Method and System for Predicting Remaining Useful Life of Lithium-Ion Batteries Based on Multidimensional Data Hybrid Deep Neural Network

Abstract

Method and system for predicting remaining useful life of lithium-ion battery based on multi-dimensional data hybrid deep neural network, which relates to the technical field of lithium-ion batteries. It solves the problem that existing methods for predicting the remaining useful life of lithium-ion batteries based on deep learning basically require data of multiple battery cycles, posing a great challenge to the data storage of the battery management system. The method includes: conducting charge and discharge aging experiments on the battery to obtain battery performance indicators and observation data, establishing an original lithium battery aging data set and performing preprocessing; organizing and extracting the data set through feature engineering; building a hybrid deep neural network model; training the hybrid deep neural network model using k-fold cross-validation; when the k-fold cross-validation training is completed, loading the model and calculating the output of each fold model on the test set data, and taking the average for predicting the remaining useful life of the lithium-ion battery. It is applied to the field of new energy vehicles.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion batteries, and in particular to a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network. Background Art

[0002] Due to its characteristics such as high energy density, wide operating temperature range, long cycle life, and environmental friendliness, lithium-ion batteries have become a widely used power source in new energy vehicles. Therefore, the performance of lithium-ion batteries directly determines the core indicators such as the driving range, charging time, and safety performance of new energy vehicles, and is one of the important influencing factors for the development of new energy vehicles. However, during the long-term operation of lithium-ion batteries, with the increase in the number of charge and discharge cycles and the change in operating temperature, problems such as battery aging and performance degradation will gradually occur. When the performance degrades to a certain extent, various safety problems such as the collapse of the internal structure of the battery and local short circuits will lead to serious safety accidents. Therefore, during the normal operation of the battery, it is necessary to accurately estimate the remaining useful life of the battery to ensure the safe and stable operation of the battery. At present, although researchers have done a lot of work in battery state assessment, prediction, and fault diagnosis, due to the complex working mechanism of lithium-ion batteries and their strong non-linear characteristics, the main electrochemical reactions inside them, such as solid-phase diffusion, liquid-phase diffusion, solid-liquid interface charge transfer, and the growth and decomposition of solid electrolyte interface films, cannot be accurately described by ordinary differential equations. Moreover, the operating conditions of new energy vehicles are complex and variable, and the mutual coupling of various external factors such as electricity, heat, and mechanics will also have an inestimable impact on the battery, resulting in a serious impact on the accurate estimation of the remaining useful life of the battery. Therefore, it is often difficult to achieve high-precision, wide-adaptability, and high-robustness life prediction of lithium-ion batteries. Generally speaking, the prediction of the remaining useful life (RUL) of lithium-ion batteries can be divided into model-based methods and data-driven methods according to technical methods.

[0003] Model-based methods usually require highly accurate models, are easily interfered by external conditions, and have a relatively slow response to load changes. Therefore, these methods are limited by the complexity of the model, limited robustness, dynamic adaptability, and prediction accuracy. In addition, the constructed battery aging models are usually only applicable to specific application environments or battery types, with limited application ranges. More critically, the performance of these methods in predicting the early life of lithium-ion batteries is usually poor.

[0004] The data-driven model estimates the state of health of the battery by collecting signals such as voltage and current from the battery management system and using advanced algorithms such as Gaussian process regression, support vector machines, and machine learning to establish a non-linear relationship model between the data and the battery state. The advantage of this method is that it does not rely on the complex physical processes of the battery aging mechanism. As an extended field of machine learning, deep learning can automatically extract effective features from large-scale datasets by constructing a multi-layer neural network architecture, showing excellent fitting ability for complex systems. In recent years, deep learning technology has been increasingly valued for its superiority in the prediction research of lithium-ion battery life. In particular, the battery capacity degradation data exhibits non-linear time series characteristics. As an effective tool for processing such data and predicting future sequences, the recurrent neural network (RNN) has been verified for its applicability and efficiency by many researchers. As an efficient feature extraction technology, the convolutional neural network (CNN) simplifies the complexity of feature extraction compared with RNN and its derivative algorithms. Through its unique structure, CNN can effectively alleviate the overfitting problem, especially when the data volume is limited or the capacity regeneration phenomenon is encountered. Therefore, CNN has become a commonly used method in the field of battery life prediction and has been widely applied in related research in recent years.

