Battery capacity estimation method, system and device and storage medium

Abstract

The invention discloses a battery capacity estimation method, system and device and a storage medium, and relates to the technical field of battery capacity estimation, and the method comprises the steps: collecting the charging current, charging voltage, charging time duration and charging temperature data of a to-be-detected battery in a constant-voltage charging process in a constant-current and constant-voltage charging mode, and constructing a battery constant-voltage charging data set; constructing a battery capacity estimation model based on a convolutional neural network CNN, an attention mechanism and a long short-term memory recurrent neural network LSTM; inputting the constant-voltage charging data set of the battery to be detected into the battery capacity estimation model to obtain an estimated value of the battery capacity; and the problem of insufficient utilization of time, space and importance information among the data sequences is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of battery capacity estimation, and in particular to a battery capacity estimation method, system, device and storage medium. Background Art

[0002] With the development of the energy storage industry and energy storage devices, lithium-ion batteries, with their high energy density, long cycle life, and low self-discharge, are gaining popularity in the market. In particular, the rapidly developing new energy vehicle industry is seeing increasing demand for ternary lithium and lithium iron phosphate batteries. However, this comes with the challenge of achieving safe and effective battery management, of which accurate estimation of lithium-ion battery capacity is a key component. Lithium-ion battery capacity is not only a crucial parameter for estimating the battery's state of charge (SOC), but also a key indicator for estimating battery life. Therefore, it plays a crucial role in assessing the battery's health.

[0003] Battery capacity estimation can be broadly categorized into model-based and data-driven approaches. Model-based approaches require building a precise battery model. However, due to the complex chemical reactions within batteries and the complex operating environment, establishing a highly accurate estimation model is extremely difficult. With the advancement of artificial intelligence, directly estimating battery capacity using collected data such as battery terminal voltage, current, and temperature is becoming an increasingly important option. For example, patent CN201911312016, a method for online rapid capacity estimation of lithium-ion batteries based on capacity increment analysis, uses charging data to extract the characteristics of the capacity increment curve, and then uses the extracted characteristics to establish a model related to the battery capacity to estimate the battery capacity; patent CN202310868393, a method for capacity estimation of lithium batteries in energy storage power stations based on multi-feature combination, uses charging data to extract the current average value, standard deviation, and the slope of the straight line at the starting point and end point of the current curve as feature quantities, and then inputs these feature quantities into a neural network for battery capacity estimation; patent CN202310082075, a method for estimating the health status of lithium batteries based on feature selection and temporal attention, uses nine parameters, including the maximum, minimum, and average values of the charging current in a specific interval during the charging and discharging process, the maximum, minimum, and average values of the voltage during the discharge process, and the maximum, minimum, and average values of the temperature during the discharge process, as feature values, and then inputs them into a hybrid neural network to estimate the battery capacity. All of the above are achieved by extracting corresponding features of the current, voltage, and temperature during the charging and discharging process, analyzing the correlation between the features and the battery capacity, and using the features with strong correlation as parameters of the neural network or model to estimate the battery capacity.

[0004] However, the existing technology for estimating battery capacity based on data-driven approach highly depends on the accurate extraction of eigenvalues, which will increase the technical complexity and inadequately utilize the correlation between different data sequences and the importance corresponding to different data, being unfavorable for improving the accuracy of battery capacity estimation. Summary of the Invention

[0005] Aiming at the deficiency of the existing technology in inadequately utilizing the correlation between different data sequences and the importance corresponding to different data, which is unfavorable for improving the accuracy of battery capacity estimation, the present invention proposes a battery capacity estimation method, system, device and storage medium. A battery capacity estimation model is constructed based on a hybrid neural network model. The features of constant voltage charging data are automatically extracted through a CNN network, and the correlation between feature sequence data is extracted by combining an LSTM and an attention mechanism, thus solving the problem that the existing technology inadequately utilizes the correlation between different data sequences and the importance corresponding to different data, being unfavorable for improving the accuracy of battery capacity estimation.

