to find the optimal hyperparameters by constructing a probability model of the objective function. Bayesian optimization uses a Gaussian process as the probability model of the objective function and determines the next sampling point by maximizing the acquisition function. The calculation formula of the acquisition function is as follows:
[0094] α(x)=μ(x)+kσ(x) (9)
[0095] Where α(x) is the value of the acquisition function, μ(x) is the Gaussian process model’s prediction of the mean of the target function at point x, σ(x) is the Gaussian process model’s prediction of the standard deviation of the target function at point x, and k is a constant used to control the trade-off between exploration and exploitation.
[0096] Therefore, a BILSTM prediction model based on dot product attention is established, and the area S is used as the input training model to obtain the maximum remaining capacity prediction value of the battery. Specifically, the following steps are included:
[0097] S31. Build a prediction model, including an input layer, a BILSTM layer, a Dropout layer, a dot product attention layer, and an FC output layer;
[0098] S32, after preprocessing the area S data, divide it into a training set and a test set, use a sliding window to generate training sequence data from the training set, use the corresponding maximum remaining capacity as a training label, and then train the model;
[0099] S33. Use the trained prediction model to predict the lithium battery test data, and use RMSE and MAE to evaluate the prediction results of the model.
[0100] Furthermore, the dot product attention layer in step S31 specifically includes the following:
[0101] S311. Compare the output matrix Y of the BILSTM network with the final hidden state h n Do the dot product to get vector e, which represents Y and h n The degree of relationship between them is shown in the following formula:
[0102] e i =Dot(Y i ,h n )
[0103] Where Dot(·) represents the dot product, Y i represents the output vectors that make up the matrix Y;
[0104] S312, the obtained e vector is passed through the softmax function to obtain the attention weight α, as shown in the following formula:
[0105] α=softmax(e)
[0106] S313. Perform a dot product of the attention weight and the output matrix Y to obtain the final output value.
[0107] Furthermore, in step S33, the trained prediction model is used to predict the lithium battery test data, and the prediction results of the model are evaluated using RMSE and MAE, which specifically include:
[0108] Output the remaining maximum capacity C of the battery through the prediction model max , and then get the corresponding charge and discharge times The prediction performance of the model is evaluated using RMSE and MAE scores:
[0109]
[0110]
[0111] Where n is the length of the sequence, m is the length of the sample generated by the training sequence, and y i Indicates the actual maximum battery capacity value, Indicates the predicted maximum battery capacity value.
[0112] S4. Calculate the RUL value of the battery using the obtained maximum remaining capacity prediction value.
[0113] In this embodiment, the RUL value of the battery is calculated using the obtained maximum remaining capacity prediction value, and the calculation method is shown in the following formula:
[0114]
[0115] In the above formula Indicates the number of charge and discharge cycles when the predicted remaining capacity reaches the retirement threshold, n s Indicates the number of charge and discharge cycles when prediction starts.
[0116] In summary, the lithium battery RUL prediction method based on area features and data-driven described in an embodiment of the present invention introduces the concept of regional triangular area on the basis of the existing DTV method, and extracts the regional area S as a health factor, establishes the connection between microphase changes and macro signals in the battery degradation process, and improves the interpretability of the model. At the same time, by introducing dot product attention in the BILSTM network, the model's ability to capture strongly correlated features is enhanced, and the prediction accuracy and generalization are improved, thereby detecting and evaluating the RUL of the battery. The method of the present invention is simple, does not require a lot of calculations, can be combined with BMS, and has the potential for online application.
[0117] According to a second aspect of the present invention, a lithium battery RUL prediction system based on area characteristics and data driving is provided, comprising:
[0118] The data acquisition module is used to obtain the voltage data and temperature data of the charging at the set rate, calculate the ratio of the temperature and voltage changes to obtain the DTV curve, and pre-process the DTV curve using a denoising and smoothing method;
[0119] The regional triangle area calculation module is used to select three feature points in the smooth DTV curve, including two peak points and one valley point, and calculate the area S of the triangle connected by the three feature points;
[0120] The data-driven model building module is used to build a BILSTM prediction model based on dot product attention, and use the area S as the input training model to obtain the maximum remaining capacity prediction value of the battery;
[0121] The battery life prediction module is used to calculate the RUL value of the battery using the obtained maximum remaining capacity prediction value.
[0122] It can be understood that the lithium battery RUL prediction system based on area characteristics and data-driven provided by the present invention corresponds to the lithium battery RUL prediction method based on area characteristics and data-driven provided in the aforementioned embodiments. The relevant technical features of a lithium battery RUL prediction system based on area characteristics and data-driven can refer to the relevant technical features of a lithium battery RUL prediction method based on area characteristics and data-driven, which will not be repeated here.
