A lithium battery state of charge evaluation method based on charge curve reconstruction

Abstract

The application relates to a lithium battery state of charge evaluation method based on charging curve reconstruction, which comprises the following steps: presetting charging cutoff conditions, discharging cutoff conditions and cycle cutoff conditions, continuously charging and discharging the lithium battery according to the preset cutoff conditions, and recording data information in the experiment process; collecting charging curve data of the lithium battery, performing data cleaning, and removing error data; pre-processing the cleaned data, filling in missing values by adopting a linear interpolation method, and obtaining voltage-time sequence data of the lithium battery; forming a data set by sliding a window to the voltage-time sequence data; training the data set, and constructing a charging curve reconstruction model; verifying the performance of the charging curve reconstruction model by using a test set and a verification set, and calculating the battery power based on the charging curve reconstruction model; and the lithium battery state of charge evaluation method can strengthen short-term time sequence feature extraction and provide data support for lithium battery health management.

Description

Technical Field

[0001] This application relates to the field of lithium-ion battery technology, specifically to a method for assessing the state of charge of lithium batteries based on charging curve reconstruction. Background Technology

[0002] Lithium-ion batteries, with their high energy density, low self-discharge rate, and long cycle life, have become the core energy storage unit in new energy power systems, electric vehicles, and other fields. A complete constant-current charging curve contains crucial information about the battery's state of charge (SOC) and is fundamental to achieving refined SOC management. In practical applications, energy storage batteries need to frequently switch between charging and discharging in response to upper-level dispatch commands or grid load changes, resulting in fragmented and discontinuous charging data, making it difficult to obtain a complete constant-current charging curve. Existing methods for estimating the SOC of lithium batteries include model-based methods that are complex to model and have poor adaptability, and traditional data-driven methods that are highly dependent on complete charging data and cannot effectively handle fragmented input. Conventional deep learning models are not specifically optimized for the temporal characteristics of battery charging data, resulting in insufficient prediction accuracy and limited generalization ability, making it difficult to meet the application requirements of real-world scenarios. Summary of the Invention

[0003] This application provides a lithium battery state of charge assessment method based on charging curve reconstruction. This lithium battery state of charge assessment method can enhance the extraction of short-term time-series features, thereby providing data support for lithium battery health management.

[0004] The lithium battery state-of-charge assessment method based on charging curve reconstruction provided in this application includes the following steps: S1. Preset charging cutoff conditions, discharging cutoff conditions, and cycle cutoff conditions. Conduct continuous charging and discharging experiments on the lithium battery according to the preset cutoff conditions and record the data information during the experiment. S2. Collect the charging curve data of the lithium battery, perform data cleaning, and remove erroneous data; S3. The cleaned data is preprocessed, and missing values ​​are filled in using linear interpolation to obtain the voltage-time series data of the lithium battery. S4. The voltage-time series data is used to form a dataset through a sliding window; S5. Train the dataset to construct a charging curve reconstruction model; S6. Using the test set and validation set, verify the performance of the charging curve reconstruction model, and calculate the battery capacity based on the charging curve reconstruction model.

[0005] In one alternative embodiment, in step S1, the charging cutoff condition is when the voltage of the lithium battery rises to 4.2V, the discharging cutoff condition is when the voltage of the lithium battery drops to 2.7V, and the cycle cutoff condition is when the state of health (SOH) reaches 70%.

[0006] In one alternative approach, step S3, when preprocessing the data, includes: Low-health data (SOH ≤ 0.7) were removed. The effective charging curves were resampled using linear interpolation to a standardized voltage-time series with a voltage range of 2.8V to 4.2V and a sampling interval of 5mV, resulting in unified format data containing n data pairs. , Data pairs; among which, k=1,2,...,n.

[0007] In one alternative approach, step S3, when preprocessing the data, further includes constructing the fragmented dataset, which involves the following steps: Within a voltage range of 2.8V to 4.2V, a voltage window with a width of 0.3V is traversed and covered with a fixed step size. For each traversed window, only the voltage-time data within the window is retained, and the time data corresponding to the voltage outside the window is filled with zeros. The dataset is expanded by fully traversing the original valid samples and dividing them into fixed windows to increase the sample size and enhance the model's adaptability to different voltage range segments.

