Lithium ion battery health state estimation method and system based on physical domain characteristics and CNN-EFC-BiLSTM

Abstract

The invention discloses a lithium ion battery health state estimation method and system based on physical domain characteristics and CNN-EFC-BiLSTM, and the method comprises the steps: measuring the constant-voltage charging health state data of each charging and discharging cycle of a lithium ion battery, and forming a data sequence; extracting physical domain features from the data sequence as health indexes to form a physical domain feature sequence; establishing a health state estimation model based on CNN-EFC-BiLSTM on the basis of the health state data and health indexes of the lithium ion battery; and after physical domain characteristics are extracted from the data of the constant-voltage charging stage of the lithium ion battery, the health state result of the lithium ion battery is calculated through the CNN-EFC-BiLSTM model. The lithium ion battery health state estimation method has better accuracy, rapidness and robustness for lithium ion battery health state estimation, and lithium ion battery health state estimation can be carried out more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion batteries, and in particular to a method and system for estimating the health state of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM. Background Art

[0002] Electric vehicles have the advantages of environmental protection, high energy efficiency, and low operating costs. The popularity of electric vehicles helps reduce carbon emissions, and thus they are widely used in personal transportation, public transportation, commercial logistics, and special-purpose vehicles. As the core component of electric vehicles, lithium-ion batteries have become a key indicator for evaluating the degree of battery aging during continuous recycling. Since the safety of batteries is closely related to their health status, a series of safety issues may occur when the battery is aged or damaged, which may lead to catastrophic events such as explosions and fires. Therefore, accurate and rapid estimation of the battery health status can effectively reduce the occurrence of a series of safety issues caused by batteries.

[0003] In recent years, lithium-ion battery health state estimation methods usually combine deep learning models such as CNN, BiLSTM, and Attention. For example, the CNN model is used to extract abstract features of time and space from the charge and discharge data of lithium-ion batteries, and the time series relationship of the data is analyzed by BiLSTM, which has become an important method for estimating the health state of lithium-ion batteries. The CNN model can effectively capture the abstract features in the battery charge and discharge data, while BiLSTM processes the historical and future data of the battery through a bidirectional information flow, which can more accurately reflect the changing trends in the time series. This combination allows the model to deeply explore the complex patterns in the data and improve the accuracy of the estimation of the health state of lithium-ion batteries.

[0004] However, existing technologies have certain limitations in feature extraction. Relying solely on CNN models to extract abstract features of time and space may not be able to fully capture the more complex physical and chemical changes in the aging process of lithium-ion batteries. In addition, these models are highly dependent on high-quality charging and discharging data, which may limit the generalization ability of the model due to problems such as high data acquisition costs and limited data volume. At the same time, although the Attention mechanism improves the ability to weight important features, some key battery aging features may still not be fully captured, resulting in affected prediction accuracy. Therefore, the existing technology needs to be further improved in feature extraction to enhance the ability to model the complex lithium-ion battery health state decay process, while improving data utilization efficiency and model accuracy. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for estimating the health state of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM, so as to more accurately estimate the health state of a lithium-ion battery.

[0006] To achieve the above purpose, the technical solution provided by the present invention is:

[0007] The first aspect of the present application provides a method for estimating the health state of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM, comprising the following steps:

[0008] Measure the constant voltage charging health status data of each charge and discharge cycle of the lithium-ion battery to form a data sequence;

[0009] Extracting physical domain features from the data sequence as health indicators to form a physical domain feature sequence;

[0010] Based on the health status data and health indicators of lithium-ion batteries, a health status estimation model based on CNN-EFC-BiLSTM is established; the health status estimation model based on CNN-EFC-BiLSTM is as follows: an EFC weight distribution layer for capturing key change trends is added after the CNN model to form a CNN-EFC layer, and a BiLSTM layer for calculating long-term dependencies of sequences is set after the CNN-EFC layer;

[0011] After extracting the physical domain features from the data of the constant voltage charging stage of the lithium-ion battery, the health status of the lithium-ion battery is calculated through the CNN-EFC-BiLSTM model.