[0005] However, most of the deep learning-based methods basically require data of multiple battery cycles, which poses a great challenge to the data storage of the battery management system, especially when continuously monitoring and real-time processing a large amount of battery data. This requires more efficient data compression and processing strategies to optimize storage and computing resources. Summary of the Invention

[0006] In view of the problem that most of the existing methods for predicting the remaining useful life of lithium-ion batteries based on deep learning basically require data of multiple battery cycles, which poses a great challenge to the data storage of the battery management system, the present invention proposes a method for predicting the remaining useful life of lithium-ion batteries based on a multi-dimensional data hybrid deep neural network. The method includes:

[0007] Step S1: Conduct charge and discharge aging experiments on the battery to obtain the performance indicators of each cycle of the battery and the battery cycle observation data, establish the original lithium battery aging dataset, and preprocess the aging dataset;

[0008] Step S2: Divide the preprocessed dataset into a training set for model training and a test set for model evaluation, and organize and extract the dataset through feature engineering;

[0009] Step S3: Build a hybrid deep neural network model based on a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, and a Transformer;

[0010] Step S4: Use k-fold cross-validation to train the hybrid deep neural network model. After the loss function of the training model no longer decreases continuously, save the model as the optimal model for this fold.

[0011] Step S5: After the k-fold cross-validation training is completed, load the model and calculate the output of each fold of the model on the test set data. Take the average and use it to predict the remaining useful life of the lithium-ion battery.

[0012] Furthermore, a preferred method is also proposed. In step S1, the performance indicators of each cycle of the battery include discharge capacity, charging time, and internal resistance; the battery cycle observation data includes voltage, current, temperature, and capacity during battery charge and discharge cycles.

[0013] Furthermore, a preferred method is also proposed. In step S1, the charge and discharge aging experiment is a multi-stage constant-current charging and constant-current discharging process at a constant temperature, including: first charging the battery to a specific state of charge with a constant current, then charging the battery in the state of charge to the cut-off voltage with another constant current, and then charging at a constant voltage until the cut-off current of the battery, and discharging to the cut-off voltage with the same current.

[0014] Furthermore, a preferred method is also proposed. In step S1, the preprocessing includes: using linear interpolation method to remove outliers and normalizing the data without outliers.

[0015] Furthermore, a preferred method is also proposed. In step S2, the feature engineering includes:

[0016] Interpolate the current, voltage, temperature, and capacity curves of each cycle linearly according to the time series into 4 curves with 100 data points each. The 4 curves form a 4×100 two-dimensional data matrix; combine the 3 values of discharge capacity, charging time, and battery internal resistance of each cycle into a 1×3 one-dimensional data matrix.

[0017] Furthermore, a preferred method is also proposed. Step S3 includes:

[0018] Input the one-dimensional data matrix into a 3-layer one-dimensional convolutional neural network to extract the battery aging information in the one-dimensional scalar features, and then process the output of the one-dimensional convolutional neural network through a fully connected layer;

[0019] Input the two-dimensional data matrix into 3 layers of two-dimensional convolutional neural networks connected in series, and further extract features from the output of the two-dimensional convolutional neural network through a Transformer network;

[0020] Integrate the outputs of the Transformer network and the fully connected layer, and then output through two fully connected layers. The obtained output is the estimated value of the remaining useful life of the lithium-ion battery;

[0021] The processing flow of the Transformer network is as follows:

[0022] The data first passes through the position encoding module to retain the time step information of the data;

[0023] The time series data passes through the multi-head self-attention mechanism to calculate the correlation of each time step in the sequence. The self-attention mechanism uses a weight matrix to perform weighted aggregation on the input features of each time step.

[0024] Furthermore, a preferred method is also proposed. The k-fold cross-validation in step S4 refers to using 10-fold cross-validation.

[0025] Based on the same inventive concept, the present invention also proposes a lithium-ion battery remaining useful life prediction system based on a multi-dimensional data hybrid deep neural network. The system includes:

[0026] A charge and discharge aging experiment unit for performing charge and discharge aging experiments on the battery to obtain the performance indicators of each cycle of the battery and the battery cycle observation data, establishing an original aging data set of the lithium battery, and preprocessing the aging data set;

[0027] A data extraction unit for dividing the preprocessed data set into a training set for model training and a test set for model evaluation, and organizing and extracting the data set through feature engineering;

[0028] A model construction unit for building a hybrid deep neural network model based on a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, and a Transformer;

[0029] A training unit for training the hybrid deep neural network model using k-fold cross-validation. After the loss function of the training model no longer decreases continuously, the model is saved as the optimal model for this fold;

[0030] A prediction unit for, after the k-fold cross-validation training is completed, loading the model and calculating the output of each fold of the model on the test set data, and taking the average for predicting the remaining useful life of the lithium-ion battery.