[0008] In an optional scheme, in step S5, the charging curve reconstruction model includes a linear embedding layer, a position encoding layer, and an encoder. The linear embedding layer is used to perform data dimensionality upscaling, the position encoding layer is used to embed absolute or relative position information into the input features, and the encoder is used to compute attention weights between all data points in the sequence in parallel.

[0009] In one optional scheme, in step S5, the charging curve reconstruction model includes a GRU layer, a 1-layer position encoder, a 5-layer stacked encoder, and a 1-layer fully connected output layer; each layer of the encoder includes a multi-head attention module, a feedforward fully connected layer, and a residual connection and layer normalization module, wherein the multi-head attention has 4 heads; the fully connected output layer is used to map the high-dimensional features output by the encoder to 281 dimensions.

[0010] In one alternative approach, the GRU network model of the charging curve reconstruction model includes a recurrent unit. This recurrent unit receives the input and the hidden state from the previous time step at each time step, and generates the output and a new hidden state for the current time step. The update gate's corresponding formula is: ; in, yes Update the gate output constantly. It is an activation function. It updates the weight matrix of the gate. It is the hidden state from the previous moment. for Input at any time; The formula for resetting the door is: ; in, yes Reset the gate output at all times. It is an activation function. It is the weight matrix of the reset gate. It is the hidden state from the previous moment. for Input at any time; The new candidate hidden state is: ; in, It is a new candidate hidden state. It is an activation function. It is the weight matrix of the new candidate hidden states. yes Reset the gate output at all times. It is the hidden state from the previous moment. for Input at any time; The hidden state of the current time step is: ; in, yes Hide your status at all times. yes Update the gate output constantly. It is a new candidate hidden state. It is the hidden state from the previous moment; The corresponding formula for output calculation is: ; in, yes Time output, It is an activation function. It is the weight matrix from the hidden state to the output. It is the hidden state at the current time step. It is the bias of the output layer.

[0011] In one alternative, step S6, when calculating the battery capacity based on the charging curve reconstruction model, includes: estimating the battery capacity using data from the reconstruction curve and the ampere-hour integration method.

[0012] The beneficial effects of this application are as follows: The lithium battery state-of-charge (SOC) assessment method in this application constructs fragmented data through voltage window sampling and expands the dataset through repeated sampling. This accurately adapts to the discontinuous data characteristics in real-world scenarios, solving the problems of inconsistent time-series dimensions and insufficient sample size during training. Furthermore, this assessment method uses a GRU to replace the linear embedding layer of the encoder, replacing the traditional embedding method that only performs dimension mapping. This strengthens the capture of short-term temporal dependencies and trends in the voltage-time series. Combined with the encoder's long-distance dependency modeling capability, this improves the model's generalization performance. The model reconstructs the complete charging curve from partial charging curve data. By completing the charging curve, the charge level at each charging voltage point can be calculated. The battery voltage and corresponding charge level provide data support for the control of the energy storage system, better adapting to the practical application needs of energy storage systems.

[0013] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0014] Figure 1 A logical diagram illustrating the lithium battery state-of-charge assessment method provided in this application; Figure 2 A model architecture diagram based on the lithium battery state-of-charge assessment method provided in this application; Figure 3 The GRU network structure diagram provided in this application; Figure 4 The encoder structure diagram provided in this application; Figure 5 The neural network prediction results for the charging segment are reconstructed based on the charging curve.

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. Detailed Implementation

[0016] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0017] It should be understood that the described embodiments are merely some embodiments of this application, and not all embodiments. All other technical solutions obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0018] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0019] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0020] like Figure 1-5 As shown in the figure, this application provides a method for assessing the state of charge (SOC) of a lithium battery based on charging curve reconstruction. This method mainly includes the following steps: S1. Preset charging cutoff conditions, discharging cutoff conditions, and cycle cutoff conditions. Conduct continuous charge and discharge experiments on the lithium battery according to the preset cutoff conditions and record the data information during the experiment. Among them, the charging cutoff condition is when the lithium battery voltage rises to 4.2V, the discharging cutoff condition is when the lithium battery voltage drops to 2.7V, and the cycle cutoff condition is when the state of health (SOH) reaches 70%.

