A lithium ion battery remaining life estimation method, device and storage medium

Abstract

The application discloses a kind of lithium ion battery remaining life estimation method, device and storage medium, its method includes: the state information corresponding to different SOC of different charge-discharge cycle of target lithium ion battery is collected;After pre-processing state information, two-dimensional table is generated with SOC and the number of charge cycles corresponding to two coordinate axes of two-dimensional coordinate system, and a plurality of table data are obtained by processing two-dimensional table using state information;Table data is input into pre-trained estimation neural network model, and the estimation result of the remaining life of lithium ion battery is output.The application fully considers the factors such as the environment of battery work, charge-discharge strategy and charge-discharge intensity, utilizes a new data processing means and corresponding convolutional neural network model, so as to realize the estimation of the remaining life of lithium ion battery in complex working environment.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery technology, specifically relating to a method, apparatus, and storage medium for estimating the remaining life of a lithium-ion battery. Background Technology

[0002] Due to their superior overall performance in terms of cost, energy density, and lifespan, lithium-ion batteries have been widely used in electric vehicles, renewable energy systems, and smart grids. As the most important and expensive component in energy storage units of new energy systems, lithium-ion batteries require careful monitoring and operation. It is well known that the performance of lithium-ion batteries inevitably declines with use. This degradation is a non-linear electrochemical process with extremely complex internal mechanisms, and these mechanisms vary greatly under different operating conditions. Generally, a battery is considered to have reached the end of its lifespan when its capacity drops to 80% of its initial value. Accurately and quickly predicting the remaining battery lifespan is crucial for BMS (Battery Management System) to ensure operational reliability and safety, and is essential for battery manufacturers or battery asset owners to determine battery maintenance strategies.

[0003] In recent years, many methods for predicting the remaining lifespan of batteries have been proposed. For example, CN114839553A discloses a method and apparatus for predicting the remaining lifespan of batteries. This method requires conducting charge-discharge experiments on multiple sample batteries to obtain experimental data from multiple charge-discharge cycles for each sample battery. Based on this data, aging characteristic data for each sample battery is obtained. The aging characteristic data for each sample battery is preprocessed and feature-engineered to form a dataset, which is then divided into a training set and a test set. A combined CNN and LSTM model is constructed, where the fully connected layers of the CNN are removed, and the output is a one-dimensional feature vector. The CNN output is used as the input to the LSTM. The combined CNN and LSTM model is trained using the training set to obtain a battery remaining lifespan predictor, and then tested using the test set. Finally, the remaining lifespan of the battery to be predicted is predicted using the final battery remaining lifespan predictor. Another example is CN113985294A, which provides a method and apparatus for estimating the remaining lifespan of batteries. The method for estimating the remaining battery life includes acquiring historical charge-discharge cycle data of the lithium battery and extracting the discharge time difference from the first voltage to the second voltage and the battery capacity of the forward cycle to form an initial feature vector. Based on the initial feature vector and a pre-trained life prediction model, iterative predictions are performed until the predicted capacity is lower than a preset proportion of the rated capacity, ultimately obtaining the predicted remaining lifespan of the lithium battery. The life prediction model is an Extreme Learning Machine (ELM) model with initialization parameters optimized using a beetle whisker search algorithm. This estimation method improves the accuracy and stability of the ELM, ultimately enhancing the accuracy of the predicted remaining lifespan of the lithium battery.

[0004] Therefore, existing battery remaining life estimation methods use data from the entire charging and / or discharging process as input, or artificially extract features to describe battery life. However, in practical applications, batteries are not always fully charged or discharged. For example, automotive power batteries often have their minimum discharge SOC or maximum charging SOC artificially limited to protect battery life. Furthermore, existing methods typically assume that battery operating conditions and charging / discharging strategies are constant, but such conditions are rare in reality; in most cases, battery charging is random. Therefore, using data from the entire charging and / or discharging process for remaining life estimation in practical applications is demanding and has low feasibility. Models trained using this data also have certain errors when compared to real-world usage environments. In summary, these factors lead to: 1. High data completeness requirements for battery remaining life estimation, often requiring complete data throughout the entire process, which is difficult to obtain in practice; 2. Existing methods are overly idealistic and do not closely match actual operating conditions, resulting in insufficient prediction accuracy.

[0005] To address the aforementioned problems, this invention fully considers factors such as different battery operating environments, different charging and discharging strategies, and different charging and discharging intensities. It proposes a battery remaining life prediction method that aligns with the actual operating conditions of batteries by utilizing a novel data processing technique and a corresponding convolutional neural network model. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a method, device and storage medium for estimating the remaining life of lithium-ion batteries, which fully considers factors such as the battery's working environment, charging and discharging strategy and charging and discharging intensity, and utilizes a novel data processing method and a corresponding convolutional neural network model to achieve the estimation of the remaining life of lithium-ion batteries under complex working environments.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The present invention provides a method for estimating the remaining life of a lithium-ion battery, comprising:

[0009] Step S01: Collect the state information of the continuous SOC value change of the target lithium-ion battery under different charge and discharge cycles; the continuous SOC value change range is 10%-30%; the state information includes the number of cycles, voltage, current, battery temperature, ambient temperature and SOC information in each charge and discharge cycle; the number of cycles is greater than 10.

[0010] Step S02: After preprocessing the status information, a two-dimensional table is generated with the SOC value and the number of charging cycles corresponding to the two coordinate axes of the two-dimensional coordinate system, respectively. The status information is used to process the two-dimensional table to obtain multiple data tables, including voltage data table, current data table and temperature data table.