[0031] Based on the same inventive concept, the present invention also proposes a computer device including a memory and a processor. A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network as described in any one of the above.

[0032] Based on the same inventive concept, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network as described in any one of the above.

[0033] The beneficial effects of the present invention are as follows:

[0034] The present invention solves the problem that existing methods for predicting the remaining useful life of lithium-ion batteries based on deep learning basically require data of multiple battery cycles, which poses a great challenge to the data storage of the battery management system.

[0035] The present invention provides a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network, and designs a hybrid deep neural network model based on a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, and a Transformer, which can effectively utilize one-dimensional scalar features and two-dimensional vector features. Through forward propagation, the two-dimensional convolutional neural network effectively extracts the time series data features of the battery, and the Transformer uses the self-attention mechanism to effectively capture the global dependence relationship by calculating the correlation between each time step in the input sequence, further enhancing the feature expression ability. The one-dimensional convolutional neural network can directly capture the key information related to battery aging from the key performance parameters of the battery. This multi-branch design ensures the synergy between different data dimensions and fully excavates potential aging features. The method of multi-branch and multi-dimensional data fusion can more accurately extract the features related to the remaining life of the battery, so as to accurately predict the remaining life of the battery, and thus assist the battery management system to effectively manage and control the battery.

[0036] Compared with the traditional method that relies on multiple cycle data, the method proposed by the present invention can more efficiently identify and extract the aging features of lithium-ion batteries because it adopts a multi-branch multi-dimensional hybrid neural network. It can accurately predict the remaining useful life of the battery only with the charge and discharge data of one cycle, thus significantly reducing the demand for data storage of the battery management system. This design not only improves the data processing efficiency but also enhances the adaptability of the model, enabling it to maintain high performance under limited data volume.

[0037] The present invention is applied to the field of new energy vehicles. Description of the Drawings

[0038] Figure 1 Flowchart of a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network according to Embodiment 1;

[0039] Figure 2Schematic diagram of the predicted remaining useful life of the shorter-life lithium-ion battery described in Embodiment XI, where the horizontal axis represents the true life and the vertical axis represents the predicted life;

[0040] Figure 3 Schematic diagram of the predicted remaining useful life of the longer-life lithium-ion battery described in Embodiment XI, where the horizontal axis represents the true life and the vertical axis represents the predicted life;

[0041] Figure 4 Schematic diagram of the error distribution of the predicted remaining useful life of all lithium-ion batteries described in Embodiment XI, where the horizontal axis represents the test set number and the vertical axis represents the prediction error;

[0042] Figure 5 Schematic diagram of the predicted remaining useful life of the shorter-life lithium-ion battery in the comparative example described in Embodiment XI, where the horizontal axis represents the true life and the vertical axis represents the predicted life;

[0043] Figure 6 Schematic diagram of the predicted remaining useful life of the longer-life lithium-ion battery in the comparative example described in Embodiment XI, where the horizontal axis represents the true life and the vertical axis represents the predicted life;

[0044] Figure 7 Schematic diagram of the error distribution of the predicted remaining useful life of all lithium-ion batteries in the comparative example described in Embodiment XI, where the horizontal axis represents the test set number and the vertical axis represents the prediction error. Specific Embodiments

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention.

[0046] Embodiment 1. Refer to Figure 1 Describe this embodiment. A method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network, the method includes:

[0047] Step S1: Conduct charge and discharge aging experiments on the battery to obtain the performance indicators of each cycle of the battery and the battery cycle observation data, establish an original lithium battery aging data set, and preprocess the aging data set;

[0048] Step S2: Divide the preprocessed data set into a training set for model training and a test set for model evaluation. The batteries can be randomly assigned to the training set and the test set in a ratio of 3:1 according to the battery life, and the data set is sorted and extracted through feature engineering;

[0049] Step S3: Build a hybrid deep neural network model based on one-dimensional convolutional neural network, two-dimensional convolutional neural network and Transformer, and summarize and output information through a fully connected layer;

[0050] Step S4: Use k-fold cross validation to train the hybrid deep neural network model. When the loss function of the training model no longer continues to decrease, save the model and record it as the optimal model of the fold;

[0051] Step S5: After the k-fold cross-validation training is completed, the model is loaded and the output of each fold of the model on the test set data is calculated, and the average is taken to predict the remaining service life of the lithium-ion battery.