[0021] S2. Collect the charging curve data of the lithium battery, perform data cleaning, and remove erroneous data.

[0022] S3. Preprocess the cleaned data, using linear interpolation to fill in missing values, to obtain lithium battery voltage-time series data. Preprocessing includes steps such as format conversion, timestamp processing, outlier handling, duplicate removal, resampling, and missing value filling, thereby transforming irregular data into standard data to obtain battery voltage-time series data. .

[0023] Specifically, data preprocessing includes: Low-health data (SOH ≤ 0.7) were removed. The effective charging curves were resampled using linear interpolation to a standardized voltage-time series with a voltage range of 2.8V to 4.2V and a sampling interval of 5mV, resulting in unified format data containing n data pairs. , Data pairs; among which, , k=1,2,...,281.

[0024] The fragmented dataset is constructed as follows: within the voltage range of 2.8V to 4.2V, a voltage window with a width of 0.3V is traversed and covered with a fixed step size; for each truncation window, only the voltage-time data within the window is retained, and the time data corresponding to the voltage outside the window is filled with zeros; the dataset is expanded by fully traversing the original valid samples and dividing them into fixed windows to increase the sample size and enhance the model's adaptability to different voltage range segments.

[0025] S4. The voltage-time series data is converted into a dataset using a sliding window.

[0026] S5. Train the dataset to build a charging curve reconstruction model.

[0027] Specifically, the charging curve reconstruction model includes a linear embedding layer, a position encoding layer, and an encoder. The linear embedding layer is used to increase the dimensionality of the data, the position encoding layer is used to embed absolute or relative position information into the input features, and the encoder is used to compute the attention weights between all data points in the sequence in parallel.

[0028] The charging curve reconstruction model includes a GRU layer, a 1-layer position encoder, a 5-layer stacked encoder, and a 1-layer fully connected output layer. Each encoder layer contains a multi-head attention module, a feedforward fully connected layer, and a residual connection and layer normalization module, where the multi-head attention module has 4 heads. The fully connected output layer is used to map the high-dimensional features output by the encoder to 281 dimensions.

[0029] In one alternative approach, the GRU network model of the charging curve reconstruction model includes a recurrent unit. This recurrent unit receives the input and the hidden state from the previous time step at each time step, and generates the output and a new hidden state for the current time step. The GRU neural network structure diagram is shown below. Figure 3 As shown.

[0030] The formula for updating the gate is: ; in, yes Update the gate output constantly. It is an activation function. It updates the weight matrix of the gate. It is the hidden state from the previous moment. for Input at any time; The formula for resetting the door is: ; in, yes Reset the gate output at all times. It is an activation function. It is the weight matrix of the reset gate. It is the hidden state from the previous moment. for Input at any time; The new candidate hidden state is: ; in, It is a new candidate hidden state. It is an activation function. It is the weight matrix of the new candidate hidden states. yes Reset the gate output at all times. It is the hidden state from the previous moment. for Input at any time; The hidden state of the current time step is: ; in, yes Hide your status at all times. yes Update the gate output constantly. It is a new candidate hidden state. It is the hidden state from the previous moment; The corresponding formula for output calculation is: ; in, yes Time output, It is an activation function. It is the weight matrix from the hidden state to the output. It is the hidden state at the current time step. It is the bias of the output layer.