[0011] Step S03: Input the voltage data table, current data table, and temperature data table into the pre-trained estimation neural network model, and output the estimation result of the remaining life of the lithium-ion battery.

[0012] Preferably, the continuous SOC variation range is 20%, and the number of cycles is 20.

[0013] Preferably, the temperature data is battery temperature information or the difference between battery temperature information and ambient temperature.

[0014] Preferably, the neural network model is trained through the following steps:

[0015] Step S1: Obtain the battery charge / discharge experiment dataset;

[0016] Step S2: After preprocessing the data in the battery charge and discharge experiment dataset, construct the training dataset and validation dataset for the battery charge and discharge characteristic neural network;

[0017] Step S3: Train the pre-built convolutional neural network model using the training dataset, and adjust the hyperparameters in the convolutional neural network model according to the Bayesian optimization algorithm to obtain the trained convolutional neural network model. The hyperparameters include the number of filters in the convolutional layer, the size of the convolutional kernel, the polling size, and the number of neurons in the fully connected layer.

[0018] Step S4: Validate the trained convolutional neural network model using the validation dataset. Determine whether the validation is successful by comparing the error between the estimated result and the actual result output by the convolutional neural network model with the preset allowable error range. If the validation fails, continue to adjust the hyperparameters in the convolutional neural network model until the validation dataset is successfully validated and the estimated neural network model is obtained.

[0019] Preferably, the convolutional neural network model, from input to output, includes: a first 3D convolutional layer, a second 3D convolutional layer, a first 2D convolutional layer, a second 2D convolutional layer, a pooling layer, and a connecting layer, constructed using a residual network. The first 3D convolutional layer initially fuses the features between current and voltage, and current and temperature, outputting an initial fused 3D matrix. The second 3D convolutional layer further fuses the initial fused 3D matrix, outputting a 2D matrix. The outputs of the second 3D convolutional layer and the first 2D convolutional layer are transferred to the residual network. The extracted feature information matrix is ​​transferred to the fully connected layer through skip connections. Pooling layers or dropout layers are used in each connection of the residual network.

[0020] Preferably, the battery charge-discharge experimental dataset is obtained through the following steps:

[0021] Step S11: Randomly divide multiple battery samples with the same specifications into n first-class groups, and adopt n different charging and discharging strategies for each of the n first-class groups.

[0022] Step S12: Divide each first group into m second groups, and assign each of the m second groups to m different ambient temperatures;

[0023] Step S13: Record the number of cycles, voltage, current, battery temperature, ambient temperature, and SOC information of the grouped batteries in each charge-discharge cycle at intervals using the battery BMS, test sensors, or external temperature sensors, thereby collecting m*n sets of full-cycle battery charge-discharge experimental datasets corresponding to different charging strategies and ambient temperatures. The interval recording reference is the SOC value or time value.

[0024] Preferably, step S2 includes:

[0025] Step S21: For any set of data in the battery charge and discharge experiment dataset, use noise reduction methods to remove the interfering data.

[0026] Step S22: Based on a predetermined set of multiple consecutive SOC change intervals, generate a two-dimensional table with the SOC change intervals and the number of cycles corresponding to the two coordinate axes of the two-dimensional coordinate system.

[0027] Step S23: Use the voltage, current and temperature data of this set of data to process the two-dimensional table to obtain table data, and use the table data to generate the first dataset;

[0028] Step S24: After determining the first dataset corresponding to each set of data in the battery charging and discharging experiment dataset, randomly allocate 80% of the first dataset as the training dataset and the remaining 20% ​​of the first dataset as the validation dataset.

[0029] Preferred options also include:

[0030] Record the fitting accuracy error values ​​of multiple first datasets in the validation dataset during the process of estimating the validation of the neural network model;

[0031] When at least one of the SOC range, temperature value, and dispersion of the charging strategy corresponding to multiple charge-discharge cycles in the status information is greater than the corresponding preset first dispersion threshold, second dispersion threshold, and third dispersion threshold, the second estimation result is obtained by weighted averaging the estimation results based on the recorded fitting accuracy error values ​​for different SOC ranges, different temperatures, and different charging strategies.

[0032] To achieve the above objectives, the present invention also provides a lithium-ion battery remaining life estimation device, comprising:

[0033] The data acquisition module is used to collect state information on the continuous SOC value change of the target lithium-ion battery under different charge and discharge cycles; the continuous SOC value change range is 10%-30%; the state information includes the number of cycles, voltage, current, battery temperature, ambient temperature and SOC information in each charge and discharge cycle; the number of cycles is greater than 10.

[0034] The data processing module is used to preprocess the status information and generate a two-dimensional table with the SOC value and the number of charging cycles corresponding to the two axes of the two-dimensional coordinate system, respectively. The status information is used to process the two-dimensional table to obtain multiple data tables, including voltage data tables, current data tables and temperature data tables.

[0035] The remaining life estimation module is used to input voltage data tables, current data tables, and temperature data tables into a pre-trained estimation neural network model and output the estimated results of the remaining life of the lithium-ion battery.

[0036] To achieve the above objectives, the present invention also provides a computer storage medium storing computer-executable instructions, which, when executed, implement the steps of any one of claims 1 to 8.