[0012] The physical domain characteristics include: thermal characteristics, time characteristics and electrical characteristics; wherein the thermal characteristics refer to the average charging temperature of the lithium-ion battery, the time characteristics refer to the current trajectory length, and the electrical characteristics refer to the incremental energy.

[0013] Furthermore, the thermal feature extraction method is as follows:

[0014] The average charging temperature of lithium-ion batteries is extracted as a thermal feature, and the temperature data of the constant voltage charging stage of each charge and discharge cycle is obtained. The average value is calculated as follows:

[0015]

[0016] Among them, T avg Indicates the average charging temperature, T1 to T n Represents the temperature of each sampling point, and n represents the number of sampling points.

[0017] Furthermore, the method for extracting the time feature is as follows:

[0018] The length of the current trajectory is extracted as the time feature, and the current data and charging time of the constant voltage charging stage of each charge and discharge cycle are obtained. The length of the curve trajectory of the current changing with time is calculated as follows:

[0019]

[0020] Where L is the length of the current trajectory, f(x) is the expression of the current changing with time, and a and b represent the start and end time of the constant voltage charging stage, respectively.

[0021] Furthermore, the method for extracting the electrical characteristics is as follows:

[0022] The incremental energy is extracted as the electrical feature to obtain the energy data of the constant voltage charging stage of each charge and discharge cycle, and the difference between the end energy and the start energy is calculated using the following formula:

[0023] ΔE=E end -E start

[0024] Where ΔE represents the incremental energy, E end and E start are the ending energy and the starting energy respectively.

[0025] Furthermore, the algorithm of the CNN-EFC layer includes:

[0026] Convolutional layer: For a one-dimensional input x(u) and a convolution kernel (filter) f(s), where u is the index in time or space and s is the index within the convolution kernel, the one-dimensional convolution operation is expressed as follows:

[0027]

[0028] Wherein, y(u) represents the output after the convolution operation, that is, the feature map; each filter in the convolution layer performs the one-dimensional convolution operation to generate a corresponding feature map;

[0029] Activation function: Select Relu as the activation function. The formula of the Relu activation function is as follows:

[0030] h(u)=max(0,y(u))

[0031] Among them, h(u) represents the output after y(u) is activated. If the convolution output y(u) is greater than 0, the output of the activation function is y(u); if it is less than or equal to 0, the output is 0;

[0032] Max pooling layer: The max pooling operation selects the maximum value in the set range as the representative of the range. For an activated feature map h(u) and pooling window size p, the max pooling operation is expressed as:

[0033]

[0034] Among them, m(u) is the output after the pooling operation, and the max operation is applied in the index s ranging from u to u+p-1, indicating that the maximum value is selected within each pooling window.

[0035] Furthermore, the algorithm of the EFC weight distribution layer includes:

[0036] To calculate the weight score, the feature vector m(u) of each time step u is converted into a weight score through a fully connected layer. The process is expressed as:

[0037] z(u)=W·m(u)+b

[0038] Among them, z(u) represents the weight score, W and b are the weight and bias of the fully connected layer respectively;

[0039] Use the Softmax function to convert to actual weights: The actual weight for each time step is calculated as follows:

[0040]

[0041] Among them, a(u) represents the actual weight, U is the total number of time steps, and the sum of the weights of all time steps is ensured to be 1;

[0042] By generating a weighted output, the original input m(u) is weighted using the obtained weight a(u) to obtain a weighted result. For each time step, the weighted result m′(u) is calculated as follows:

[0043] m′(u)=a(u)·m(u).

[0044] The second aspect of the present application provides a lithium-ion battery health state estimation system based on physical domain features and CNN-EFC-BiLSTM, including:

[0045] The battery health status data acquisition module is used to measure the constant voltage charging health status data of each charge and discharge cycle of the lithium-ion battery to form a data sequence;

[0046] A physical domain feature extraction module is used to extract physical domain features from a data sequence as health indicators to form a physical domain feature sequence;

[0047] A health state estimation model building module is used to establish a health state estimation model based on CNN-EFC-BiLSTM based on the health state data and health indicators of the lithium-ion battery; the health state estimation model based on CNN-EFC-BiLSTM is: an EFC weight distribution layer for capturing key change trends is added after the CNN model to form a CNN-EFC layer, and a BiLSTM layer for calculating the long-term dependency of the sequence is set after the CNN-EFC layer;

[0048] The battery health status determination module is used to extract physical domain features from the constant voltage charging stage data of the lithium-ion battery, and then calculate the health status results of the lithium-ion battery through the CNN-EFC-BiLSTM model.