[0123] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0124] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A lithium battery RUL prediction method based on area characteristics and data driven, characterized in that: The following steps are involved: S1. Obtaining voltage data and temperature data charged at a set rate, calculating the ratio of temperature and voltage changes to obtain a DTV curve, and preprocessing the DTV curve using a denoising and smoothing method; S2. Select three feature points in the smooth DTV curve, including two peak points and one valley point, and calculate the area S of the triangle connected by the three feature points; S3. Establish a BILSTM prediction model based on dot product attention, use the area S as input to train the model, and obtain the maximum remaining capacity prediction value of the battery; including: S31. Establish a BILSTM prediction model based on dot product attention, including an input layer, a BILSTM layer, a Dropout layer, a dot product attention layer, and an FC output layer; S311. Compare the output matrix Y of the BILSTM network with the final hidden state Do the dot product to get the vector , Indicates that Y and The degree of relationship between them is shown in the following formula: in, represents the dot product, represents the output vectors that make up the matrix Y; S312, what you will get Vector by Function gets the attention weight , as shown below: ; S313, do a dot product of the attention weight and the output matrix Y to obtain the final output value; S32, after preprocessing the area S data, divide it into a training set and a test set, use a sliding window to generate training sequence data from the training set, use the corresponding maximum remaining capacity as a training label, and then train the model; S33, using the trained prediction model to predict the lithium battery test data, and using RMSE and MAE to evaluate the prediction results of the model; S4. Calculate the RUL value of the battery using the obtained maximum remaining capacity prediction value.
2. A lithium battery RUL prediction method based on area characteristics and data drive according to claim 1, characterized in that: The step S1 comprises the following steps: S11, charging the lithium battery at a set charging rate, obtaining corresponding voltage data and temperature data, and calculating the ratio of temperature to voltage change to obtain a DTV curve; S12. Preprocess the DTV curve using a denoising and smoothing method.
3. A lithium battery RUL prediction method based on area characteristics and data drive according to claim 2, characterized in that: The step S11 comprises the following steps: S111, charging the lithium battery under constant current and constant voltage conditions, and obtaining voltage data and temperature data by appropriately adjusting the sampling interval; S112, calculating the ratio of temperature and voltage change by the following formula to obtain a DTV curve: In the formula, Indicates The temperature value of the sampling Indicates The terminal voltage value of the sampling time is Indicates the sampling interval.
4. A lithium battery RUL prediction method based on area characteristics and data drive according to claim 2, characterized in that: In step S12, the DTV curve is preprocessed by using the GF denoising and smoothing method to obtain a filtered DTV curve; wherein the GF algorithm is: in, represents the standard deviation, x represents the DTV data value before smoothing filtering, Indicates the average value.
5. The lithium battery RUL prediction method based on area characteristics and data drive according to claim 1, characterized in that: The step S2 comprises the following steps: S21. According to the obtained smooth DTV curve, two peak points are selected, which are recorded as point A and point B, and a trough point is selected, which is recorded as point C; S22. Calculate the area of the triangle formed by the three points according to the following formula, recorded as ; in, represents the half side length of the triangle, It represents the distance between point A and point B. The same goes for BC and AC.
6. The lithium battery RUL prediction method based on area characteristics and data drive according to claim 1, characterized in that: The use of RMSE and MAE to evaluate the prediction results of the model includes: Output the remaining maximum capacity of the battery through the prediction model , and then get the corresponding charge and discharge times , use RMSE and MAE scores to evaluate the prediction performance of the model: Among them, n is the length of the sequence, m is the length of the sample generated by the training sequence, Indicates the actual maximum battery capacity value, Indicates the predicted maximum battery capacity value.
7. The lithium battery RUL prediction method based on area characteristics and data drive according to claim 1, characterized in that: The RUL calculation formula of the battery is: in, Indicates the number of charge and discharge cycles when the remaining capacity reaches the retirement threshold. Indicates the number of charge and discharge cycles when prediction starts.
8. A lithium battery RUL prediction system based on area characteristics and data drive, characterized in that: include: The data acquisition module is used to obtain the voltage data and temperature data of the charging at the set rate, calculate the ratio of the temperature and voltage changes to obtain the DTV curve, and pre-process the DTV curve using a denoising and smoothing method; The regional triangle area calculation module is used to select three feature points in the smooth DTV curve, including two peak points and one valley point, and calculate the area S of the triangle connected by the three feature points; The data-driven model building module is used to build a BILSTM prediction model based on dot product attention, and use the area S as the input training model to obtain the maximum remaining capacity prediction value of the battery; including: S31. Establish a BILSTM prediction model based on dot product attention, including an input layer, a BILSTM layer, a Dropout layer, a dot product attention layer, and an FC output layer; S311. Compare the output matrix Y of the BILSTM network with the final hidden state Do the dot product to get the vector , Indicates that Y and The degree of relationship between them is shown in the following formula: in, represents the dot product, represents the output vectors that make up the matrix Y; S312, what you will get Vector by Function gets the attention weight , as shown below: ; S313, do a dot product of the attention weight and the output matrix Y to obtain the final output value; S32, after preprocessing the area S data, divide it into a training set and a test set, use a sliding window to generate training sequence data from the training set, use the corresponding maximum remaining capacity as a training label, and then train the model; S33, using the trained prediction model to predict the lithium battery test data, and using RMSE and MAE to evaluate the prediction results of the model; The battery life prediction module is used to calculate the RUL value of the battery using the obtained maximum remaining capacity prediction value.