[0052] Compared with the traditional method that relies on multiple cycle data, the method for predicting the remaining service life of lithium-ion batteries based on a multi-dimensional data hybrid deep neural network proposed in this embodiment can more efficiently identify and extract the aging characteristics of lithium-ion batteries because it uses a multi-branch multi-dimensional hybrid neural network. It only needs one cycle of charge and discharge data to accurately predict the remaining service life of the battery, thereby significantly reducing the demand for data storage in the battery management system. This design not only improves data processing efficiency, but also enhances the adaptability of the model, enabling it to maintain high performance when the amount of data is limited.

[0053] At the same time, the multi-branch, multi-dimensional data fusion hybrid deep neural network designed in this embodiment can more efficiently identify and extract the aging characteristics of lithium-ion batteries. Through one-dimensional convolutional neural networks and two-dimensional convolutional neural networks, the key performance parameters and time series data of the battery can be processed respectively, effectively improving the feature expression ability of the hybrid deep neural network model.

[0054] Furthermore, the method for predicting the remaining service life of lithium-ion batteries proposed in this embodiment combines the advantages of convolutional neural networks (CNN) and Transformer. Transformer captures the correlation between each time step in the input sequence through the self-attention mechanism, enhancing the understanding of global dependencies. This enables the model to maintain high performance even when the amount of data is limited. The stability and accuracy of the hybrid deep neural network model can be ensured through k-fold cross-validation training and multi-branch design. The optimal model of each fold can integrate the characteristics of different data and improve the reliability of the prediction. The multi-dimensional data fusion design ensures the synergy between different data dimensions, and can fully explore the potential features related to battery aging, thereby improving the accuracy of the prediction of the remaining service life.

[0055] Embodiment 2. This embodiment further defines a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in Embodiment 1. In step S1, the performance indicators of each battery cycle include discharge capacity, charging time, and internal resistance; the battery cycle observation data includes voltage, current, temperature, and capacity during battery charge and discharge cycles.

[0056] In this embodiment, performance indicators such as discharge capacity, charging time, and internal resistance are introduced, enabling the model to more comprehensively reflect the health status and performance degradation of the battery. These indicators help to more accurately capture the aging characteristics of the battery. The battery cycle observation data includes multiple observation data such as voltage, current, temperature, and capacity, which can provide detailed information about the battery under different working conditions and enhance the model's understanding of the battery behavior. By comprehensively considering various performance indicators and observation data, the model can identify and capture more complex characteristic relationships. For example, changes in voltage and temperature may be directly related to changes in internal resistance, and comprehensive analysis helps to reveal the potential mechanism of battery aging. By considering different charge and discharge cycle conditions (such as different voltages and temperatures), the model has stronger adaptability and can maintain good prediction performance under different working environments and conditions. This is particularly important for battery management systems in practical applications.

[0057] Embodiment 3. This embodiment further defines a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in Embodiment 1. In step S1, the charge and discharge aging experiment is a multi-stage constant-current charging and constant-current discharging process at a constant temperature, including: first charging the battery to a specific state of charge using a constant current, then charging the battery at the state of charge to the cut-off voltage using another constant current, and then charging at a constant voltage until the cut-off current of the battery, and discharging to the cut-off voltage with the same current.

[0058] In this embodiment, the state of charge (SOC) of the battery represents the ratio of the remaining capacity of the battery after being used for a period of time or left unused for a long time to the capacity of its fully charged state. The specific state of charge described in this embodiment is 60%, 70%, and 80%. These three specific states of charge.

[0059] In this embodiment, through the processes of multi-stage charging and constant-current discharging, the charging state of the battery can be precisely controlled, providing richer battery behavior data. This helps to more accurately model the performance degradation characteristics of the battery and improve the accuracy of prediction. The multi-stage charging and discharging processes can better capture the dynamic characteristics of the battery under different working conditions. By analyzing the performance of the battery at different states of charge, the key factors affecting the battery life can be identified. Different charging currents and states of charge are adopted, enabling the experiment to evaluate the impact of different charging strategies on the battery performance, thereby providing a basis for optimizing the battery usage and charging schemes. At the same time, the hybrid deep neural network model has the ability of adaptive learning and can continuously optimize the model parameters with the input of new data. This flexibility enables the prediction model to adapt to different types and brands of lithium-ion batteries, improving the universality of its application. Through the precise prediction method, the dependence on a large amount of experimental data can be reduced, and the R & D and testing costs can be lowered. The prediction of the battery life can be carried out in the design stage, thus saving time and resources.

