A method and system for predicting the state of health of lithium-ion batteries based on their DC internal resistance.

Abstract

This invention discloses a method and system for predicting the state of health of a lithium-ion battery based on its DC internal resistance, belonging to the technical field of battery health state prediction. The method includes the following steps: collecting pulse voltage data of a single lithium-ion battery throughout its entire life cycle and preprocessing the pulse voltage data; extracting the battery voltage rebound characteristic curve based on the preprocessed pulse voltage data; evaluating the current battery health state based on the extracted battery voltage rebound characteristic curve and outputting the evaluation result; predicting the battery life based on the current battery health state evaluation result and outputting the prediction result. Through the method and system of this invention, the process of predicting battery health state can be effectively simplified, meeting the requirements for widespread application in practical situations, and effectively improving the accuracy and efficiency of battery health state prediction. This, in turn, effectively reflects the internal aging and safety status of the battery so that abnormal batteries can be replaced and handled in a timely manner.

Description

Technical Field

[0001] This invention relates to the technical field of battery health state prediction, and in particular to a method and system for predicting battery health state based on the DC internal resistance of a lithium-ion battery. Background Technology

[0002] Lithium-ion batteries have been widely used in various fields in recent years due to their high energy density, long lifespan, and relatively low cost, becoming an important energy storage device for electric vehicles. However, with the increase in the number of charging cycles, batteries undergo irreversible aging, such as reduced driving range and even safety hazards. Therefore, accurate prediction of the state of health (SOH) and remaining useful life (RUL) of batteries will bring new technological advancements to battery manufacturing and BMS optimization. Furthermore, accurate prediction of the remaining battery life helps to achieve battery recycling, improve resource utilization, reduce the environmental harm of waste batteries, and thus promote the green development of the new energy vehicle industry.

[0003] Existing research on lithium-ion battery health diagnosis methods mainly falls into three categories: physical models, semi-empirical models, and data-driven methods. Physical and semi-empirical models aim to reveal various mechanisms related to battery capacity decay, such as the formation of the solid electrolyte interphase (SEI) film, lithium deposition, active material consumption, and increased internal resistance. While these models can predict battery life to some extent, in practical applications, battery charge-discharge behavior is often very complex and irregular. Diverse cycling conditions (such as charge-discharge rate and cycling temperature) cause different aging behaviors in batteries. Therefore, constructing physical or semi-empirical models suitable for actual user behavior to accurately predict battery health diagnosis and safety warnings is extremely challenging.

[0004] With the rise of artificial intelligence, data-driven methods based on machine learning and neural networks are widely used in battery health diagnosis. These methods do not require a deep understanding of the battery's electrochemical principles; they only need a large amount of data to train the model to achieve predictive capabilities. Data-driven methods place high demands on the battery's aging characteristic factors; the more accurately these factors reflect the battery's aging patterns, the higher the accuracy. Commonly used characteristic factors in current research include charge-discharge curves, incremental capacity curves (IC), and electrochemical impedance spectroscopy (EIS).

[0005] However, the methods described above have significant limitations in predicting battery health. The charge-discharge curve requires complete recording of voltage, current, and time data for each charge-discharge cycle, a cumbersome process with high computational demands. Incremental capacity curve analysis requires high-frequency data acquisition and differential calculations, which is not conducive to widespread application in real-world scenarios. EIS measurement is complex and the equipment is expensive, increasing testing costs and hindering large-scale deployment. These factors result in low accuracy and efficiency in battery health prediction. Summary of the Invention

[0006] To overcome the problems of existing battery health state prediction technologies, such as cumbersome prediction processes that hinder widespread application in real-world scenarios and consequently low accuracy and efficiency, this invention proposes a battery health state prediction method and system based on the DC internal resistance of lithium-ion batteries. This method effectively simplifies the battery health state prediction process, meets the requirements for widespread application in real-world scenarios, and thus effectively improves the accuracy and efficiency of battery health state prediction.

[0007] To achieve the objectives of this invention, the following technical solution is adopted:

[0008] A method for predicting the state of health of a lithium-ion battery based on its DC internal resistance, the method comprising the following steps:

[0009] Collect pulse voltage data of a single lithium-ion battery throughout its entire life cycle, and preprocess the pulse voltage data.

[0010] Based on the preprocessed pulse voltage data, extract the battery voltage rebound characteristic curve;

[0011] Based on the extracted battery voltage rebound characteristic curve, the current battery health status is assessed, and the assessment results are output.

[0012] Predict battery life based on the current battery health status assessment results and output the prediction results.

[0013] In the above technical solution, preprocessing the collected pulse voltage data can effectively improve the stability and reliability of the data. Based on the preprocessed pulse voltage data, the battery voltage rebound characteristic curve is extracted. This involves applying a pulse current and collecting voltage rebound data, which greatly simplifies the experimental process, allows for real-time testing, and reduces instrument and measurement costs. Furthermore, the extracted battery voltage rebound characteristic curve is used to assess the battery health status, effectively simplifying the assessment process and meeting the requirements for widespread application in real-world situations. It also effectively predicts battery life based on the current battery health status assessment results, improving the accuracy and efficiency of battery health status prediction. This, in turn, effectively reflects the internal aging and safety status of the battery, enabling timely replacement and handling of abnormal batteries.

[0014] Furthermore, the preprocessing of the pulse voltage data includes:

[0015] The collected pulse voltage data is uploaded to the lithium battery health database, and the historical data in the battery health database is cleaned.