[0037] The present invention has achieved at least the following beneficial effects:

[0038] 1. Taking into full account factors such as the battery's operating environment, charging and discharging strategies, and charging and discharging intensity, a novel data processing method is used through a convolutional neural network model to estimate the remaining lifespan of lithium-ion batteries under complex operating environments.

[0039] 2. The constructed model integrates three-dimensional and two-dimensional neural networks, taking the battery's voltage, current, and temperature curves as inputs instead of treating them as three different channels of model input. This allows for a better capture of the complex relationships within the battery, effectively extracting and modeling the hidden features, and using the differences between cycles as a parallel network structure to improve accuracy.

[0040] 3. By utilizing residual networks to preserve features at different depths, the overall trend of the curve and the differences between cycles are effectively captured, further improving prediction performance. Results show that this method can predict battery life from its early stages to the end of its lifespan using relatively short, continuous SOC value variation intervals and a smaller number of cycles of data.

[0041] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0042] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration:

[0043] Figure 1 This is a flowchart illustrating the steps of a method for estimating the remaining life of a lithium-ion battery in an embodiment of the present invention.

[0044] Figure 2 This is a graph showing the relationship between the discharge capacity of battery No. 1 and the number of cycle periods in an embodiment of the present invention.

[0045] Figure 3This is a graph showing the relationship between the discharge capacity of battery No. 2 and the number of cycle periods in an embodiment of the present invention.

[0046] Figure 4 This is a graph showing the data collected from the battery charge-discharge test in an embodiment of the present invention.

[0047] Figure 5 This is a schematic diagram of the structure of a two-dimensional table in an embodiment of the present invention;

[0048] Figure 6 This is a structural block diagram of the convolutional neural network model in an embodiment of the present invention;

[0049] Figure 7 This refers to the prediction accuracy of the remaining life estimate of battery obtained by fitting battery No. 1 in the 20%-40% SOC range in the embodiment of the present invention.

[0050] Figure 8 This refers to the prediction accuracy of the remaining battery life estimate obtained by fitting battery No. 1 in the 40%-60% SOC range in this embodiment of the invention.

[0051] Figure 9 This refers to the prediction accuracy of the remaining battery life estimate obtained by fitting battery No. 1 in the 60%-80% SOC range in this embodiment of the invention.

[0052] Figure 10 This is a flowchart illustrating the training process of the convolutional neural network model in an embodiment of the present invention. Detailed Implementation

[0053] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0054] In real-world scenarios closely resembling daily use of electric vehicles, battery charging typically begins when the State of Charge (SOC) is between 20% and 60%, and stops when the SOC reaches 70% to 100%. Furthermore, most manufacturers, considering the need to extend battery life, recommend maintaining the electric vehicle battery's SOC between 20% and 80% to avoid overcharging or over-discharging.

[0055] In the actual testing process, for the battery cell: lithium iron phosphate (LFP) / graphite, model: 18650M1A, nominal capacity: 1.1Ah, rated voltage: 3.3V, the corresponding No. 1 battery test information is as follows:

[0056] The experimental environment temperature was 10 degrees Celsius. The charging strategy was as follows: for SOC < 40%, a 4C rate constant current (CC) fast charging method was used, followed by a 1C rate constant current constant voltage (CC-CV) charging method. The experiment yielded a curve showing the relationship between the discharge capacity and the number of cycles for battery No. 1. Figure 2 And the curve showing the relationship between the discharge capacity of battery No. 2 and the number of cycle periods. Figure 3 .

[0057] Battery 2 was tested at an ambient temperature of 30 degrees Celsius; charging strategy: the battery was charged at a constant current constant voltage (CC-CV) rate of 1C throughout the entire test. It can be seen that as the number of cycles gradually increases, the battery's discharge capacity gradually and non-linearly decreases from approximately 1.1 Ah (the initial nominal capacity) to 0.88 Ah, which is 80% of the initial capacity. The test data for batteries 1 and 2 also show that the number of cycles required for the battery to reach 80% capacity differs depending on the charging strategy and ambient temperature.

[0058] like Figure 4 As shown, the data curves collected during battery charge-discharge tests illustrate the changes in voltage and temperature with capacity ratio during charging. Since the battery changes at the 40% SOC threshold, the changes in the curves can be seen in the graph. The deviation in the aging trajectory in the graph clearly shows that different aging characteristics can be distinguished by the changes in the battery voltage and temperature curves. The graph shows that the rate of voltage increase increases with the degree of battery aging. This is because the internal characteristics of the battery (such as battery resistance) change with different degrees of aging. Simultaneously, the battery's operating temperature also affects its performance. Therefore, this invention selects the battery voltage, current, and temperature curves during the charging process as the model input for this work. This also allows the prediction to be applicable to different charging conditions.

[0059] It can be observed that when the battery ages to a certain extent, the charging voltage reaches 3.6V (upper cutoff voltage) before the capacity reaches 80%. Figure 4 As shown in (a), although the ambient temperature is kept constant, the battery surface temperature is relatively high at the beginning of the charging process due to heat accumulation from the previous discharge cycle. When the capacity ratio reaches 80%, the temperature drops significantly due to the longer charging time and relatively lower charging rate. Figure 4 As shown in (b).

[0060] To make the model applicable to the aforementioned scenarios, this invention proposes a novel model that uses segmented data (20%-80% capacity ratio) as input. The proposed model uses only data corresponding to 20% charging capacity during the charging process, demonstrating excellent applicability in various practical applications. Since the internal characteristics of the battery during charging determine all points on the external performance curve, points corresponding to the same capacity ratio range must be able to distinguish degradation characteristics across different cycles, such as... Figure 4 As shown in (a), this 20% charging capacity can be any segment of the charging process where the capacity ratio is within the range of 20%.