[0049] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for estimating the health status of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM as provided in the first aspect of the present application is implemented.

[0050] The fourth aspect of the present application provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the lithium-ion battery health status estimation method based on physical domain features and CNN-EFC-BiLSTM as provided in the first aspect of the present application.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] The present invention proposes a comprehensive method for extracting multiple physical features based on constant voltage charging stage data, comprehensively considering various influencing factors of lithium-ion batteries in the constant voltage charging stage, and extracting multiple physical features based on this data, including thermal features, time features and electrical features. It has been verified that multiple physical features have a strong correlation with the battery health status, which can more comprehensively reflect the subtle internal changes of the battery during the aging process, and can establish a more accurate mapping model with the battery health status through deep learning.

[0053] The present invention proposes a CNN-EFC-BiLSTM model, which utilizes the unique ability of the CNN model to extract sequence periodicity and key change trends, and considering the disadvantage of the CNN model having instability, adds an EFC weight distribution layer after the CNN model to form a CNN-EFC model structure, alleviates the instability problem caused by the hyperparameter setting of the CNN model, enhances the ability to combine the original feature sequence, and reduces the calculation burden for subsequent models; after the CNN-EFC model structure, a BiLSTM layer with the ability to capture the long-term dependency of the sequence is utilized, and the model can keenly capture the change trend of the battery health state by calculating the forward and reverse directions of the sequence respectively; therefore, the CNN-EFC-BiLSTM model can demonstrate excellent performance in the field of battery health state estimation, including accuracy, rapidity and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the lithium-ion battery health state estimation method based on physical domain features and CNN-EFC-BiLSTM of the present invention.

[0055] Figure 2 It is a current trajectory curve diagram of different cycle numbers at 0.5C rate in an embodiment of the present invention.

[0056] Figure 3 It is a curve diagram of the change of the length of each circulating current trajectory at a rate of 0.5C in an embodiment of the present invention.

[0057] Figure 4 It is a curve diagram of the change of each cycle incremental energy at 0.5C rate in the embodiment of the present invention.

[0058] Figure 5 This is a diagram of the SOH estimation results of the CNN-EFC-BiLSTM model at a 0.5C rate in an embodiment of the present invention.

[0059] Figure 6 It is a comparison chart of SOH estimation results of different models at 0.5C rate in the embodiment of the present invention. DETAILED DESCRIPTION

[0060] The above contents of the present invention are further described in detail below in the form of embodiments, but this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the above contents of the present invention belong to the scope of the present invention.

[0061] Lithium-ion batteries are generally charged in a constant-current and constant-voltage manner, which is divided into two stages: constant-current charging and constant-voltage charging. Considering that electric vehicles generally have a non-zero charge during charging, it is impossible to obtain complete charging data of the electric vehicle power battery. The present invention flexibly uses the constant-voltage charging stage data to extract health factors, and combines the excellent performance of the CNN-EFC-BiLSTM deep learning model to greatly improve the accuracy of battery health status estimation and expand its application range.

[0062] In one embodiment of the present invention, a method for estimating the health status of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM is proposed. Figure 1 As shown, the following steps are included:

[0063] Measure the constant voltage charging health status data of each charge and discharge cycle of the lithium-ion battery to form a data sequence;

[0064] Extracting physical domain features from the data sequence as health indicators to form a physical domain feature sequence;

[0065] Based on the health status data and health indicators of lithium-ion batteries, a health status estimation model based on CNN-EFC-BiLSTM is established; the health status estimation model based on CNN-EFC-BiLSTM is as follows: an EFC weight distribution layer for capturing key change trends is added after the CNN model to form a CNN-EFC layer, and a BiLSTM layer for calculating the long-term dependency of the sequence is set after the CNN-EFC layer;

[0066] After extracting the physical domain features from the data of the constant voltage charging stage of the lithium-ion battery, the health status of the lithium-ion battery is calculated through the CNN-EFC-BiLSTM model.