[0016] Furthermore, the process of cleaning and processing historical data in the battery health database includes:

[0017] The historical data in the battery health database were sequentially processed by tabulating, filling in missing values, splitting rows and columns, deleting duplicate data, normalizing data, and integrating data to obtain voltage and capacity data during battery pulse discharge.

[0018] The expression for normalizing historical data in the battery health database is as follows:

[0019]

[0020] Where x1 represents unnormalized voltage data; x max and x min This represents the maximum and minimum voltage values ​​at different cycle numbers within the same time period.

[0021] In the above technical solution, the collected pulse voltage data is processed by tabulating, filling in missing values, splitting rows and columns, deleting duplicate data, normalizing data, and integrating data, which can effectively improve the stability and reliability of the data.

[0022] Furthermore, the process of extracting the battery voltage rebound characteristic curve includes:

[0023] For batteries that cycle throughout their entire lifespan, a complete pulse charging signal is applied in each cycle, and the battery voltage rebound characteristic curve is extracted during the entire pulse charging process.

[0024] Parameter identification is performed on the voltage rebound characteristic curve during a pulse charging process to obtain DC internal resistance data, and the DC internal resistance data is divided into training set and test set.

[0025] The battery capacity data is calculated using the ampere-hour integration method through the battery cabinet's own algorithm.

[0026] In the above technical solution, applying a pulse current and collecting voltage rebound data can greatly simplify the experimental process, enable real-time testing, and reduce instrument and measurement costs. Furthermore, the DC internal resistance of the battery is reflected by the voltage rebound data, and the experiment can be conducted at any state of charge (SOC) without the need for a complete charge and discharge process. This can effectively simplify the process of predicting the battery health status. At the same time, the prediction accuracy varies at different SOCs, and the prediction results for a more stable electrochemical state are also more stable. Moreover, battery health diagnosis based on DC internal resistance can greatly simplify the experimental process, enable real-time testing, and reduce instrument and measurement costs.

[0027] Furthermore, the process of assessing the current battery health status based on the extracted battery voltage rebound characteristic curve includes:

[0028] A battery health status assessment model is constructed, which includes a convolution module, a bidirectional long short-term memory module, and an assessment module.

[0029] The training set is input into the battery health status assessment model, and several rounds of parameter iterative optimization training are set to optimize the parameters of the battery health status assessment model.

[0030] The battery health status features in the training set are extracted using a convolutional module.

[0031] Long-term correlation features in the pulse charge and discharge process of batteries in the training set are extracted using a bidirectional long short-term memory module.

[0032] The prediction module assesses the battery's health status based on the extracted battery health status features and long-term correlation features, and outputs the assessment results.

[0033] The prediction results are tested using a test set, and a loss function is set to optimize the parameters of the battery health status assessment model after each training round. When the preset iterative training rounds end or the loss function converges, the trained battery health status assessment model is obtained.

[0034] Among them, battery health status is represented by the battery's maximum capacity in the current cycle divided by the battery's initial capacity, and the evaluation results include early battery capacity data.

[0035] Furthermore, the process of extracting long-term correlated features from the battery pulse charge-discharge process in the training set using a bidirectional long short-term memory module includes:

[0036] The bidirectional long short-term memory module includes an input layer, a hidden layer, and an output layer;

[0037] The input layer, hidden layer, and output layer are used to extract long-term sequences during the battery's cyclic charging and discharging process. Long-term correlation features are then extracted based on these long-term sequences, expressed as follows:

[0038] f(t)=σ(W g x(t)+W f h(t-1)+b f );

[0039] i(t)=σ(W i x(t)+W i h(t-1)+b i );

[0040] o(t)=σ(W o x(t)+W o h(t-1)+b o );

[0041] C(t)=f(t)⊙C(t-1)+i(t)⊙tanh(W c x(t)+W c h(t-1)+b c );

[0042] y(t) = o(t) ⊙ tanh(C(t));

[0043] The expression for setting the loss function to optimize the parameters of the battery health status assessment model after each training round is as follows:

[0044]

[0045] Where x(t) and y(t) represent the input and output vectors at the current time step t, respectively, f(t) represents the activation value of the forget gate, i(t) represents the activation value of the input gate, o(t) represents the activation value of the output gate, and Y... i , and R represents the true value, the predicted value, and the average of the true values, respectively. 2 The coefficient of determination is represented by , and RMSE represents the root mean square error.

[0046] In the above technical solution, the constructed battery health status assessment model, after training on the training set, testing on the test set, and optimization of the loss function, can effectively grasp the changes in pulse voltage data throughout the entire life cycle of the lithium-ion battery. Then, based on the cooperation between the convolution module, the bidirectional long short-term memory module, and the assessment module, the current battery health status is assessed, which effectively simplifies the process of predicting the full life cycle health status of the battery and can meet the requirements of wide application in practical situations, thereby effectively improving the accuracy and efficiency of battery health status prediction.

[0047] Furthermore, the process of predicting battery life based on the current battery health status assessment results includes:

[0048] Predict the battery's capacity degradation trajectory based on early battery capacity data;

[0049] Based on the battery's capacity degradation trajectory, predict the battery's end-of-life and current cycle count, and then predict the remaining battery life based on the end-of-life and current cycle count.

[0050] Furthermore, the process of predicting the battery's capacity degradation trajectory includes:

[0051] A battery capacity degradation trajectory prediction model is constructed, which includes a self-attention layer, a feedforward neural network layer, and a battery capacity degradation trajectory prediction layer.

[0052] The training set is input into the battery capacity degradation trajectory prediction model, and several rounds of parameter iteration optimization training are set to optimize the parameters of the battery capacity degradation trajectory prediction model.