[0061] The present invention provides a method for estimating the remaining life of a lithium-ion battery, referring to... Figure 1 ,include:

[0062] Step S01: Collect the state information of the continuous SOC value change of the target lithium-ion battery under different charge and discharge cycles; the continuous SOC value change range is 10%-30%; the state information includes the number of cycles, voltage, current, battery temperature, ambient temperature and SOC information in each charge and discharge cycle; the number of cycles is greater than 10.

[0063] Step S02: After preprocessing the status information, a two-dimensional table is generated with the SOC value and the number of charging cycles corresponding to the two coordinate axes of the two-dimensional coordinate system, respectively. The status information is used to process the two-dimensional table to obtain multiple data tables, including voltage data table, current data table and temperature data table.

[0064] Step S03: Input the voltage data table, current data table, and temperature data table into the pre-trained estimation neural network model, and output the estimation result of the remaining life of the lithium-ion battery.

[0065] The working principle and beneficial effects of the above technical solution are as follows: By collecting state information of the target lithium-ion battery corresponding to different SOCs under different charge and discharge cycles, the state information includes at least data such as battery SOC, battery temperature, ambient temperature, charging voltage, and discharging voltage; after preprocessing the state information, a two-dimensional table is generated with SOC and the number of charging cycles corresponding to the two axes of a two-dimensional coordinate system (see the structure of the two-dimensional table). Figure 5 This method processes a two-dimensional table using state information to obtain multiple tabular data sets. The data within any cell of each table includes, but is not limited to, discrete data and continuous data curves. This tabular data is then input into a pre-trained estimation neural network model, which outputs an estimate of the remaining lifespan of the lithium-ion battery. This technical solution fully considers factors such as the battery's operating environment, charging / discharging strategy, and charging / discharging intensity. It utilizes a novel data processing method through a convolutional neural network model to estimate the remaining lifespan of lithium-ion batteries under complex operating environments. Unlike the RGB channels in images and videos, the relationship between voltage, current, and temperature is complex. Therefore, the proposed method does not treat them as three different input channels to the model; instead, it uses the battery voltage, current, and temperature curves as a third dimension for input.

[0066] In a preferred embodiment, the continuous SOC variation range is 20%, and the number of cycles is 20.

[0067] In a preferred embodiment, the temperature data is battery temperature information or the difference between battery temperature information and ambient temperature.

[0068] In a preferred embodiment, refer to Figure 10 The neural network model is trained through the following steps:

[0069] Step S1: Obtain the battery charge / discharge experiment dataset;

[0070] Step S2: After preprocessing the data in the battery charge and discharge experiment dataset, construct the training dataset and validation dataset for the battery charge and discharge characteristic neural network;

[0071] Step S3: Train the pre-built convolutional neural network model using the training dataset, and adjust the hyperparameters in the convolutional neural network model according to the Bayesian optimization algorithm to obtain the trained convolutional neural network model. The hyperparameters include the number of filters in the convolutional layer, the size of the convolutional kernel, the polling size, and the number of neurons in the fully connected layer.

[0072] Step S4: Validate the trained convolutional neural network model using the validation dataset. Determine whether the validation is successful by comparing the error between the estimated result and the actual result output by the convolutional neural network model with the preset allowable error range. If the validation fails, continue to adjust the hyperparameters in the convolutional neural network model until the validation dataset is successfully validated and the estimated neural network model is obtained.

[0073] The working principle and beneficial effects of the above technical solution are as follows: A battery charging and discharging experiment dataset is acquired; after preprocessing the data in the battery charging and discharging experiment dataset, a training dataset and a validation dataset for a battery charging and discharging characteristic neural network are constructed; the training dataset is input into a pre-built convolutional neural network model for training, and the hyperparameters in the convolutional neural network model are adjusted according to the Bayesian optimization algorithm to obtain a trained convolutional neural network model. The hyperparameters include the number of filters in the convolutional layer, the size of the convolutional kernel, the polling size, and the number of neurons in the fully connected layer; the trained convolutional neural network model is validated using the validation dataset. The validation is judged based on the error between the estimated result output by the convolutional neural network model and the actual result within a preset allowable error range. If the validation fails, the hyperparameters in the convolutional neural network model are adjusted until the validation dataset passes the validation, thus obtaining the estimation neural network model. This achieves the training of the estimation neural network model. In a preferred embodiment, refer to... Figure 6 and Figure 10The convolutional neural network model, from input to output, sequentially includes: a first 3D convolutional layer, a second 3D convolutional layer, a first 2D convolutional layer, a second 2D convolutional layer, a pooling layer, and a connecting layer, constructed using a residual network. The first 3D convolutional layer initially fuses features between current and voltage, and current and temperature, outputting a preliminary fused 3D matrix. The second 3D convolutional layer further fuses the preliminary fused 3D matrix, outputting a 2D matrix. The outputs of the second 3D convolutional layer and the first 2D convolutional layer are transferred to the residual network. Through skip connections, the extracted feature information matrix is ​​transferred to the fully connected layer. Pooling layers or dropout layers are used in each connection of the residual network. This allows the residual network to retain features at different depths, effectively capturing the overall trend of the curve and the differences between periods, further improving prediction performance. Specifically, it includes:

[0074] 3D Convolutional Layer: A 3D convolutional layer performs a three-dimensional convolution operation on the input data to extract local features. The mathematical expression for the convolution operation is:

[0075]

[0076] Among them, y ijk W represents an element in the convolution output feature map. mnp x represents the weights in the convolution kernel. (i+m)(j+n)(k+p) The element representing the input data, b ijk This represents the bias term. In this process, local features of the input data are captured by sliding convolution kernels.