[0031] The encoder consists of three stacked encoding layers, with the core module being the encoder multi-head attention module, as shown in the structure diagram below. Figure 4 As shown, it generates a query vector Q, a key vector K, and a value vector V by linearly transforming the input matrix, and then performs independent attention calculations. The outputs of each head are weighted and fused to obtain the global self-attention feature map. Due to the presence of multiple heads, the multi-head attention mechanism can fuse features that focus on different aspects, better handle long sequence data, and more effectively focus on global information, thereby improving the model's performance and generalization ability. The self-attention mechanism calculation formula is as follows: , ; In the formula, It is the input of the multi-head self-attention module, and its function is to give Q, K, and V a learnable attention weight, and to increase the computational weight of data with large attention weights in network training. Let be a trainable projection matrix, and SoftMax be a normalized exponential function. is the scaling factor for the dot product attention. Each input passes through h attention heads, each generating an attention matrix. All matrices are then merged with the weight matrix used during network training. Multiply them to obtain the output result. The calculation formula is as follows: ; .

[0032] In this example, the charging curve reconstruction model trained yields the following optimal model parameters: 64 GRU neurons, 5 stacked encoder layers, and 4 heads for multi-head attention. The length of a single sample is 281, the training batch size is 10, the Adam optimizer is selected, the learning rate is 0.001, the number of training epochs is 1000, the activation function for each layer is leaky_relu, and the accumulated mean absolute error is used as the loss function.

[0033] S6. Using the test set and validation set, verify the performance of the charging curve reconstruction model, and calculate the battery capacity based on the charging curve reconstruction model.

[0034] The verification process includes: reconstructing the model using the trained charging curves, making predictions using 1,000 charging curve data from the same type of battery, obtaining test results, calculating the absolute percentage error at a single point, and calculating the mean absolute percentage error to verify the practicality of the model.

[0035] From all the predicted results, take one set of predicted results as follows: Figure 5 As shown, this is a segment of data used as input. The curve fitting is excellent, indicating that the model has learned the inherent relationships between the data and has good application value. Furthermore, the absolute error between the predicted and actual values ​​is very small, which meets the requirements for data size and trend.

[0036] The error metric used is the coefficient of determination (R²). 2 The calculation formula is as follows: ; In the formula, Let be the number of sequences, and i be the number of points predicted for each sequence. For predicted values, This is the corresponding actual value.

[0037] The results for all test sets are shown in Table 1. After calculation, the predicted results R for all charging curves are... 2 The error is 0.998, and the MAE is 14.44. Both errors are very small, meeting the requirements for predicting battery charging curve data.

[0038] Furthermore, the calculation of battery capacity based on the charging curve reconstruction model includes: estimating the battery capacity using data from the reconstructed curve and the ampere-hour integration method. The ampere-hour integration method calculation formula is as follows: ; In the formula, This is the initial capacity of the battery. Let be the charging and discharging current of the battery at time t.

[0039] As can be seen from the above embodiments, the lithium battery state-of-charge assessment method in this application constructs fragmented data through voltage window sampling and expands the dataset through repeated sampling. This accurately adapts to the discontinuous data characteristics in real-world scenarios, solving the problems of inconsistent time-series dimensions and insufficient sample size during training. Furthermore, this assessment method uses a GRU to replace the linear embedding layer of the encoder, replacing the traditional embedding method that only performs dimension mapping. This strengthens the capture of short-term time-series dependencies and trends in voltage-time series data. Combined with the encoder's long-distance dependency modeling capability, this improves the model's generalization performance. The model reconstructs the complete charging curve from partial charging curve data. By completing the charging curve, the charge level at each charging voltage point can be calculated. The battery voltage and corresponding charge level provide data support for the control of the energy storage system, better adapting to the actual application needs of energy storage systems.

[0040] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for assessing the state of charge of a lithium battery based on charging curve reconstruction, characterized in that, Includes the following steps: S1. Preset charging cutoff conditions, discharging cutoff conditions, and cycle cutoff conditions. Conduct continuous charging and discharging experiments on the lithium battery according to the preset cutoff conditions and record the data information during the experiment. S2. Collect the charging curve data of the lithium battery, perform data cleaning, and remove erroneous data; S3. The cleaned data is preprocessed, and missing values ​​are filled in using linear interpolation to obtain the voltage-time series data of the lithium battery. S4. The voltage-time series data is used to form a dataset through a sliding window; S5. Train the dataset to construct a charging curve reconstruction model; S6. Using the test set and validation set, verify the performance of the charging curve reconstruction model, and calculate the battery capacity based on the charging curve reconstruction model.