[0053] The self-attention layer extracts battery capacity features from the training set, the feedforward neural network layer processes the extracted battery capacity features point by point, and the battery capacity degradation trajectory prediction layer predicts the battery capacity degradation trajectory based on the battery capacity features after point-by-point processing and outputs the prediction results.

[0054] The prediction results are tested using a test set, and a loss function is set to optimize the parameters of the battery capacity degradation trajectory prediction model after each training round. When the preset iterative training rounds end or the loss function converges, the trained battery capacity degradation trajectory prediction model is obtained.

[0055] Furthermore, the self-attention layer uses a self-attention mechanism to extract battery capacity features from the training set, expressed as:

[0056]

[0057] The feedforward neural network layer processes the output of the self-attention layer point by point, as expressed in the following expression:

[0058] FFN(x)=max(0,xW1+b1)W2+b2;

[0059] The battery capacity degradation trajectory prediction layer predicts the battery capacity degradation trajectory based on the point-by-point processed battery capacity characteristics. The expression is as follows:

[0060] Output = softmax(W) out ·DecoderOutput+b iut );

[0061] The formula for predicting the remaining battery life based on the battery's end-of-life and current cycle count is:

[0062] RUL = Battery Life - Number of Current Charge-Discharge Cycles;

[0063] Where Q is the query matrix, K is the key matrix, and V is the value matrix; d k denoted as the dimension of the key vector, x represents the input feature, W1 and W2 both represent the weight matrices of the feedforward neural network, b1 and b2 represent the bias vectors of the feedforward neural network, and RUL ultimately calculates the remaining battery life.

[0064] In the above technical solution, the constructed battery capacity degradation trajectory prediction model, after training on the training set, testing on the test set, and optimization of the loss function, can effectively grasp the changes in pulse voltage data throughout the entire life cycle of a lithium-ion battery. Furthermore, based on the cooperation between the self-attention layer, the feedforward neural network layer, and the battery capacity degradation trajectory prediction layer, it can predict the battery capacity and lifespan, effectively simplifying the battery health status prediction process and meeting the requirements for wide application in practical situations, thereby effectively improving the accuracy and efficiency of battery health status prediction.

[0065] A battery health state prediction system based on the DC internal resistance of a lithium-ion battery, the system comprising:

[0066] The data acquisition module is used to collect pulse voltage data of a single lithium-ion battery throughout its entire life cycle.

[0067] The data processing module is used to preprocess the pulse voltage data;

[0068] The data extraction module is used to extract the battery voltage rebound characteristic curve based on the preprocessed pulse voltage data.

[0069] The status assessment module is used to assess the current battery health status based on the extracted battery voltage rebound characteristic curve and output the assessment results.

[0070] The lifespan prediction module is used to predict battery lifespan based on the current battery health status assessment results and output the prediction results.

[0071] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0072] This invention proposes a method and system for predicting the health status of lithium-ion batteries based on their DC internal resistance. By preprocessing the collected pulse voltage data, the stability and reliability of the data can be effectively improved. Based on the preprocessed pulse voltage data, the battery voltage rebound characteristic curve is extracted. That is, by applying a pulse current and collecting voltage rebound data, the experimental process can be greatly simplified, real-time testing can be performed, and instrument and measurement costs can be reduced. Based on the extracted battery voltage rebound characteristic curve, the battery health status is evaluated, which can effectively simplify the battery health status evaluation process, meet the requirements of wide application in practical situations, and effectively predict battery life based on the current battery health status evaluation results. This improves the accuracy and efficiency of battery health status prediction, and effectively reflects the internal aging and safety status of the battery so as to replace and handle abnormal batteries in a timely manner. Attached Figure Description

[0073] Figure 1 A flowchart illustrating the steps of a battery health state prediction method based on the DC internal resistance of a lithium-ion battery, provided in this application embodiment;

[0074] Figure 2 A schematic diagram illustrating the state of pulse charging segments collected during battery cycling in the lithium battery health diagnosis method based on the DC internal resistance of lithium-ion batteries provided in this application embodiment;

[0075] Figure 3 A flowchart for assessing the health status of a lithium battery based on its DC internal resistance, provided in this application embodiment;

[0076] Figure 4 A graph showing the SOH estimation results of a lithium battery based on the DC internal resistance of the lithium-ion battery, provided in an embodiment of this application;

[0077] Figure 5 A flowchart of lithium battery capacity degradation trajectory and RUL prediction based on early capacity data of lithium-ion batteries is provided for the embodiments of this application.

[0078] Figure 6 The following diagram illustrates the lithium battery capacity degradation trajectory and RUL prediction results based on early capacity data of lithium-ion batteries, provided for embodiments of this application.

[0079] Figure 7 This is a schematic diagram of a battery health status prediction system based on the DC internal resistance of a lithium-ion battery, provided in an embodiment of this application. Detailed Implementation

[0080] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0081] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0082] Example 1:

[0083] A method for predicting the state of health of lithium-ion batteries based on their DC internal resistance, see [link to relevant documentation]. Figure 1 The method includes the following steps:

[0084] S1: Collect pulse voltage data of a single lithium-ion battery throughout its entire life cycle, and preprocess the pulse voltage data;

[0085] S2: Extract the battery voltage rebound characteristic curve based on the preprocessed pulse voltage data;

[0086] S3: Based on the extracted battery voltage rebound characteristic curve, assess the current battery health status and output the assessment results.

[0087] S4: Predict battery life based on the current battery health status assessment results and output the prediction results.