[0077] In this embodiment, the input to the neural network is the voltage, current, and temperature data curves of the battery, which are collected during different charging cycles. First, these curves are processed using 3D convolutional layers. This invention uses two 3D convolutional layers with the following hyperparameters: the first convolutional layer has 3 input channels corresponding to the voltage, current, and temperature data, 16 output channels, a 3x3x3 kernel size, and 1 padding; the second convolutional layer has 16 input channels, 32 output channels, a 3x3x3 kernel size, and 1 padding. Through these 3D convolutional layers, the network is able to extract and model hidden features from the battery curves.

[0078] 2D Convolutional Layers: Similar to 3D convolutional layers, 2D convolutional layers perform two-dimensional convolution operations on the input data. The mathematical expression for the convolution operation is:

[0079]

[0080] Among them, y ij W represents an element in the convolution output feature map. mn x represents the weights in the convolution kernel.(i+m)(j+n) The element representing the input data, b ij This represents the bias term. In this process, local features of the input data are captured by sliding convolution kernels.

[0081] In this embodiment, after passing through the 3D convolutional layers, the output is reshaped into a 2D shape. Then, these 2D feature maps are processed using 2D convolutional layers. This invention uses two 2D convolutional layers with the following hyperparameters: the first convolutional layer has 32 input channels, 64 output channels, a 3x3 kernel size, and 1 padding. The second convolutional layer has 64 input channels, 128 output channels, a 3x3 kernel size, and 1 padding. Through these 2D convolutional layers, the network can further extract and model features from the battery curve.

[0082] Pooling Layer: The pooling layer is used to reduce data dimensionality and computational cost. There are generally two types of pooling operations: max pooling and average pooling. Here, we prefer average pooling.

[0083] Fully connected layer: A fully connected layer connects all output nodes of the previous layer to all input nodes of the current layer. The mathematical expression for a fully connected layer is:

[0084]

[0085] Among them, y i W represents the value of the output node of the fully connected layer. ij Indicates the connection weight, x j b represents the value of the input node. i This indicates the bias term.

[0086] In this embodiment, after passing through the 2D convolutional layer, the output is flattened into a 1D shape. Then, the flattened features are processed by fully connected layers. This invention uses two fully connected layers: the first fully connected layer has an input size of 128 and an output size of 64; the second fully connected layer has an input size of 64 and an output size of 1, which is the final prediction result. Through these fully connected layers, the network can map the extracted features to the final prediction result.

[0087] Activation functions: The ReLU activation function is used to introduce non-linearity between layers of the network. The ReLU function sets negative values ​​to zero and keeps positive values ​​unchanged, thereby increasing the expressive power of the network.

[0088] During training, the backpropagation algorithm is used to update the network parameters (weights and biases).

[0089] First, calculate the loss function. Here, we take the mean squared error loss function as an example:

[0090]

[0091] Where N represents the number of samples, y i Indicates network output, t i This represents the target value.

[0092] Next, we update the weights and biases using gradient descent:

[0093]

[0094]

[0095] Where η represents the learning rate.

[0096] Validating Convolutional Neural Networks:

[0097] Validate the trained convolutional neural network using a validation dataset. This step involves validating the trained convolutional neural network using a validation set. By comparing the network's predictions with actual results, the model's performance can be evaluated, and further adjustments and training can be performed as needed. The purpose of this step is to ensure that the model can accurately predict remaining battery life. Validation results are typically evaluated using various metrics, such as mean squared error (MSE) and mean absolute error (MAE). If the validation results are not as expected, methods such as readjusting the neural network structure, increasing or decreasing training data, etc., can be considered to improve model performance. To measure model performance, evaluation metrics such as accuracy, precision, and recall can be calculated. If the model performs as expected on the validation set, the validation is successful. Otherwise, it is necessary to continue adjusting network parameters or structure and repeat the training and validation process until satisfactory performance is achieved.

[0098] Throughout the training and validation process, Bayesian optimization algorithms can be used to tune the hyperparameters of convolutional neural networks, such as the number of filters in convolutional layers, kernel size, pooling size, and the number of neurons in fully connected layers. The goal of Bayesian optimization is to find the hyperparameter combination that minimizes the validation loss within a finite number of trials. It does this by constructing a probabilistic model (such as a Gaussian process) of the objective function (validation loss) and then using Bayesian inference to select the hyperparameter combination for the next trial, thus finding the optimal hyperparameter combination with the fewest possible attempts.

[0099] Suppose we want to minimize the validation loss function f(x), where x is a combination of hyperparameters. The objective is to find the hyperparameter combination x that minimizes the validation loss within a finite number of trials. * .

[0100] To achieve this goal, the objective function f(x) is modeled as a Gaussian process, i.e.:

[0101] f(x)~GP(μ(x),σ 2 (x))

[0102] Where GP represents a Gaussian process, μ(x) and σ 2 (x) represent the mean and variance at the hyperparameter combination x, respectively.

[0103] In the Bayesian optimization algorithm, a prior distribution P(x) is used to represent the distribution of the hyperparameter combination x. Then, the prior distribution is updated by observing some data points to obtain the posterior distribution P(x|D), where D is the observed data points.