[0006] A battery capacity estimation method includes the following steps:

[0007] Collect the charging current, charging voltage, charging time length and charging temperature data during the constant voltage charging process in the constant current and constant voltage charging mode of the battery to be detected, and construct a battery constant voltage charging data set;

[0008] Construct a battery capacity estimation model based on a convolutional neural network CNN, two attention mechanism layers and a long short-term memory recurrent neural network LSTM;

[0009] Input the battery constant voltage charging data set into the battery capacity estimation model to obtain an estimated value of the battery capacity;

[0010] Among them, the step of inputting the battery constant voltage charging data set into the battery capacity estimation model to obtain an estimated value of the battery capacity includes the following steps:

[0011] Input the battery constant voltage charging data set into the CNN to extract a feature sequence;

[0012] Input the feature sequence into the first attention mechanism layer to calculate the weights corresponding to the feature sequence, and output a corrected feature sequence;

[0013] Input the corrected feature sequence into the long short-term memory recurrent neural network LSTM to extract the correlation between feature sequence data, and output a sequence (h t ,h t-1 ,···h t-m );

[0014] Input the sequence (h t ,h t-1 ,···h t-m) In the second attention mechanism layer, the correlations between sequences (h t , h t-1 , ··· h t-m ) at different times are calculated, and different weight values are given to output the corrected sequence

[0015] According to the sequence The capacity of the battery is calculated through a fully connected layer.

[0016] Further, the battery constant voltage charging dataset is collected by the battery under different charge and discharge cycle numbers.

[0017] Further, the battery constant voltage charging dataset includes: a voltage dataset, which is expressed as: V cv_M = [v cv_M (1), v cv_M (2), ··· v cv_M (N)]; a current dataset, which is expressed as: C cv_M = [c cv_M (1), c cv_M (2), ··· c cv_M (N)]; a charging time dataset, which is expressed as: Ti cv_M = [ti cv_M (1), ti cv_M (2), ··· ti cv_M (N)]; a charging temperature dataset, which is expressed as: Te cv_M = [te cv_M (1), te cv_M (2), ··· te cv_M (N)]; where, V cv_M represents the voltage dataset corresponding to the Mth constant voltage charging, N represents the length of the dataset, C cv_M represents the current dataset of the Mth constant voltage charging, Ti cv_M represents the charging time dataset during the Mth constant voltage charging, and Te cv_M represents the charging temperature dataset during the Mth constant voltage charging.

[0018] Further, the step of inputting the feature sequence F into the first attention mechanism layer to calculate the weight corresponding to the feature sequence and output the corrected feature sequence specifically includes the following steps:

[0019] Perform max pooling MaxPool and average pooling AvgPool operations on the input feature sequence on each channel;

[0020] Input the results of max pooling and average pooling into a multi-layer perceptron MLP respectively to learn the channel dimension features and the importance of each channel;

[0021] The two C×1×1 vectors output by the MLP are concatenated and processed through the Sigmoid function mapping to obtain the final channel attention weight M c (F), and its expression is:

[0022]

[0023] where W1 is the weight of the second layer of the multi-layer perceptron, and W0 is the weight of the first layer of the multi-layer perceptron, is the average pooling feature among channels, is the maximum pooling feature among channels;

[0024] Using the channel attention weight M c (F) and F are calculated to obtain the channel feature F′, and its calculation formula is:

[0025] Perform maximum pooling and average pooling operations on the channel feature F′;

[0026] The two results of maximum pooling and average pooling are concatenated, a convolution operation is performed using a convolutional neural network, and the result of the convolution is passed through the Sigmoid function to generate the spatial attention weight M s (F), which is expressed as:

[0027]

[0028] where f n×n represents a convolutional neural network with a size of n×n, and σ represents the Sigmoid function, is the average pooling feature in the space, is the maximum pooling feature in the space;

[0029] Using the spatial attention weight M s (F) is multiplied by F′ to obtain the corrected feature F″.

[0030] Furthermore, the Relu function is used as the activation function of the convolutional neural network CNN.

[0031] Furthermore, the battery capacity is calculated through the fully connected layer according to the sequence , and its calculation formula is:

[0032]

[0033] where C battery is the estimated battery capacity, ω i$w_i$ is the weight value of each neuron in the fully connected neural network, where $i$ represents the $i$-th neuron in the fully connected layer and $m$ represents the total number of neurons in the fully connected layer.