[0088] In step S1, pulse voltage data of a single lithium-ion battery throughout its entire life cycle is collected, specifically as follows:

[0089] For batteries that have a full life cycle, such as Figure 2The diagram illustrates four different states of charge (SOC) within a single cycle. Periodic positive and negative pulse signals are applied to control the flow of charging current, acquiring pulse voltage data of the lithium-ion battery throughout its entire lifespan. Battery capacity data is derived using the battery cabinet's own algorithm based on the ampere-hour integration method and uploaded to the battery health database. The pulse voltage data reflects the battery's DC internal resistance; therefore, this invention is essentially based on DC internal resistance. The entire lifespan data acquisition involves constant current and constant voltage charging and discharging of a brand-new lithium-ion battery until its capacity drops to 80% of its rated capacity. During this period, the current for constant current and constant voltage charging and discharging is set to any multiple. A pulse charging process involves applying a positive pulse to the battery for a certain duration, resting for a certain time, then applying a negative pulse for a certain duration, and resting for a certain time (where the certain duration can be any value within 2 to 600 seconds). The magnitude of the positive and negative pulse currents is greater than the charging and discharging currents set for the entire lifespan cycle. The pulse charging acquisition nodes are mainly divided into four charging and discharging states: charging start, rest after full charge, discharging start, and rest after full discharge. The entire experiment was conducted at three temperatures: 10℃, 25℃, and 45℃, with the temperature controlled by a constant temperature chamber.

[0090] like Figure 2 As shown, this embodiment uses data from state 3 for illustration. The data usage method for other states is consistent with state 3, but the specific prediction results will differ somewhat. Because the DC internal resistance is reflected in the pulse voltage, this invention does not use the traditional method of defining the voltage ratio of the resistor, but only uses the raw pulse voltage data, including the pulse voltage of the battery at 10°C, 25°C, and 45°C.

[0091] In this embodiment, there are 7 batteries at 45°C, numbered 01-07; 9 batteries at 25°C, numbered 08-16; and 8 batteries at 10°C, numbered 17-24. Data from either the pulse charging or pulse discharging phase can be used. In this step, to obtain more accurate prediction results, voltage data from the more electrochemically stable discharge phase is collected and used.

[0092] Further, in step S1, the preprocessing of the pulse voltage data includes:

[0093] The collected pulse voltage data is uploaded to the lithium battery health database, and the historical data in the battery health database is cleaned.

[0094] Specifically, in step S11, the process of cleaning the historical data in the battery health database includes:

[0095] The historical data in the battery health database were sequentially processed by tabulating, filling in missing values, splitting rows and columns, deleting duplicate data, normalizing data, and integrating data to obtain voltage and capacity data during battery pulse discharge.

[0096] The expression for normalizing historical data in the battery health database is as follows:

[0097]

[0098] Where, x i This represents unnormalized voltage data; x max and x min This represents the maximum and minimum voltage values ​​at different cycle numbers within the same time period.

[0099] In some embodiments, the obtained voltage and capacity data are preprocessed, including removing missing values, duplicate values, and non-numerical data, and processing abnormal noise in the data using polynomial interpolation to ensure data stability and reliability. To more intuitively detect data anomalies, images of the original data are plotted based on the preprocessed battery data, including images of pulse voltage changes and battery capacity decay, with the excitation time and cycle number on the x-axis, respectively. The changes in characteristic data during battery aging are observed. After confirming that there are no obvious anomalies in the data, normalization is performed using the MIN-MAX method, satisfying the formula:

[0100]

[0101] Where, x i It is unnormalized voltage data, x max and x min These are the maximum and minimum voltage values ​​across different cycle numbers at the same time. Normalizing the data helps the model learn the essential characteristics of battery aging, avoids errors caused by excessive data differences, and accelerates convergence.

[0102] Understandably, performing tabular processing, filling in missing values, splitting rows and columns, deleting duplicate data, normalizing data, and integrating data on the collected pulse voltage data can effectively improve the stability and reliability of the data.

[0103] In step S2, see Figure 3 The process of extracting battery voltage rebound characteristic curves and battery capacity data includes:

[0104] For batteries that cycle throughout their entire lifespan, a complete pulse charging signal is applied in each cycle, and the battery voltage rebound characteristic curve is extracted during the entire pulse charging process.

[0105] Parameter identification is performed on the voltage rebound characteristic curve during a pulse charging process to obtain DC internal resistance data, and the DC internal resistance data is divided into training set and test set.

[0106] The battery capacity data is calculated using the ampere-hour integration method and the battery cabinet's own algorithm.

[0107] Specifically, in step S12, the complete battery dataset (i.e., the DC internal resistance dataset) is divided into a training set and a test set. In this embodiment, leave-one-out cross-validation is used, with the data from one battery serving as the test set and the data from the remaining batteries serving as the training set. First, cross-validation is performed on batteries within the same temperature range, and then cross-temperature predictions are made.

[0108] Specifically, in this embodiment, for a 45°C battery, one battery from numbers 01-07 is used as the test battery, and the pulse voltage data of the remaining batteries are merged to form a training set. During merging, it is ensured that the features correspond, that is, the same column represents the pulse voltage value at the same time. For example, 01-06 are selected for training, and 07 is used for testing.

[0109] In this embodiment, for the 25℃ battery, one battery from numbers 08-16 is used as the test battery, and the pulse voltage data of the remaining batteries are merged to form the training set. During merging, it is ensured that the features correspond, that is, the same column represents the pulse voltage value at the same time. For example, 08-15 is selected for training, and 16 is used for testing.