[0060] Embodiment 4. This embodiment further limits a method for predicting the remaining service life of a lithium-ion battery based on a hybrid deep neural network with multi-dimensional data described in Embodiment 1. The preprocessing in step S1 includes: using the linear interpolation method to remove outliers and normalizing the data without outliers, that is, standardizing the data without outliers to the "0-1" interval.

[0061] In this embodiment, linear interpolation can effectively fill in the missing values or outliers in the data, ensuring the integrity of the data set. This helps to avoid the bias introduced by abnormal data, thereby improving the prediction accuracy of the model. Removing outliers can improve the overall quality of the data set, making the data input into the model more consistent and reliable, and contributing to enhancing the learning effect and generalization ability of the model. Further, normalizing the data without outliers to the "0-1" interval enables the data of different features to be on the same scale, avoiding the model training problems caused by the differences in the feature value ranges. This is particularly important for deep neural networks because it can accelerate the convergence speed and improve the training efficiency. Through the preprocessing steps, the sensitivity of the model to the fluctuations and noises of the input data is reduced, thereby enhancing the robustness of the model in practical applications and improving the stability of the prediction results.

[0062] Embodiment 5. This embodiment further limits a method for predicting the remaining service life of a lithium-ion battery based on a hybrid deep neural network with multi-dimensional data described in Embodiment 1. The feature engineering in step S2 includes:

[0063] Interpolate the current, voltage, temperature, and capacity curves for each cycle into 4 curves with 100 data points according to the time series. The 4 curves form a 4×100 two-dimensional data matrix. Combine the discharge capacity, charge time, and internal resistance of each cycle into a 1×3 one-dimensional data matrix.

[0064] In this embodiment, interpolating the current, voltage, temperature, and capacity curves of each cycle to the same number of data points (100) helps to eliminate data inconsistencies between different cycles, standardize the input data, and facilitate subsequent processing and analysis. The linear interpolation method used in this embodiment can effectively retain the trend and changes of the original data, enabling the hybrid deep neural network model to capture the dynamic characteristics of the battery during charge and discharge and avoid information loss caused by sparse data points. Further, by forming a 4×100 two-dimensional data matrix from the 4 curves, the hybrid deep neural network model can simultaneously consider the mutual influence of multiple features, improving the understanding of battery performance. The introduction of such multi-dimensional data can enhance the learning ability of the hybrid deep neural network model for complex relationships. Combining the discharge capacity, charge time, and internal resistance into a 1×3 one-dimensional data matrix makes the subsequent feature extraction and modeling processes more concise and efficient, helps reduce computational complexity, and improves the training speed of the hybrid deep neural network model.

[0065] At the same time, combining time-series data with static features (such as discharge capacity, charge time, and internal resistance) can enable the hybrid deep neural network model to consider both dynamic changes and static characteristics simultaneously, thereby enhancing its learning ability and generalization ability. Deep neural networks usually require high-dimensional feature inputs. Organizing data in such a matrix form can better meet the input requirements of the hybrid deep neural network model and improve the performance of the model.

[0066] Further, data organized in matrix form can be more conveniently visualized, which helps analyze the performance of the battery under different charge and discharge states.

[0067] Embodiment 6: This embodiment further limits a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in Embodiment 5. The step S3 includes:

[0068] Input the one-dimensional data matrix into a 3-layer one-dimensional convolutional neural network to extract the battery aging information in the one-dimensional scalar features, and then process the output of the one-dimensional convolutional neural network through a fully connected layer;

[0069] Input the two-dimensional data matrix into 3-layer two-dimensional convolutional neural networks connected in series, and further extract features from the output of the two-dimensional convolutional neural network through a Transformer network;

[0070] After integrating the outputs of the Transformer network and the fully connected layer, and then passing them through two fully connected layers for output, the obtained output is the estimated remaining useful life of the lithium-ion battery;

[0071] The processing flow of the Transformer network is as follows:

[0072] The data first passes through the position encoding module to retain the time step information of the data;

[0073] The time series data passes through the multi-head self-attention mechanism to calculate the correlation of each time step in the sequence. The self-attention mechanism weights and aggregates the input features of each time step through a weight matrix. This mechanism ensures that the model can pay attention to the long-range dependencies between any two points in the time series. Subsequently, through the feed-forward neural network and normalization operations, deep features are extracted and redundant noise is reduced. After multiple layers of this structure, the model finally outputs rich feature representations for subsequent life prediction.