[0034] Further, the long short-term memory recurrent neural network LSTM is trained using the Adam gradient descent algorithm.

[0035] Further, a battery capacity estimation system includes:

[0036] An acquisition module, configured to acquire charging current, charging voltage, charging time length, and charging temperature data during the constant voltage charging process in the constant current and constant voltage charging mode of the battery to be detected, and construct a battery constant voltage charging data set;

[0037] A model construction module, configured to construct a battery capacity estimation model based on a convolutional neural network CNN, two attention mechanism layers, and a long short-term memory recurrent neural network LSTM;

[0038] An estimation module, configured to input the battery constant voltage charging data set of the battery to be detected into the battery capacity estimation model to obtain an estimated value of the battery capacity; wherein, inputting the battery constant voltage charging data set of the battery to be detected into the battery capacity estimation model to obtain an estimated value of the battery capacity; the estimation module includes:

[0039] A first feature extraction unit, configured to input the battery constant voltage charging data set into the CNN to extract a feature sequence;

[0040] A first correction unit, configured to input the feature sequence into the first attention mechanism layer to calculate the weight corresponding to the feature sequence, and output a corrected feature sequence;

[0041] A second feature extraction unit, configured to input the corrected feature sequence into the long short-term memory recurrent neural network LSTM to extract the correlation between the feature sequence data, and output a sequence ($h$ t , $h$ t-1 , ···, $h$ t-m );

[0042] A second correction unit, configured to input the sequence ($h$ t , $h$ t-1 , ···, $h$ t-m ) into the second attention mechanism layer, calculate the correlation between different time sequences ($h$ t , $h$ t-1 , ···, $h$ t-m ), and give different weight values, and output a corrected sequence

[0043] A calculation unit, configured to calculate the capacity of the battery according to the sequence through a fully connected layer.

[0044] Further, a computer device for estimating battery capacity includes: a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, the steps of the battery capacity estimation method are implemented.

[0045] Further, a readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the battery capacity estimation method.

[0046] The present invention provides a battery capacity estimation method, system, device, and storage medium, which have the following beneficial effects:

[0047] In the present invention, a battery capacity estimation model is constructed based on a convolutional neural network (CNN), two attention mechanism layers, and a long short-term memory recurrent neural network (LSTM). The CNN automatically extracts a feature sequence from the input battery constant voltage charging data, and by combining LSTM and the attention mechanism, the correlation between the feature sequence data is extracted, reducing the complexity of implementing the battery capacity estimation technology. At the same time, it avoids the problem of insufficient utilization of time, space, and importance information between data sequences, thereby improving the accuracy of battery capacity estimation. Description of the Drawings

[0048] Figure 1 It is an architecture diagram of the battery capacity estimation model in an embodiment of the present invention;

[0049] Figure 2 It is a flowchart of the battery capacity estimation method in an embodiment of the present invention;

[0050] Figure 3 It is a terminal voltage curve diagram of the battery during the constant current-constant voltage charging process at different cycle charge and discharge times in an embodiment of the present invention;

[0051] Figure 4 It is a charging current curve of the battery during the constant current-constant voltage charging process at different cycle charge and discharge times in an embodiment of the present invention;

[0052] Figure 5 It is a structure diagram corresponding to the LSTM network in an embodiment of the present invention;

[0053] Figure 6 It is a principle block diagram of the first attention mechanism layer in an embodiment of the present invention;

[0054] Figure 7 It is a principle block diagram of the channel attention mechanism (CAM) in an embodiment of the present invention;

[0055] Figure 8This is the principle block diagram of the spatial attention mechanism (SAM) in the embodiments of the present invention;

[0056] Figure 9 This is the principle block diagram of the second attention mechanism layer in the embodiments of the present invention. Detailed implementation manners

[0057] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0058] As Figure 1 shown, the present invention proposes a battery capacity estimation method, including the following steps:

[0059] 1. Obtain datasets such as charging current (as Figure 4 shown), charging voltage (as Figure 3 shown), charging time length, charging temperature, discharge capacity, etc. of the lithium-ion battery under different charge and discharge cycle numbers.