[0110] In this embodiment, for a 10°C battery, one battery from numbers 17-24 is used as the test battery, and the pulse voltage data of the remaining batteries are merged to form a training set. During merging, it is ensured that the features correspond, that is, the same column represents the pulse voltage value at the same time. For example, 17-23 are selected for training, and 24 is used for testing.

[0111] Understandably, applying a pulsed current and collecting voltage rebound data can greatly simplify the experimental process, enabling real-time testing and reducing instrument and measurement costs. Furthermore, the battery's DC internal resistance is reflected by the voltage rebound data, allowing the experiment to be conducted at any state of charge (SOC) without requiring a complete charge-discharge process. This effectively simplifies the process of predicting battery health status. At the same time, the prediction accuracy varies across different SOCs, and the prediction results for more stable electrochemical states are also more stable. Moreover, battery health diagnosis based on DC internal resistance can greatly simplify the experimental process, enabling real-time testing and reducing instrument and measurement costs.

[0112] In step S3, the process of assessing the battery health status based on the extracted battery voltage rebound characteristic curve includes:

[0113] S31: Construct a battery health status assessment model, which includes a convolution module, a bidirectional long short-term memory module, and a prediction module;

[0114] S32: Input the training set into the battery health status assessment model, and set up several rounds of parameter iterative optimization training to optimize the parameters of the battery health status assessment model;

[0115] S33: Extract battery health status features from the training set using a convolutional module;

[0116] S34: Use a bidirectional long short-term memory module to extract long-term correlation features from the battery cycle charge and discharge process in the training set;

[0117] S35: The prediction module performs a pre-assessment of the battery's health status based on the extracted battery health status features and long-term correlation features, and outputs the assessment results.

[0118] S36: Test the prediction results using the test set, and set the loss function to optimize the parameters of the battery health status assessment model after each training round. When the preset iterative training rounds end or the loss function converges, the trained battery health status assessment model is obtained.

[0119] Among them, battery health status is represented by the battery's maximum capacity in the current cycle divided by the battery's initial capacity, and the evaluation results include early battery capacity data.

[0120] In some embodiments, the battery health status assessment model constructed in this application may be a CNN-BiLSTM model.

[0121] Specifically, in step S33, the convolutional neural network (CNN) module consists of convolutional layers, pooling layers, and fully connected layers. Each convolutional and pooling layer can be viewed as multiple two-dimensional matrices, combined into a tensor. The feature plane realizes data input, output, and backpropagation through unconnected neurons, which share a series of filters to extract data features. The pooling layer works similarly to the filters, enhancing the feature extraction effect. As the network structure deepens, the extracted data topology features gradually exhibit non-specific characteristics, ultimately obtaining the data features of the battery health status (i.e., battery health status features).

[0122] Specifically, in step S34, in order to better address the extraction of long-term correlated features during battery cyclic charging and discharging, a bidirectional long short-term memory (BiLSTM) module is linked after the fully connected layer of the convolutional module. The process of using the BiLSTM module to extract long-term correlated features during battery cyclic charging and discharging in the training set includes:

[0123] The bidirectional long short-term memory module includes an input layer, a hidden layer, and an output layer;

[0124] The input layer, hidden layer, and output layer are used to extract long-term sequences during the battery's cyclic charging and discharging process. Long-term correlation features are then extracted based on these long-term sequences, expressed as follows:

[0125] f(t)=σ(W f x(t)+W f h(t-1)+b f );

[0126] i(t)=σ(W i x(t)+W i h(t-1)+b i );

[0127] ο(t)=σ(W ο x(t)+W ο h(t-1)+b ο );

[0128] C(t)=f(t)⊙C(t-1)+i(t)⊙tanh(W c x(t)+W c h(t-1)+b c );

[0129] y(t)=ο(t)⊙tanh(C(t));

[0130] The expression for setting the loss function to optimize the parameters of the battery health status assessment model after each training round is as follows:

[0131]

[0132] Where x(t) and y(t) represent the input and output vectors at the current time step t, respectively, f(t) represents the activation value of the forget gate, i(t) represents the activation value of the input gate, o(t) represents the activation value of the output gate, and Y... i , and R represents the true value, the predicted value, and the average of the true values, respectively. 2 R represents the coefficient of determination. 2 The coefficient of determination is represented by , and RMSE represents the root mean square error.

[0133] Specifically, in this embodiment, the same DC internal resistance data is used as input. The estimation results in this embodiment are compared with the estimation results of traditional neural network models such as BP, LSTM, GPR, and CNN-LSTM, proving that the present solution has higher accuracy and is more stable.

[0134] In this embodiment, parameter identification is performed on experimental data from different temperatures to establish and train a battery health status assessment model, demonstrating the feasibility of the method. Finally, to achieve cross-temperature assessment, battery data from different temperatures are combined as input to the neural network, with one battery selected for testing at each temperature. For ease of illustration, the selected test set consists of three batteries: 02, 15, and 21, corresponding to 45℃, 25℃, and 10℃ respectively. The predicted results are as follows... Figure 4 As shown. The neural network used is the CNN-BILSTM described above, with minor adjustments to the network parameters. Figure 4 The horizontal axis represents the number of cycles, and the vertical axis represents SOH. True value represents the actual SOH value; CNN-BiLSTM Predicted represents the SOH value estimated by the CNN-BiLSTM neural network; R 2 This represents the coefficient of determination.