[0074] The hybrid deep neural network model described in this embodiment is a multi-branch feature extraction structure. This structure effectively captures the multi-dimensional features during the aging process of the lithium-ion battery through the splitting process of one-dimensional and two-dimensional data. Specifically: The hybrid deep neural network model contains two branches:

[0075] Branch 1: Processes the key performance indicators (internal resistance, discharge capacity, and charging time) of the battery through a one-dimensional convolutional neural network (CNN1d) to extract the aging feature parameters at different cycle scales. The design of this branch enables the network to quickly analyze the performance changes of the battery at different cycle stages.

[0076] Branch 2: Processes the cycle observation data (current, voltage, temperature, and capacity curve) through a two-dimensional convolutional neural network (CNN2d). This part of the data reflects the more complex spatio-temporal features that can be reflected within the cycle during the battery aging process. CNN2d can very effectively extract the local features of these data. The Transformer layer after CNN2d can further enhance the ability to capture the global features of the data through its self-attention mechanism, pay attention to the important features at different cycle stages during the battery aging process, and thus effectively improve the accuracy and robustness of battery life prediction.

[0077] Therefore, the hybrid deep neural network model proposed in this embodiment can achieve the life prediction of the lithium-ion battery with only one cycle of data input through its powerful feature extraction ability for multi-dimensional data. Compared with the traditional method, the method proposed in this embodiment reduces the dependence on a large amount of historical data and greatly improves the calculation efficiency and flexibility in practical applications.

[0078] Therefore, the core innovation of this embodiment does not lie in the changes at the basic network level, but in the design of the multi-branch feature extraction architecture, so as to utilize the powerful feature extraction capabilities of CNN and Transformer to achieve more efficient and accurate life prediction.

[0079] Combined with Embodiment 1 to illustrate the advantages of this embodiment, traditional methods usually require data of multiple battery cycles for training, which poses a great challenge to the data storage of the battery management system. The hybrid deep neural network model designed in this embodiment can achieve accurate prediction of the remaining useful life of the battery only with the charge and discharge data of one cycle, significantly reducing the data storage requirements.

[0080] Embodiment 7: This embodiment further limits the method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in Embodiment 1. The k-fold cross-validation in step S4 refers to using 10-fold cross-validation.

[0081] In this embodiment, by dividing the data into 10 subsets and using 9 training sets and 1 validation set each time, it can effectively reduce the dependence of the hybrid deep neural network model on specific data partitioning, and improve the stability and reliability of the prediction results. Through 10-fold cross-validation, the training data can be utilized more fully, ensuring that each data point can play a role in model training and validation, and improving the data utilization efficiency. Through multiple trainings and validations, the performance of the hybrid deep neural network model on unseen data can be better evaluated, thereby reducing the risk of model overfitting and ensuring its generalization ability. At the same time, the 10-fold cross-validation adopted in this embodiment is particularly effective in the case of a small sample size, which can maximize the use of limited data resources and help build a more accurate prediction model.

[0082] Embodiment 8: A system for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in this embodiment, the system includes:

[0083] A charge and discharge aging experiment unit for conducting charge and discharge aging experiments on the battery to obtain performance indicators of each cycle of each battery and battery cycle observation data, establishing an original aging data set of the lithium battery, and preprocessing the aging data set;

[0084] A data extraction unit for dividing the preprocessed data set into a training set for model training and a test set for evaluating the model, and sorting and extracting the data set through feature engineering;

[0085] A model construction unit for building a hybrid deep neural network model based on a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, and a Transformer, and summarizing and outputting information through a fully connected layer;

[0086] A training unit for training the hybrid deep neural network model using k-fold cross-validation. After the loss function of the training model no longer decreases continuously, the model is saved as the optimal model for this fold.

[0087] A prediction unit for, after the k-fold cross-validation training is completed, loading the model and calculating the outputs of each fold of the model on the test set data, and taking the average for predicting the remaining useful life of the lithium-ion battery.

[0088] Embodiment 9. A computer device described in this embodiment includes a memory and a processor. A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in any one of Embodiments 1 to 7.

[0089] Embodiment 10. A computer-readable storage medium described in this embodiment has a computer program stored thereon. When the computer program is run by a processor, it executes the steps of a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in any one of Embodiments 1 to 7.