[0104] Bayes' theorem is expressed as:

[0105]

[0106] Where P(D|x) represents the probability of observing data point D given the hyperparameter combination x, P(x) is the prior distribution, and P(D) is the normalization constant.

[0107] We use a Gaussian process to model P(D|x), that is:

[0108] P(D│x)~N(f(x),δ 2 )

[0109] Where N represents a normal distribution, f(x) is the value of the objective function at the hyperparameter combination x, and δ 2 It is the noise variance of the observed data points.

[0110] Define a hyperparameter selection strategy x t+1 =argmin x EI(x), where EI(x) is the expected improvement function, defined as:

[0111] EI(x)=E[max(f(x t )-f(x),0)]

[0112] Where x t Let E be the combination of hyperparameters from the previous t attempts, and E represent the expected value. The expected improvement function represents the value at the current optimal point y. t =min i≤t f(x i Under the given conditions, determine the degree of improvement that choosing the hyperparameter combination x can bring. Use the EI function to select the next hyperparameter combination x to try. t+1 .

[0113] In each iteration, a new combination of hyperparameters x will be selected. t+1 We make an attempt and update the prior distribution P(x) based on the observed data points to obtain the posterior distribution P(x│D). t This process can be repeated many times until the predetermined number of attempts is reached or the optimal combination of hyperparameters is found.

[0114] In a preferred embodiment, the battery charge-discharge experimental dataset is obtained through the following steps:

[0115] Step S11: Randomly divide multiple battery samples with the same specifications into n first-class groups, and adopt n different charging and discharging strategies for each of the n first-class groups.

[0116] Step S12: Divide each first group into m second groups, and assign each of the m second groups to m different ambient temperatures;

[0117] Step S13: Record the number of cycles, voltage, current, battery temperature, ambient temperature, and SOC information of the grouped batteries in each charge-discharge cycle at intervals using the battery BMS, test sensors, or external temperature sensors, thereby collecting m*n sets of full-cycle battery charge-discharge experimental datasets corresponding to different charging strategies and ambient temperatures. The interval recording reference is the SOC value or time value.

[0118] The working principle and beneficial effects of the above technical solution are as follows: Multiple battery samples with the same specifications are randomly divided into n groups, with m test samples assigned to each group. Different test samples in each group use different charge / discharge strategies n for each charge / discharge cycle (e.g., constant current charging (CC) for sample group 1, constant voltage charging (CV) for sample group 2, or different charge / discharge rates or combinations thereof for different sample groups n). Each different test environment temperature in each group corresponds to a different ambient temperature m (e.g., -30°C to 40°C, as close as possible to the actual possible operating environment temperature of the battery). The battery cells under test are cyclically charged and discharged on a test bench until the capacity is less than 80% of the initial value, at which point the battery life is considered to have ended. Information such as the number of cycles, voltage, current, battery temperature, ambient temperature, and SOC are recorded at intervals during each charge / discharge cycle using the battery BMS, test sensors, or external temperature sensors. The preferred reference for interval recording is a fixed interval SOC value. However, for ease of data acquisition, more data can be collected at regular intervals, and data points can be selected based on the continuously changing SOC values ​​in subsequent data processing. Through the above data collection, a total of m*n sets of battery full-cycle data corresponding to different charging strategies and ambient temperatures were collected. The data associated with each battery is divided into three categories: descriptors, summary data, and cycle data. The descriptors for each battery include charging strategy, ambient temperature, and battery model information. The summary data includes information for each charge-discharge cycle, including cycle number, discharge capacity, charge capacity, internal resistance, maximum temperature, average temperature, minimum temperature, and charging time. The cycle data includes information for a single charge-discharge cycle, including time, charge capacity, current, voltage, temperature, and discharge capacity. This enables the acquisition of battery charge-discharge experimental datasets under all ambient temperatures and all charging strategies.

[0119] In a preferred embodiment, step S2 includes:

[0120] Step S21: For any set of data in the battery charge and discharge experiment dataset, use noise reduction methods to remove the interfering data.

[0121] Step S22: Based on a predetermined set of multiple consecutive SOC change intervals, generate a two-dimensional table with the SOC change intervals and the number of cycles corresponding to the two coordinate axes of the two-dimensional coordinate system.

[0122] Step S23: Use the voltage, current and temperature data of this set of data to process the two-dimensional table to obtain table data, and use the table data to generate the first dataset;

[0123] Step S24: After determining the first dataset corresponding to each set of data in the battery charging and discharging experiment dataset, randomly allocate 80% of the first dataset as the training dataset and the remaining 20% ​​of the first dataset as the validation dataset.

[0124] The working principle and beneficial effects of the above technical solution are as follows: For each set of data, a denoising method (such as wavelet denoising) is first used to remove interference data that significantly deviates from the normal value. Multiple continuous SOC variation intervals are selected (preferably 20% as one interval, specifically 20%-40%, 40%-60%, and 60%-80%. Note that the SOC variation interval can be any continuous 20% interval, which facilitates the acquisition of relevant data in practical work. Furthermore, the values ​​of the continuous intervals can be adjusted according to the actual situation, such as 10%-30%). Using SOC and cycle number as the two axes of a two-dimensional coordinate system, the voltage, current, and temperature data are processed into... Figure 5 The two-dimensional table shown (a further improvement could reflect the impact of battery operation at different temperatures; temperature data would be the difference between battery temperature and ambient temperature, thus further incorporating the impact of ambient temperature data on battery life). The resulting dataset is randomly allocated 80% as the training set and the remaining 20% ​​as the validation set. This achieves the generation of both the training and validation datasets.