[0067] The constant voltage charging health status data of each charge and discharge cycle of the lithium-ion battery includes: charging temperature, charging current, charging time and charging energy data during the constant voltage charging stage.

[0068] Measuring the constant voltage charging health status data of each charge and discharge cycle of the lithium-ion battery to form a data sequence, including: using the lithium-ion battery health status data of each charge and discharge cycle as a sequence H1, H2, ..., H n express, Among them, H i is the battery health status of the i-th (i=1, 2, ..., n) charge-discharge cycle, n is the number of charge-discharge cycles, C i is the maximum discharge capacity of the lithium-ion battery in the i-th charge and discharge cycle, and C is the rated capacity of the lithium-ion battery;

[0069] The temperature data of the constant voltage charging stage of the i-th charge and discharge cycle is the sequence The current data of the constant voltage charging stage of the i-th charge and discharge cycle is the sequence The energy data of the constant voltage charging stage of the i-th charge and discharge cycle is the sequence Among them, T i，0 is the initial charging temperature of the constant voltage charging stage of the i-th charge and discharge cycle, T i，t is the charging temperature sampled after the interval t in the constant voltage charging phase of the i-th charge and discharge cycle, is the termination charging temperature of the constant voltage charging stage of the i-th charge and discharge cycle; I i，0 is the initial current of the constant voltage charging stage of the i-th charge and discharge cycle, I i，t is the charging current sampled after the interval t in the constant voltage charging phase of the i-th charge and discharge cycle, is the cut-off current of the constant voltage charging stage of the i-th charge and discharge cycle; E i，0 is the initial energy of the constant voltage charging stage of the i-th charge and discharge cycle, E i，t is the charging energy sampled after the interval t in the constant voltage charging phase of the i-th charge and discharge cycle, is the termination energy of the constant voltage charging stage of the i-th charge-discharge cycle; t is the data sampling interval during charging, z i is the sampling number of the constant voltage charging stage in the i-th charge and discharge cycle, z i ×t is the total sampling time of the constant voltage charging stage of the i-th charge and discharge cycle.

[0070] Physical domain characteristics include: thermal characteristics, time characteristics and electrical characteristics; among them, thermal characteristics refer to the average charging temperature of lithium-ion batteries, time characteristics refer to the length of current trajectory, and electrical characteristics refer to incremental energy.

[0071] The thermal feature extraction method is as follows:

[0072] The average charging temperature of lithium-ion batteries is extracted as a thermal feature, and the temperature data of the constant voltage charging stage of each charge and discharge cycle is obtained. The average value is calculated as follows:

[0073]

[0074] Among them, T avg Indicates the average charging temperature, T1 to T n Represents the temperature of each sampling point, and n represents the number of sampling points.

[0075] In a disclosed embodiment of the present invention, temperature data of the constant voltage charging stage of each charge and discharge cycle is obtained at rates of 0.1C, 0.2C, 0.3C, and 0.5C, respectively.

[0076] The method for extracting time features is as follows:

[0077] The length of the current trajectory is extracted as the time feature, and the current data and charging time of the constant voltage charging stage of each charge and discharge cycle are obtained. The length of the curve trajectory of the current changing with time is calculated as follows:

[0078]

[0079] Where L is the length of the current trajectory, f(x) is the expression of the current changing with time, and a and b represent the start and end time of the constant voltage charging stage, respectively.

[0080] In a disclosed embodiment of the present invention, the constant voltage charging phase current data and charging time of each charge and discharge cycle are obtained at 0.1C, 0.2C, 0.3C, and 0.5C rates, respectively.

[0081] The electrical feature extraction method is as follows:

[0082] The incremental energy is extracted as the electrical feature to obtain the energy data of the constant voltage charging stage of each charge and discharge cycle, and the difference between the end energy and the start energy is calculated using the following formula:

[0083] ΔE=E end -E start

[0084] Where ΔE represents the incremental energy, E end and E start are the ending energy and the starting energy respectively.