[0090] Embodiment 11. Refer to Figures 2 to 7 This embodiment. This embodiment provides a specific example for a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in Embodiment 1, and is also used to explain Embodiments 2 to 7. Specifically:

[0091] Step 1: At (30 ± 1) °C, perform charge-discharge aging experiments on lithium iron phosphate 18650 type lithium-ion batteries. The charging protocol is as follows: First, charge at rates of 3C, 4C, 5C, and 6C to 60%, 70%, 80% of the rated capacity. Among them, 10 batteries at each charging rate are charged to three different states of charge, totaling 120 batteries. Then, charge at a constant current of 1C to 3.65V, and then charge at a constant voltage until the cut-off current is 1 / 20C; the discharging protocol is: all batteries are discharged at a rate of 4C to 2.0V. The battery termination condition is that the rated capacity drops to 80% of the initial capacity. At the end of each cycle, record various battery performance indicators such as the discharge capacity, charging time, internal resistance, etc. of the battery in this cycle, and collect the change curves of parameters such as battery current, voltage, temperature, capacity, etc. over time during the cycle to obtain the observed data of each battery in each cycle, and establish the original aging dataset of the lithium battery. And use the mean value before and after to replace the outliers collected during the test process, and at the same time perform normalization preprocessing on the non-abnormal data.

[0092] Step 2: Based on the life values of each battery, divide the dataset preprocessed in Step 1 into a training set for model training and a test set for model evaluation according to a ratio of 3:1. Then linearly interpolate the four curves of current, voltage, temperature, and capacity in each cycle into four curves with 100 data points according to the time series. The four curves form a 4×100 two-dimensional data matrix. Combine the three values of discharge capacity, charging time, and battery internal resistance in each cycle into a 1×3 one-dimensional data matrix. The remaining useful life of the battery in each cycle is used as the label.

[0093] Step 3: Build a hybrid deep neural network model based on one-dimensional convolutional neural network, two-dimensional convolutional neural network, and Transformer, and summarize and output information through a fully connected layer. The specific method is as follows:

[0094] Input the two-dimensional data matrix obtained in Step 2 into three consecutive two-dimensional convolutional neural networks, and further extract features from the output of the two-dimensional convolutional neural network through the Transformer network. Input the one-dimensional data matrix obtained in Step 2 into three one-dimensional convolutional neural networks, and then process the output of the one-dimensional convolutional neural network through a fully connected layer. Integrate the outputs of the above Transformer and fully connected layer, and then pass through two fully connected layers. The output of the hybrid deep neural network model is the estimated value of the remaining useful life of the lithium-ion battery.

[0095] Step 4: Use 10-fold cross-validation to train the constructed hybrid deep neural network model. After the loss function of the training model no longer decreases continuously, save the model as the optimal model for this fold. After the 10-fold cross-validation training is completed, load the model and calculate the output of each fold model on the test set data, and take the average for predicting the remaining useful life of the lithium-ion battery. The life prediction value is as shown in Attachment Figure 2 、Attachment Figure 3 as shown, and the error distribution diagram is as shown in Attachment Figure 4 as shown.

[0096] In this embodiment, Gaussian process regression is used for prediction of the remaining useful life of the lithium-ion battery, which specifically includes the following steps:

[0097] Step 1: Steps 2 are exactly the same as those in the embodiment, using the same dataset and the same division of the training set and test set.

[0098] Step 3: Build a lithium-ion battery life prediction model based on Gaussian process regression. The specific method is as follows: After reducing the dimension of the two-dimensional data matrix obtained in Step 2, connect it with the one-dimensional data matrix and input it into the Gaussian process regression model. The output of the model is the estimated value of the remaining useful life of the lithium-ion battery.

[0099] Step 4: Train the constructed Gaussian process regression model. After the loss of the training model no longer decreases continuously, extract the best-performing model for predicting the remaining useful life of the lithium-ion battery. The life prediction values are shown in Appendix Figure 5 , Appendix Figure 6 as shown, and the error distribution diagram is shown in Appendix Figure 7 as shown.

[0100] According to the prediction results of the present invention and the comparative examples, it can be clearly seen that the method for predicting the remaining useful life of the lithium-ion battery proposed by the present invention significantly improves the accuracy and robustness of the prediction of the remaining useful life of the lithium-ion battery. It can more accurately extract the features related to the remaining life of the battery, thereby accurately predicting the remaining life of the battery, and assisting the battery management system to effectively manage and control the battery.

[0101] Those skilled in the art should understand that the embodiments of the present disclosure may be provided as a method, a system, or a computer program product. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks or multiple blocks. Figure 1 These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks

[0103] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the steps specified in one process or multiple processes and / or blocks Figure 1 one process or multiple processes and / or blocks Figure 1 steps for the functions specified in one block or multiple blocks.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure rather than to limit the scope of its protection. Although the present disclosure has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: after reading the present disclosure, those skilled in the art can still make various changes, modifications or equivalent replacements to the specific implementation manners of the invention, but these changes, modifications or equivalent replacements are all within the scope of protection of the pending claims of the disclosure.