[0125] In a preferred embodiment, it further includes:

[0126] This record tracks the fitting accuracy error values ​​of multiple first datasets within the validation dataset during the validation process of the estimated neural network model. For example, it shows the fitting accuracy error values ​​for different SOC ranges. The following is a reference chart showing the prediction accuracy of the estimated battery remaining life. Figures 7-9 Fill the recorded data into the average fitting accuracy error value table for each SOC interval;

[0127]

[0128] The table records the average accuracy error values ​​of multiple data points in the same SOC range during the estimation of the neural network model validation process. Similarly, there is a table of average fitting accuracy error values ​​for different charging strategies and different temperature environments (including battery temperature, ambient temperature, and temperature difference), which will not be demonstrated here.

[0129] When at least one of the following parameters in the status information—the SOC range, temperature value, and dispersion of the charging strategy corresponding to multiple charge-discharge cycles—is greater than the corresponding preset first, second, and third dispersion thresholds, a second estimation result is obtained by weighted averaging the estimation results based on the recorded fitting accuracy error values ​​for different SOC ranges, temperatures, and charging strategies. This second estimation result provides a more accurate estimate of the remaining battery life.

[0130] To achieve the above objectives, the present invention also provides a lithium-ion battery remaining life estimation device, comprising:

[0131] The data acquisition module is used to collect state information on the continuous SOC value change of the target lithium-ion battery under different charge and discharge cycles; the continuous SOC value change range is 10%-30%; the state information includes the number of cycles, voltage, current, battery temperature, ambient temperature and SOC information in each charge and discharge cycle; the number of cycles is greater than 10.

[0132] The data processing module is used to preprocess the status information and generate a two-dimensional table with the SOC value and the number of charging cycles corresponding to the two axes of the two-dimensional coordinate system, respectively. The status information is used to process the two-dimensional table to obtain multiple data tables, including voltage data tables, current data tables and temperature data tables.

[0133] The remaining life estimation module is used to input voltage data tables, current data tables, and temperature data tables into a pre-trained estimation neural network model and output the estimated results of the remaining life of the lithium-ion battery.

[0134] The working principle and beneficial effects of the above technical solution are as follows: The data acquisition module collects state information of the target lithium-ion battery under different charge-discharge cycles corresponding to different SOCs, including at least battery SOC, battery temperature, ambient temperature, charging voltage, and discharging voltage. The data processing module preprocesses the state information and generates a two-dimensional table with SOC and the number of charging cycles corresponding to the two axes of a two-dimensional coordinate system. Multiple tables are then processed using the state information. The remaining life estimation module inputs the table data into a pre-trained estimation neural network model and outputs the estimated remaining life of the lithium-ion battery. This technical solution fully considers factors such as the battery's operating environment, charging and discharging strategy, and charging and discharging intensity. It utilizes a novel data processing method through a convolutional neural network model to estimate the remaining life of lithium-ion batteries under complex operating environments.

[0135] To achieve the above objectives, the present invention also provides a computer storage medium storing computer-executable instructions, which, when executed, implement the steps of any of the methods described in the above embodiments.

[0136] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0137] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0138] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0139] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0140] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0141] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0142] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0143] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A method for estimating the remaining life of a lithium-ion battery, characterized in that, include: Step S01: Collect state information on the continuous SOC value change of the target lithium-ion battery under different charge-discharge cycles; the continuous SOC value change range is 10%-30%; the state information includes the number of cycles, voltage, current, battery temperature, ambient temperature and SOC information in each charge-discharge cycle; the number of cycles is greater than 10. Step S02: After preprocessing the status information, a two-dimensional table is generated with the SOC value and the number of charging cycles corresponding to the two coordinate axes of the two-dimensional coordinate system, respectively. The two-dimensional table is then processed using the status information to obtain multiple data tables, including a voltage data table, a current data table, and a temperature data table. Step S03: Input the voltage data table, current data table and temperature data table into the pre-trained estimation neural network model, and output the estimation result of the remaining life of the lithium-ion battery. The estimation neural network model is trained through the following steps: Step S1: Obtain the battery charge / discharge experiment dataset; Step S2: After preprocessing the data in the battery charge and discharge experiment dataset, construct the training dataset and validation dataset for the battery charge and discharge characteristic neural network; Step S3: Train the pre-built convolutional neural network model using the training dataset, and adjust the hyperparameters in the convolutional neural network model according to the Bayesian optimization algorithm to obtain the trained convolutional neural network model. The hyperparameters include the number of filters in the convolutional layer, the size of the convolutional kernel, the polling size, and the number of neurons in the fully connected layer. Step S4: Validate the trained convolutional neural network model using the validation dataset. Determine whether the validation is successful by comparing the error between the estimated result and the actual result output by the convolutional neural network model with the preset allowable error range. If the validation fails, continue to adjust the hyperparameters in the convolutional neural network model until the validation dataset is successful and the estimated neural network model is obtained. The convolutional neural network model, from input to output, includes: a first 3D convolutional layer, a second 3D convolutional layer, a first 2D convolutional layer, a second 2D convolutional layer, a pooling layer, and a connecting layer, constructed using a residual network. The first 3D convolutional layer initially fuses the features between current and voltage, and current and temperature, outputting a preliminary fused 3D matrix. The second 3D convolutional layer further fuses the preliminary fused 3D matrix, outputting a 2D matrix. The outputs of the second 3D convolutional layer and the first 2D convolutional layer are transferred to the residual network. The extracted feature information matrix is ​​transferred to the fully connected layer through skip connections. Pooling layers or dropout layers are used in each connection of the residual network.