[0135] Understandably, the constructed battery health status assessment model, after training on the training set, testing on the test set, and optimizing the loss function, can effectively grasp the changes in pulse voltage data throughout the entire life cycle of a lithium-ion battery. Furthermore, based on the cooperation between the convolution module, the bidirectional long short-term memory module, and the prediction module, it can predict the battery's health status, effectively simplifying the battery health status prediction process and meeting the requirements for wide application in real-world situations, thereby effectively improving the accuracy and efficiency of battery health status prediction.

[0136] In this embodiment, preprocessing the collected pulse voltage data effectively improves the stability and reliability of the data. Based on the preprocessed pulse voltage data, the battery voltage rebound characteristic curve is extracted—that is, by applying a pulse current and collecting voltage rebound data—which greatly simplifies the experimental process, allows for real-time testing, and reduces instrument and measurement costs. The extracted battery voltage rebound characteristic curve is used to assess the battery health status, effectively simplifying the assessment process and meeting the requirements for widespread application in real-world scenarios. Furthermore, it effectively predicts battery life based on the current battery health status assessment results, improving the accuracy and efficiency of battery health status prediction. This, in turn, effectively reflects the internal aging and safety status of the battery, enabling timely replacement and handling of abnormal batteries.

[0137] Example 2:

[0138] This embodiment is based on the battery health state prediction method based on the DC internal resistance of lithium-ion batteries described in Embodiment 1, and further explains step S4 as follows:

[0139] Specifically, pulse voltage rebound data and capacity data of the battery life cycle described in Example 1 were collected. In Example 2, there were a total of 7 batteries at 45°C, numbered 01-07; a total of 9 batteries at 25°C, numbered 08-16; and a total of 8 batteries at 10°C, numbered 17-24.

[0140] Furthermore, the obtained capacity data is preprocessed, including removing missing values, duplicate values, and non-numerical data. Abnormal noise in the data is processed using polynomial interpolation to ensure data stability and reliability. To more intuitively detect data anomalies and differences between batteries, a battery capacity curve is plotted with the number of battery cycles on the x-axis and battery capacity on the y-axis.

[0141] The complete battery capacity dataset (the early battery capacity data estimated in Example 1) is divided into a training set and a test set. In this example, leave-one-out cross-validation is used, with the capacity data of one battery used as the test set and the capacity data of the remaining batteries used as the training set, allowing the model to learn the decay pattern of the capacity curve. To closely reflect real-world usage, the input data for the capacity degradation trajectory is the battery capacity value estimated by the CNN-BiLSTM neural network in Example 1, based on which the future capacity curve of the battery is deduced.

[0142] Specifically, in Example 2, for the 45°C battery, the capacity estimation results of one battery from numbers 01-07 are used as the test set, and the capacity data of the remaining batteries are merged as the training set. For example, 01-06 are selected for training, and 07 is used for testing.

[0143] In Example 2, for the 25°C battery, the capacity estimation results of one battery from numbers 08-16 are used as the test set, and the capacity data of the remaining batteries are merged as the training set. For example, 08-15 are selected for training, and 16 is used for testing.

[0144] In Example 2, for the 10°C battery, the capacity estimation results of one battery from numbers 17-24 are used as the test set, and the capacity data of the remaining batteries are merged as the training set. For example, 17-23 are selected for training, and 24 is used for testing.

[0145] The process of predicting battery life based on the current battery health status assessment results includes:

[0146] S41: Predict the battery's capacity degradation trajectory based on early battery capacity data;

[0147] S42: Based on the battery's capacity degradation trajectory, predict the battery's end-of-life and current cycle count, and predict the battery's remaining lifespan based on the battery's end-of-life and current cycle count.

[0148] In step S41, see Figure 5 The process of predicting the battery's capacity degradation trajectory includes:

[0149] S411: Construct a battery capacity degradation trajectory prediction model, which includes a self-attention layer, a feedforward neural network layer, and a battery capacity degradation trajectory prediction layer.

[0150] S412: Input the training set into the battery capacity degradation trajectory prediction model, and set up several rounds of parameter iteration optimization training to optimize the parameters of the battery capacity degradation trajectory prediction model;

[0151] S413: The self-attention layer extracts battery capacity features from the training set, the feedforward neural network layer processes the extracted battery capacity features point by point, and the battery capacity degradation trajectory prediction layer predicts the battery capacity degradation trajectory based on the battery capacity features after point-by-point processing and outputs the prediction results.

[0152] S414: Test the prediction results using the test set, and set the loss function to optimize the parameters of the battery capacity degradation trajectory prediction model after each training round. When the preset iterative training rounds end or the loss function converges, the trained battery capacity degradation trajectory prediction model is obtained.

[0153] In step S413, the self-attention layer uses a self-attention mechanism to extract battery capacity features from the training set, expressed as:

[0154]

[0155] The feedforward neural network layer processes the output of the self-attention layer point by point, as expressed in the following expression:

[0156] FFN(x)=max(0,xW1+b1)W2+b2;

[0157] The battery capacity degradation trajectory prediction layer predicts the battery capacity degradation trajectory based on the point-by-point processed battery capacity characteristics. The expression is as follows:

[0158] Output = softmax(W) out ·DecoderOutput+b out );

[0159] In step S42, the remaining battery life is predicted based on the battery's end-of-life and the current number of cycles, expressed as:

[0160] RUL = Battery Life - Number of Current Charge-Discharge Cycles;

[0161] Where Q is the query matrix, K is the key matrix, and V is the value matrix; d kLet x be the dimension of the key vector, x represent the input feature, W1 and W2 both represent the weight matrices of the feedforward neural network, b1 and b2 represent the bias vectors of the feedforward neural network, and RUL finally calculates the remaining battery life.