[0060] 2. In the obtained datasets, select the charging current, charging voltage, charging time length, and charging temperature during the constant voltage charging process in the constant current-constant voltage charging mode as the alternative datasets of the input data.

[0061] 3. Select the datasets with a length of N under different cycle numbers in the alternative datasets as the formal datasets. The voltage dataset can be expressed as: V cv_M =[v cv_M (1), v cv_M (2), ··· v cv_M (N)]; the current dataset can be expressed as: C cv_M =[c cv_M (1), c cv_M (2), ··· c cv_M (N)]; the charging time dataset can be expressed as: Ti cv_M =[ti cv_M (1), ti cv_M (2), ··· ti cv_M (N)]; the charging temperature dataset can be expressed as: Te cv_M =[te cv_M (1), te cv_M (2), ··· te cv_M (N)]. In the above datasets, V cv_M represents the voltage dataset corresponding to the Mth constant voltage charging, and N represents the length of the dataset; in the above datasets, C cv_M represents the Mth constant voltage charging current dataset, and N represents the length of the dataset; Ti cv_MDenote the charging time data set during the M-th constant voltage charging process, and N represents the length of the data set; Te cv_M Denote the charging temperature data set during the M-th constant voltage charging process, and N represents the length of the data set.

[0062] 4. Concatenate the data sets {V cv , C cv , Ti cv , Te cv} together and perform normalization processing.

[0063] 5. Input the normalized data set into the hybrid neural network as shown in Figure 1 . Use the convolutional neural network (CNN) in the hybrid neural network to automatically extract features. For the used CNN network, the size and number of its filters are comprehensively determined according to the length of the input data sequence, and the Relu function is used as the activation function.

[0064] 6. Input the feature data F ∈ R C×H×W (where C represents the number of channels, and H and W represent the height and width of the feature data respectively) into the first layer of the attention mechanism for channel attention mechanism (CAM) calculation. The principle block diagram is as shown in Figure 7 . The calculation expression is as shown in Equation (1). In the equation, for the input feature data F, first perform max pooling MaxPool and average pooling AvgPool operations on each channel, and then send the results into a multi-layer perceptron MLP (usually composed of two fully connected layers) respectively to learn the channel dimension features and the importance of each channel. Finally, connect the two C*1*1 vectors output by the MLP together and perform mapping processing through the Sigmoid function to obtain the final channel attention weight M c (F)

[0065]

[0066] 7. Use the channel attention weight M c (F) to calculate with the input feature F to obtain the channel feature: Then perform spatial attention mechanism (SAM) calculation on the channel feature F′. The principle block diagram is as shown in Figure 8 . The calculation process is as shown in Equation (2). In this equation, first perform max pooling and average pooling operations on F′, then connect the two results together, perform convolution operations using a convolutional neural network, and generate the spatial attention weight M s (F) through the Sigmoid function, where f n×n represents a convolutional neural network with a size of n×n, and σ represents the Sigmoid function.

[0067]

[0068] 8. Use the spatial attention weight M s (F) is multiplied by F′ to obtain the corrected feature F″.

[0069] 9. Flatten the corrected feature sequence F″, and then input it into a long short-term memory recurrent neural network (the LSTM structure is as Figure 5 shown), and use the LSTM network to extract the correlation between the input feature data (including temporal and spatial relationships), and output through (h t , h t-1 , ··· h t-m ). The Adam gradient descent algorithm is used during the training process of the LSTM network.

[0070] 10. Input the sequence (h t , h t-1 , ··· h t-m ) output by the LSTM network into the second attention mechanism layer (as Figure 9 shown). This layer calculates the correlation between different sequences (h t , h t-1 , ··· h t-m ) and gives different weight values, and finally outputs the corrected

[0071] 11. Use the output by the attention mechanism layer to obtain the capacity of the battery through a fully connected layer. The calculation method is: where C battery is the estimated battery capacity, and ω i is the weight value of each neuron in the fully connected neural network.