2. The lithium battery state-of-charge assessment method based on charging curve reconstruction according to claim 1, characterized in that, In step S1, the charging cutoff condition is when the voltage of the lithium battery rises to 4.2V, the discharging cutoff condition is when the voltage of the lithium battery drops to 2.7V, and the cycle cutoff condition is when the state of health (SOH) reaches 70%.

3. The lithium battery state-of-charge assessment method based on charging curve reconstruction according to claim 1 or 2, characterized in that, In step S3, the data preprocessing includes: Low-health data (SOH ≤ 0.7) were removed. The effective charging curves were resampled using linear interpolation to a standardized voltage-time series with a voltage range of 2.8V to 4.2V and a sampling interval of 5mV, resulting in unified format data containing n data pairs. , Data pairs; among which, k=1,2,...,n.

4. The lithium battery state-of-charge assessment method based on charging curve reconstruction according to claim 3, characterized in that, In step S3, the data preprocessing also includes constructing the fragmented dataset, which includes the following steps: Within a voltage range of 2.8V to 4.2V, a voltage window with a width of 0.3V is traversed and covered with a fixed step size. For each traversed window, only the voltage-time data within the window is retained, and the time data corresponding to the voltage outside the window is filled with zeros. The dataset is expanded by fully traversing the original valid samples and dividing them into fixed windows to increase the sample size and enhance the model's adaptability to different voltage range segments.

5. The lithium battery state-of-charge assessment method based on charging curve reconstruction according to claim 1, 2, or 4, characterized in that, In step S5, the charging curve reconstruction model includes a linear embedding layer, a position encoding layer, and an encoder. The linear embedding layer is used to perform data dimensionality upscaling, the position encoding layer is used to embed absolute or relative position information into the input features, and the encoder is used to calculate the attention weights between all data points in the sequence in parallel.

6. The lithium battery state-of-charge assessment method based on charging curve reconstruction according to claim 5, characterized in that, In step S5, the charging curve reconstruction model includes a GRU layer, a 1-layer position encoder, a 5-layer stacked encoder, and a 1-layer fully connected output layer. Each layer of the encoder includes a multi-head attention module, a feedforward fully connected layer, and a residual connection and layer normalization module, wherein the multi-head attention module has 4 heads. The fully connected output layer is used to map the high-dimensional features output by the encoder to 281 dimensions.

7. The lithium battery state-of-charge assessment method based on charging curve reconstruction according to claim 6, characterized in that, The GRU network model of the charging curve reconstruction model includes a recurrent unit. This recurrent unit receives the input and the hidden state from the previous time step at each time step, and generates the output and a new hidden state for the current time step. The update gate's corresponding formula is: ; in, yes Update the gate output constantly. It is an activation function. It updates the weight matrix of the gate. It is the hidden state from the previous moment. for Input at any time; The formula for resetting the door is: ; in, yes Reset the gate output at all times. It is an activation function. It is the weight matrix of the reset gate. It is the hidden state from the previous moment. for Input at any time; The new candidate hidden state is: ; in, It is a new candidate hidden state. It is an activation function. It is the weight matrix of the new candidate hidden states. yes Reset the gate output at all times. It is the hidden state from the previous moment. for Input at any time; The hidden state of the current time step is: ; in, yes Hide your status at all times. yes Update the gate output constantly. It is a new candidate hidden state. It is the hidden state from the previous moment; The corresponding formula for output calculation is: ; in, yes Time output, It is an activation function. It is the weight matrix from the hidden state to the output. It is the hidden state at the current time step. It is the bias of the output layer.

8. The lithium battery state-of-charge assessment method based on charging curve reconstruction according to claim 1, 2, 4, or 7, characterized in that, In step S6, calculating the battery capacity based on the charging curve reconstruction model includes: estimating the battery capacity using the data from the reconstruction curve and the ampere-hour integration method.

Images