[0085] In a disclosed embodiment of the present invention, energy data of the constant voltage charging stage of each charge and discharge cycle is obtained at 0.1C, 0.2C, 0.3C, and 0.5C rates, respectively.

[0086] The algorithms of the CNN-EFC layer include:

[0087] Convolutional layer: For a one-dimensional input x(u) and a convolution kernel (filter) f(s), where u is the index in time or space and s is the index within the convolution kernel, the one-dimensional convolution operation is expressed as follows:

[0088]

[0089] Among them, y(u) represents the output after the convolution operation, that is, the feature map; each filter in the convolution layer performs a one-dimensional convolution operation to generate the corresponding feature map;

[0090] Activation function: Select Relu as the activation function. The formula of the Relu activation function is as follows:

[0091] h(u)=max(0,y(u))

[0092] Among them, h(u) represents the output after y(u) is activated. If the convolution output y(u) is greater than 0, the output of the activation function is y(u); if it is less than or equal to 0, the output is 0;

[0093] Max pooling layer: The max pooling operation selects the maximum value in the set range as the representative of the range. For an activated feature map h(u) and pooling window size p, the max pooling operation is expressed as:

[0094]

[0095] Among them, m(u) is the output after the pooling operation, and the max operation is applied in the index s ranging from u to u+p-1, indicating that the maximum value is selected within each pooling window.

[0096] The algorithms of the EFC weight distribution layer include:

[0097] To calculate the weight score, the feature vector m(u) of each time step u is converted into a weight score through a fully connected layer. The process is expressed as:

[0098] z(u)=W·m(u)+b

[0099] Among them, z(u) represents the weight score, W and b are the weight and bias of the fully connected layer respectively;

[0100] Use the Softmax function to convert to actual weights: The actual weight for each time step is calculated as follows:

[0101]

[0102] Among them, a(u) represents the actual weight, U is the total number of time steps, and the sum of the weights of all time steps is ensured to be 1;

[0103] By generating a weighted output, the original input m(u) is weighted using the obtained weight a(u) to obtain a weighted result. For each time step, the weighted result m′(u) is calculated as follows:

[0104] m′(u)=a(u)·m(u).

[0105] The algorithm of the BiLSTM layer includes:

[0106] First, use the external state h of the previous time step u-1 and the input x at the current time step u , calculate the forget gate f u , input gate i u , output gate o u and candidate state units The formula is as follows:

[0107] f u =σ(W f ·[h u-1 , x u ]+b f )

[0108] i u =σ(W i ·[h u-1 , x u ]+b i )

[0109] o u =σ(W o ·[h u -1,x u ]+b o )

[0110]

[0111] Among them, W f , W i , W o , W C and b f , b i , b o , b C Respectively represent the weights and biases of the forget gate, input gate, output gate, and candidate state unit, σ is the Sigmoid activation function, which is used to compress the result between 0 and 1, and tanh is the hyperbolic tangent activation function, which compresses the result between -1 and 1;

[0112] Combined with forget gate f u and input gate i u To update memory cell C u , the formula is as follows:

[0113]

[0114] Among them, C u-1 represents the memory unit before updating, Represents a candidate state unit.

[0115] Combined with output gate o u , passing the information of the internal state to the external state h u , the formula is as follows:

[0116] h u =o u tanh(C u )

[0117] For the BiLSTM layer, each time step simultaneously calculates the forward and reverse external states, and then combines these two external states to form the final external state.

[0118] A part of the extracted multiple physical domain feature sequences is used as the test set, the specific contents of which are as follows:

[0119] Take k+1 to i sample values ​​of various physical domain feature sequences Input the trained CNN-EFC-BiLSTM battery health state estimation model to obtain the estimated battery health state sequence

[0120] Among them, T avgk+1 is the k+1th value of the thermal feature sequence, T avgi is the ith value of the thermal feature sequence, L k+1 is the k+1th value of the time feature sequence, L i is the i-th value of the time characteristic sequence, ΔE k+1 is the k+1th value of the electrical characteristic sequence, ΔE i is the i-th value of the electrical characteristic sequence; is the k+1th value of the estimated battery health status sequence, is the i-th value of the estimated battery health status sequence.