[0162] Specifically, in some embodiments, the battery capacity degradation trajectory prediction model uses the Transformer model. The Transformer model consists of encoder and decoder modules. Each encoder and decoder module contains multiple self-attention layers and feedforward neural network layers, combined to form a complex network architecture. The self-attention mechanism allows the model to establish global connections between different positions in the input sequence, extracting key features from the data through weighted summation. The feedforward neural network layer further processes and transforms the extracted features. Through the stacking of multiple encoder and decoder layers, the extracted data features gradually exhibit high-level and abstract characteristics, ultimately obtaining the data features of the battery health status.

[0163] In this embodiment, the self-attention mechanism is implemented through the following formula:

[0164]

[0165] Multi-head self-attention is the result of parallel computation of multiple self-attention mechanisms:

[0166] MultiHead(Q,K,V)=Concat(head1,head 1, ...,head h W o

[0167] The feedforward neural network layer processes the output of the self-attention layer point by point:

[0168] FFN(x) = max(0, xW1+b1)W2+b2

[0169] The calculation formula for the cross-attention layer is similar to that for the self-attention layer of the encoder, but it uses the encoder's output as the key and value:

[0170]

[0171] Positional codes are added to the input embedding vector, and the positional codes are generated using sine and cosine functions:

[0172]

[0173] By stacking multiple encoders and decoders, the model gradually extracts high-level and abstract features, ultimately obtaining data features related to battery health status. The final output of the decoder is then passed through a linear transformation and a softmax layer to generate a prediction result.

[0174] Output = softmax(W) out ·DecoderOutput+b out )

[0175] In this embodiment, considering the extreme performance capabilities of the detection model, the validation set is selected from early cyclic data predicting future data. Specifically, data from approximately the first 20% of the battery's lifespan is selected. Since the inherent capacity degradation patterns of batteries differ significantly at different temperatures, and the model's learning is based on the macroscopic degradation patterns of the capacity curve, this embodiment considers batteries at different temperatures in groups rather than mixing them together.

[0176] In this embodiment, to test the extreme capabilities of the model, early battery capacity data from previous tests were selected for extrapolation. Specifically, for a 10°C battery, capacity data from the first 45 cycles was used to predict the capacity for the next 200 cycles; for a 25°C battery, capacity data from the first 50 cycles was used to predict the capacity for the next 400 cycles; and for a 45°C battery, capacity data from the first 100 cycles was used to predict the capacity for the next 350 cycles. This embodiment only provides one possible scheme for selecting the number of early battery cycles, and the scheme for selecting the number of early battery cycles in this invention is not limited to this.

[0177] Based on the above steps, the future capacity data of the battery at different temperatures is predicted using the Transformer model, and the battery capacity degradation curve is plotted accordingly. In this embodiment, the capacity degradation curve is a composite of two parts: the early capacity data estimated in Embodiment 1 and the future capacity data predicted in this embodiment. Based on the overall capacity degradation curve, the maximum number of cycles to 75% of the initial capacity value is found as the battery's end lifespan, and the battery's maximum lifespan (N) is taken as the battery's maximum lifespan. EOL The predicted value of the battery is then used to predict the remaining useful life (RUL) for a specific cycle. For a specific cycle, based on its DCIR data, the model in Example 1 can be used to obtain an estimated capacity value, which is then mapped to the capacity degradation curve predicted in Example 2 to obtain the cycle number corresponding to that cycle, as the "current cycle number". Based on the battery's maximum lifespan (N... EOL Subtract the current cycle number (N) Current This allows us to obtain a predicted value for the remaining battery life. The prediction results of this scheme are as follows: Figure 6 As shown. Figure 6The horizontal axis represents the number of battery cycles, and the vertical axis represents the battery capacity value. True Data represents the actual capacity value; Prediction represents the early capacity data estimated in Example 1; Extrapolation represents the predicted battery degradation trajectory; Threhold is the threshold line for battery capacity to decrease to 75% of the initial value; Uncertainty represents the uncertainty of the prediction; R 2 RMSE and EoL_error represent the coefficient of determination, root mean square error, and termination lifetime error, respectively.

[0178] In this embodiment, the constructed battery capacity degradation trajectory prediction model, after training on the training set, testing on the test set, and optimization of the loss function, can effectively grasp the changes in pulse voltage data throughout the entire life cycle of a lithium-ion battery. Furthermore, based on the cooperation between the self-attention layer, the feedforward neural network layer, and the battery capacity degradation trajectory prediction layer, it can predict the battery capacity and lifespan, effectively simplifying the battery health status prediction process and meeting the requirements for widespread application in practical situations, thereby effectively improving the accuracy and efficiency of battery health status prediction.

[0179] Example 3:

[0180] A battery health state prediction system based on the DC internal resistance of lithium-ion batteries, see [link to relevant documentation]. Figure 7 The system includes:

[0181] The data acquisition module is used to collect pulse voltage data of a single lithium-ion battery throughout its entire life cycle.

[0182] The data processing module is used to preprocess the pulse voltage data;

[0183] The data extraction module is used to extract the battery voltage rebound characteristic curve based on the preprocessed pulse voltage data.

[0184] The status assessment module is used to assess the current battery health status based on the extracted battery voltage rebound characteristic curve and output the assessment results.

[0185] The lifespan prediction module is used to predict battery lifespan based on the current battery health status assessment results and output the prediction results.