2. The method for estimating the remaining life of a lithium-ion battery according to claim 1, characterized in that, The continuous SOC value variation range is 20%, and the number of cycles is 20.

3. The method for estimating the remaining life of a lithium-ion battery according to claim 1, wherein the temperature data is battery temperature information or the difference between battery temperature information and ambient temperature.

4. The method for estimating the remaining life of a lithium-ion battery according to claim 1, characterized in that, The battery charge / discharge experimental dataset was obtained through the following steps: Step S11: Randomly divide multiple battery samples with the same specifications into n first-class groups, and adopt n different charging and discharging strategies for each of the n first-class groups. Step S12: Divide each first group into m second groups, and assign each of the m second groups to m different ambient temperatures; Step S13: Record the number of charge / discharge cycles, voltage, current, battery temperature, ambient temperature, and SOC information of the grouped batteries at intervals using the battery BMS, test sensors, or external temperature sensors, thereby collecting data. The dataset consists of full-cycle battery charge-discharge experiments with different charging strategies and ambient temperatures. The interval records are referenced by SOC values ​​or time values.

5. The method for estimating the remaining life of a lithium-ion battery according to claim 1, characterized in that, Step S2 includes: Step S21: For any set of data in the battery charge and discharge experiment dataset, use noise reduction methods to remove the interfering data. Step S22: Based on a predetermined set of multiple consecutive SOC change intervals, generate a two-dimensional table with the SOC change intervals and the number of cycles corresponding to the two coordinate axes of the two-dimensional coordinate system. Step S23: Use the voltage, current and temperature data of this set of data to process the two-dimensional table to obtain table data, and use the table data to generate the first dataset; Step S24: After determining the first dataset corresponding to each set of data in the battery charging and discharging experiment dataset, randomly allocate 80% of the first dataset as the training dataset and the remaining 20% ​​of the first dataset as the validation dataset.

6. The method for estimating the remaining life of a lithium-ion battery according to claim 1, characterized in that, Also includes: Record the fitting accuracy error values ​​of the first dataset generated by multiple first datasets in the validation dataset during the process of estimating the validation of the neural network model; When at least one of the SOC range, temperature value, and dispersion of the charging strategy corresponding to multiple charge-discharge cycles in the status information is greater than the corresponding preset first dispersion threshold, second dispersion threshold, and third dispersion threshold, the second estimation result is obtained by weighted averaging the estimation results based on the recorded fitting accuracy error values ​​for different SOC ranges, different temperatures, and different charging strategies.

7. A device for estimating the remaining life of a lithium-ion battery, characterized in that, include: The data acquisition module is used to collect state information on the continuous SOC value change of the target lithium-ion battery under different charge-discharge cycles; the continuous SOC value change range is 10%-30%; the state information includes the number of cycles, voltage, current, battery temperature, ambient temperature and SOC information in each charge-discharge cycle; the number of cycles is greater than 10. The data processing module is used to preprocess the status information and generate a two-dimensional table with the SOC value and the number of charging cycles corresponding to the two coordinate axes of the two-dimensional coordinate system, respectively. The status information is used to process the two-dimensional table to obtain multiple data tables, including a voltage data table, a current data table and a temperature data table. The remaining life estimation module is used to input the voltage data table, current data table and temperature data table into a pre-trained estimation neural network model and output the estimation result of the remaining life of the lithium-ion battery. The estimation neural network model is trained through the following steps: Step S1: Obtain the battery charge / discharge experiment dataset; Step S2: After preprocessing the data in the battery charge and discharge experiment dataset, construct the training dataset and validation dataset for the battery charge and discharge characteristic neural network; Step S3: Train the pre-built convolutional neural network model using the training dataset, and adjust the hyperparameters in the convolutional neural network model according to the Bayesian optimization algorithm to obtain the trained convolutional neural network model. The hyperparameters include the number of filters in the convolutional layer, the size of the convolutional kernel, the polling size, and the number of neurons in the fully connected layer. Step S4: Validate the trained convolutional neural network model using the validation dataset. Determine whether the validation is successful by comparing the error between the estimated result and the actual result output by the convolutional neural network model with the preset allowable error range. If the validation fails, continue to adjust the hyperparameters in the convolutional neural network model until the validation dataset is successful and the estimated neural network model is obtained. The convolutional neural network model, from input to output, includes: a first 3D convolutional layer, a second 3D convolutional layer, a first 2D convolutional layer, a second 2D convolutional layer, a pooling layer, and a connecting layer, constructed using a residual network. The first 3D convolutional layer initially fuses the features between current and voltage, and current and temperature, outputting a preliminary fused 3D matrix. The second 3D convolutional layer further fuses the preliminary fused 3D matrix, outputting a 2D matrix. The outputs of the second 3D convolutional layer and the first 2D convolutional layer are transferred to the residual network. The extracted feature information matrix is ​​transferred to the fully connected layer through skip connections. Pooling layers or dropout layers are used in each connection of the residual network.

8. A computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, which, when executed, implement the steps of the method according to any one of claims 1 to 6.

Images