[0072] Based on the same inventive concept, the present invention proposes a battery capacity estimation system, including:

[0073] A collection module, configured to collect the charging current, charging voltage, charging time length, and charging temperature data during the constant voltage charging process in the constant current and constant voltage charging mode of the battery to be detected, and construct a battery constant voltage charging data set.

[0074] A model construction module, configured to construct a battery capacity estimation model based on a convolutional neural network CNN, two attention mechanism layers, and a long short-term memory recurrent neural network LSTM.

[0075] An estimation module for inputting a constant voltage charging data set of a battery to be detected into a battery capacity estimation model to obtain an estimated value of the battery capacity; wherein, inputting the constant voltage charging data set of the battery to be detected into the battery capacity estimation model to obtain the estimated value of the battery capacity; the estimation module includes:

[0076] A first feature extraction unit for inputting the battery constant voltage charging data set into a CNN to extract a feature sequence.

[0077] A first correction unit for inputting the feature sequence into the first attention mechanism layer to calculate the weights corresponding to the feature sequence and output a corrected feature sequence.

[0078] A second feature extraction unit for inputting the corrected feature sequence into a long short-term memory recurrent neural network (LSTM) to extract the correlation between the feature sequence data and output a sequence (h t , h t-1 , ··· h t-m ).

[0079] A second correction unit for inputting the sequence (h t , h t-1 , ··· h t-m ) into the second attention mechanism layer, calculating the correlation between the sequences (h t , h t-1 , ··· h t-m ) at different times and giving different weight values, and outputting a corrected sequence

[0080] A calculation unit for calculating the capacity of the battery according to the sequence through a fully connected layer.

[0081] Based on the same inventive concept, the present invention also provides a computer device for estimating battery capacity, including: a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, the steps of the battery capacity estimation method are implemented.

[0082] Based on the same inventive concept, the present invention also provides a readable storage medium storing a computer program. The computer program includes program instructions, and when the program instructions are executed by a processor, they are used to execute the steps of the battery capacity estimation method.

[0083] As mentioned above, the above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for estimating battery capacity, characterized in that, Including the following steps: Collect the charging current, charging voltage, charging time length, and charging temperature data during the constant voltage charging process of the battery to be detected in the constant current and constant voltage charging mode, and construct a battery constant voltage charging data set; Construct a battery capacity estimation model based on the convolutional neural network CNN, two attention mechanism layers, and the long short-term memory recurrent neural network LSTM; Input the battery constant voltage charging data set of the battery to be detected into the battery capacity estimation model to obtain an estimated value of the battery capacity; Among them, the step of inputting the battery constant voltage charging data set of the battery to be detected into the battery capacity estimation model to obtain an estimated value of the battery capacity includes the following steps: Input the battery constant voltage charging data set into the CNN to extract a feature sequence; Input the feature sequence into the first attention mechanism layer to calculate the weight corresponding to the feature sequence, and output a corrected feature sequence; Input the corrected feature sequence into a long short-term memory recurrent neural network (LSTM) to extract the correlation between the feature sequence data, and output the sequence (h t , h t-1 , ··· h t-m ); Input the sequence (h t , h t-1 , ··· h t-m ) into the second attention mechanism layer, calculate the correlation between sequences (h t , h t-1 , ··· h t-m ) at different times, and give different weight values to output the corrected sequence According to the sequence The capacity of the battery is calculated through the fully connected layer.

2. The battery capacity estimation method according to claim 1, wherein The battery constant voltage charging data set is collected by the battery under different charge and discharge cycle numbers.

3. A method for estimating battery capacity according to claim 1, wherein, The battery constant-voltage charging dataset includes: a voltage dataset, which is expressed as: V cv-M =[v cv-M (1), v cv-M (2), ··· v cv-M (N)]; a current dataset, which is expressed as: C cv-M =[c cv-M (1), c cv-M (2), ··· c cv-M (N)]; a charging time dataset, which is expressed as: Ti cv-M =[ti cv-M (1), ti cv-M (2), ··· ti cv-M (N)]; a charging temperature dataset, which is expressed as: Te cv-M =[te cv-M (1), te cv-M (2), ··· te cv-M (N)]; where, V cv-M represents the voltage dataset corresponding to the M-th constant-voltage charging, N represents the length of the dataset, C cvM represents the current dataset of the M-th constant-voltage charging, Ti cv_M represents the charging time dataset during the M-th constant-voltage charging, and Te cv_M represents the charging temperature dataset during the M-th constant-voltage charging.