Claims

1. A method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network, characterized in that, The method includes the following steps: Step S1: Conduct charge and discharge aging experiments on the battery to obtain the performance indicators of each cycle of the battery and the battery cycle observation data, establish the original lithium battery aging dataset, and preprocess the aging dataset; Step S2: Divide the preprocessed dataset into a training set for model training and a test set for model evaluation, and organize and extract the dataset through feature engineering; Step S3: Build a hybrid deep neural network model based on one-dimensional convolutional neural network, two-dimensional convolutional neural network and Transformer; Step S4: Use k-fold cross-validation to train the hybrid deep neural network model. After the loss function of the training model no longer decreases continuously, save the model as the optimal model for this fold; Step S5: After the k-fold cross-validation training is completed, load the model and calculate the output of each fold of the model on the test set data, and take the average to predict the remaining service life of the lithium-ion battery; In step S1, the performance indicators of each cycle of the battery include discharge capacity, charging time, and internal resistance; the battery cycle observation data includes voltage, current, temperature, and capacity during battery charge and discharge cycles; The feature engineering in step S2 includes: Interpolate the current, voltage, temperature, and capacity curves of each cycle linearly according to the time series into 4 curves with 100 data points, and the 4 curves form a 4×100 two-dimensional data matrix; combine the discharge capacity, charging time, and battery internal resistance values of each cycle into a 1×3 one-dimensional data matrix; Step S3 includes: Input the one-dimensional data matrix into a 3-layer one-dimensional convolutional neural network to extract the battery aging information in the one-dimensional scalar features, and then process the output of the one-dimensional convolutional neural network through a fully connected layer; Input the two-dimensional data matrix into 3 layers of two-dimensional convolutional neural networks connected in series, and further extract features from the output of the two-dimensional convolutional neural network through the Transformer network; Integrate and process the outputs of the Transformer network and the fully connected layer, and then output through two fully connected layers. The obtained output is the estimated value of the remaining service life of the lithium-ion battery; The processing flow of the Transformer network is: The data first passes through the position encoding module to retain the time step information of the data; The time series data passes through the multi-head self-attention mechanism to calculate the correlation of each time step in the sequence. The self-attention mechanism uses a weight matrix to weightedly summarize the input features of each time step.

2. A method for predicting the remaining service life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network according to claim 1, wherein In step S1, the charge and discharge aging experiment is a multi-stage constant current charging and constant current discharging process at a constant temperature, including: first charging the battery to a specific state of charge with a constant current, then charging the battery in the state of charge to the cut-off voltage with another constant current, and then charging at a constant voltage until the cut-off current of the battery, and discharging to the cut-off voltage with the same current.

3. A method for predicting the remaining service life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network according to claim 1, characterized in that The preprocessing in step S1 includes: using the linear interpolation method to remove outliers and normalizing the data without outliers.

4. A method for predicting the remaining service life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network according to claim 1, characterized in that The k-fold cross-validation in step S4 refers to using 10-fold cross-validation.

5. A remaining useful life prediction system for lithium-ion batteries based on a multi-dimensional data hybrid deep neural network, characterized in that, The system is implemented based on a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in claim 1. The system includes: A charge-discharge aging experiment unit, which is used to conduct charge-discharge aging experiments on the battery to obtain the performance indicators of each cycle of the battery and the battery cycle observation data, establish an original aging data set of the lithium battery, and preprocess the aging data set; A data extraction unit, which is used to divide the preprocessed data set into a training set for model training and a test set for evaluating the model, and organize and extract the data set through feature engineering; A model construction unit, which is used to build a hybrid deep neural network model based on a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, and a Transformer; A training unit, which is used to train the hybrid deep neural network model using k-fold cross-validation. After the loss function of the training model no longer decreases continuously, save the model as the optimal model for this fold; A prediction unit, which is used to load the model and calculate the output of each fold of the model on the test set data after the k-fold cross-validation training is completed, and take the average for predicting the remaining useful life of the lithium-ion battery.

6. A computer device, characterized in that: It includes a memory and a processor. A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is run by the processor, it executes the steps of a method for predicting the remaining useful life of a lithium-ion battery based on a multi-dimensional data hybrid deep neural network described in any one of claims 1-4.

Images