[0186] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting the state of health of a battery based on the DC internal resistance of a lithium-ion battery, characterized in that, The method includes the following steps: Collect pulse voltage data of a single lithium-ion battery throughout its entire life cycle, and preprocess the pulse voltage data. Based on the preprocessed pulse voltage data, extract the battery voltage rebound characteristic curve; Based on the extracted battery voltage rebound characteristic curve, the current battery health status is assessed, and the assessment results are output. Predict battery life based on the current battery health status assessment results and output the prediction results; The process of extracting the battery voltage rebound characteristic curve includes: For batteries that cycle throughout their entire lifespan, a complete pulse charging signal is applied in each cycle, and the battery voltage rebound characteristic curve is extracted during the entire pulse charging process. Parameter identification is performed on the voltage rebound characteristic curve during a pulse charging process to obtain DC internal resistance data, and the DC internal resistance data is divided into training set and test set. The battery capacity data is calculated using the ampere-hour integration method through the battery cabinet's own algorithm. The process of assessing the current battery health status based on the extracted battery voltage rebound characteristic curve includes: A battery health status assessment model is constructed, which includes a convolution module, a bidirectional long short-term memory module, and an assessment module. The training set is input into the battery health status assessment model, and several rounds of parameter iterative optimization training are set to optimize the parameters of the battery health status assessment model. The battery health status features in the training set are extracted using a convolutional module. Long-term correlation features in the pulse charge and discharge process of batteries in the training set are extracted using a bidirectional long short-term memory module. The prediction module assesses the battery's health status based on the extracted battery health status features and long-term correlation features, and outputs the assessment results. The prediction results are tested using a test set, and a loss function is set to optimize the parameters of the battery health status assessment model after each training round. When the preset iterative training rounds end or the loss function converges, the trained battery health status assessment model is obtained. Among them, battery health status is represented by the battery's maximum capacity in the current cycle divided by the battery's initial capacity, and the evaluation results include early battery capacity data.

2. The battery health state prediction method based on the DC internal resistance of a lithium-ion battery according to claim 1, characterized in that, The preprocessing process for the pulse voltage data includes: The collected pulse voltage data is uploaded to the lithium battery health database, and the historical data in the battery health database is cleaned.

3. The battery health state prediction method based on the DC internal resistance of a lithium-ion battery according to claim 2, characterized in that, The process of cleaning historical data in the battery health database includes: The historical data in the battery health database were sequentially processed by tabulating, filling in missing values, splitting rows and columns, deleting duplicate data, normalizing data, and integrating data to obtain voltage and capacity data during battery pulse discharge. The expression for normalizing historical data in the battery health database is as follows: ； in, This indicates unnormalized voltage data; and This represents the maximum and minimum voltage values ​​at different cycle numbers within the same time period.

4. The battery health state prediction method based on the DC internal resistance of a lithium-ion battery according to any one of claims 1, characterized in that, The process of predicting battery life based on the current battery health status assessment results includes: Predict the battery's capacity degradation trajectory based on early battery capacity data; Based on the battery's capacity degradation trajectory, predict the battery's end-of-life and current cycle count, and then predict the remaining battery life based on the end-of-life and current cycle count.

5. The battery health state prediction method based on the DC internal resistance of a lithium-ion battery according to claim 4, characterized in that, The process of predicting the battery's capacity degradation trajectory includes: A battery capacity degradation trajectory prediction model is constructed, which includes a self-attention layer, a feedforward neural network layer, and a battery capacity degradation trajectory prediction layer. The training set is input into the battery capacity degradation trajectory prediction model, and several rounds of parameter iteration optimization training are set to optimize the parameters of the battery capacity degradation trajectory prediction model. The self-attention layer extracts battery capacity features from the training set, the feedforward neural network layer processes the extracted battery capacity features point by point, and the battery capacity degradation trajectory prediction layer predicts the battery capacity degradation trajectory based on the battery capacity features after point-by-point processing and outputs the prediction results. The prediction results are tested using a test set, and a loss function is set to optimize the parameters of the battery capacity degradation trajectory prediction model after each training round. When the preset iterative training rounds end or the loss function converges, the trained battery capacity degradation trajectory prediction model is obtained.

6. The battery health state prediction method based on the DC internal resistance of a lithium-ion battery according to claim 5, characterized in that, The self-attention layer uses a self-attention mechanism to extract battery capacity features from the training set, expressed as: ； The feedforward neural network layer processes the output of the self-attention layer point by point, as expressed in the following expression: ； The battery capacity degradation trajectory prediction layer predicts the battery capacity degradation trajectory based on the point-by-point processed battery capacity characteristics. The expression is as follows: ； The formula for predicting the remaining battery life based on the battery's end-of-life and current cycle count is: RUL = Battery Life - Number of Current Charge-Discharge Cycles; in, For querying the matrix, The key matrix, It is a value matrix; Let x be the dimension of the key vector, and let x represent the input feature. and Both represent the weight matrix of the feedforward neural network. and RUL represents the bias vector of the feedforward neural network, which is the final calculated remaining battery life.

7. A battery health state prediction system based on the DC internal resistance of a lithium-ion battery, the system being based on the method described in any one of claims 1-6, characterized in that, The system includes: The data acquisition module is used to collect pulse voltage data of a single lithium-ion battery throughout its entire life cycle. The data processing module is used to preprocess the pulse voltage data; The data extraction module is used to extract the battery voltage rebound characteristic curve based on the preprocessed pulse voltage data. The status assessment module is used to assess the current battery health status based on the extracted battery voltage rebound characteristic curve and output the assessment results. The lifespan prediction module is used to predict battery lifespan based on the current battery health status assessment results and output the prediction results.

Images