4. A battery capacity estimation method according to claim 1, wherein The step of inputting the feature sequence F into the first attention mechanism layer to calculate the weight corresponding to the feature sequence and output a corrected feature sequence specifically includes the following steps: Perform maximum pooling MaxPool and average pooling AvgPool operations on the input feature sequence on each channel; Input the results of maximum pooling and average pooling into a multi-layer perceptron MLP respectively to learn the channel dimension features and the importance of each channel; The two C*1*1 vectors output by the MLP are concatenated and processed through the Sigmoid function mapping to obtain the final channel attention weight M c (F), and its expression is: Among them, W1 is the weight of the second layer of the multi-layer perceptron, and W0 is the weight of the first layer of the multi-layer perceptron. is the average pooling feature between channels, is the maximum pooling feature between channels; Using the channel attention weight M c (F) is calculated with F to obtain the channel feature F′, and its calculation formula is: Perform maximum pooling and average pooling operations on the channel feature F'; Concatenate the results of max pooling and average pooling, perform convolution operations using a convolutional neural network, and generate spatial attention weight M through the convolution result using the Sigmoid function s (F), which is expressed as: where f n×n represents a convolutional neural network of size n×n, and σ represents the Sigmoid function, is the average pooling feature in the space, is the max pooling feature in the space; Utilize the spatial attention weight M s (F) is multiplied by F′ to obtain the corrected feature F″.

5. A method for estimating battery capacity according to claim 1, characterized in that, Use the Relu function as the activation function of the convolutional neural network CNN.

6. The battery capacity estimation method according to claim 1, wherein The said according to the sequence The capacity of the battery is calculated through a fully connected layer, and its calculation formula is: Among them, C battery is the estimated battery capacity, ω i is the weight value of each neuron in the fully connected neural network. i represents the i-th neuron in the fully connected layer, and m represents the total number of neurons in the fully connected layer.

7. A method for estimating battery capacity according to claim 1, characterized in that, Use the Adam gradient descent algorithm to train the long short-term memory recurrent neural network LSTM.

8. A battery capacity estimation system, characterized in that Including: A collection module for collecting the charging current, charging voltage, charging time length, and charging temperature data during the constant voltage charging process of the battery to be detected in the constant current and constant voltage charging mode, and constructing a battery constant voltage charging data set; A model construction module for constructing a battery capacity estimation model based on the convolutional neural network CNN, two attention mechanism layers, and the long short-term memory recurrent neural network LSTM; An estimation module for inputting the battery constant voltage charging data set of the battery to be detected into the battery capacity estimation model to obtain an estimated value of the battery capacity; among them, the step of inputting the battery constant voltage charging data set of the battery to be detected into the battery capacity estimation model to obtain an estimated value of the battery capacity; the estimation module includes: A first feature extraction unit for inputting the battery constant voltage charging data set into the CNN to extract a feature sequence; A first correction unit for inputting the feature sequence into the first attention mechanism layer to calculate the weight corresponding to the feature sequence and output a corrected feature sequence; The second feature extraction unit is used to input the corrected feature sequence into a long short-term memory recurrent neural network (LSTM) to extract the correlation between the feature sequence data, and output a sequence (h t , h t-1 , ··· h t-m ); The second correction unit is used to input the sequence (h t , h t-1 , ··· h t-m ) into the second attention mechanism layer, calculate the correlation between sequences (h t , h t-1 , ··· h t-m ) at different times, and give different weight values, and output the corrected sequence A calculation unit for calculating the capacity of a battery based on a sequence through a fully connected layer.

9. A computer device for estimating battery capacity, characterized in that, Including: A memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, the steps of the battery capacity estimation method according to any one of claims 1-7 are implemented.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed by the processor, they are used to execute the steps of the method for battery capacity estimation according to any one of claims 1-7.

Images