[0121] The present invention verifies through the Pearson correlation coefficient method that multiple physical features have a strong correlation with the battery health status. Compared with the more singular features extracted by the traditional feature extraction method, the multiple physical features extracted by the present invention can more comprehensively reflect the subtle internal changes of the battery during the aging process, and thus can establish a more accurate mapping model with the battery health status through deep learning.

[0122] In order to demonstrate the feasibility of a lithium-ion battery health status estimation method based on physical domain features and CNN-EFC-BiLSTM of the present invention, the present invention is described here with the following embodiments, and the training time and error of various models are compared to illustrate the accuracy and rapidity of the present invention:

[0123] The present invention uses battery testing equipment (Xinwei BTS-4000) to conduct battery aging experiments on cylindrical lithium-ion batteries at room temperature. The rated capacity of the battery is 2.5Ah. Charge at a constant current of 1250mA (0.5C) until the battery terminal voltage reaches the upper cut-off voltage, then keep the voltage constant and perform constant voltage charging until the current drops to the cut-off current. After standing for a period of time, perform 0.5C constant current discharge until the battery terminal voltage drops to the lower cut-off voltage. The battery health status aging parameters are shown in Table 1:

[0124] Table 1 Battery aging parameters

[0125]

[0126] Using the above 0.5C rate charging data, after extracting multiple physical domain features, the corresponding SOH values ​​are combined to form a complete data sequence. The first 50% of the data sequence is used to train the model, and the following 50% of the data sequence is used to test the performance of the model. Figure 5 Shown are the estimation results of the CNN-EFC-BiLSTM model.

[0127] In order to verify the superiority of the CNN-EFC-BiLSTM model constructed in this application, the CNN-BiLSTM-AM model, as well as the BiLSTM, BiGRU and SVR models were selected for comparative experiments; the above-mentioned 0.5C rate charging data was used to extract a variety of physical domain features, and the corresponding SOH values ​​were combined to form a complete data sequence; the first 50% of the data sequence was used to train the model, and the subsequent 50% of the data sequence was used to test the performance of the model; Figure 6 As shown in Figure 2, the SOH estimation results of different models are compared; Table 2 is the comparison of the error indicators of each model; Table 3 is the comparison of the training time of each model; In the comparative experiment, the MAE values ​​of CNN-BiLSTM-AM, BiLSTM, BiGRU and SVR are 0.5822%, 0.6049%, 1.1449% and 0.8812% respectively, while that of CNN-EFC-BiLSTM is only 0.3219%; The RMSE values ​​of CNN-BiLSTM-AM, BiLSTM, BiGRU and SVR are 0.7016%, 0.7611%, 1.5008% and 1.0100% respectively, while that of CNN-EFC-BiLSTM is only 0.4236%; and the R 2 It reached 98.24%; the CNN-EFC-BiLSTM model had the best performance in all three indicators; in addition, the training time of the CNN-EFC-BiLSTM model was 9.24s, which was about 30% lower than that of the BiLSTM model.

[0128] Table 2 Comparison of error indicators of each model

[0129]

[0130] Table 3 Comparison of training time of each model

[0131]

[0132] The test diagram of this embodiment is as follows Figure 2-6 shown.

[0133] The present invention establishes a mapping relationship between features and battery health status through a deep learning model CNN-EFC-BiLSTM, which has both rapidity and accuracy; an EFC weight distribution layer is added after CNN-EFC is a CNN model, and the ability of the CNN model to combine with the original feature sequence is enhanced by allocating weights to the output of the CNN model; a BiLSTM layer with the ability to capture the long-term dependency of the sequence is arranged after the CNN-EFC layer, and the forward and reverse directions of the sequence are calculated respectively; CNN-EFC is used to capture key change trends from the original feature sequence, and the original feature sequence is combined into an abstract feature sequence, the BiLSTM layer is used to calculate the long-term dependency of the sequence, and the output of the CNN-EFC layer is used as the input of the BiLSTM layer, which effectively reduces the calculation burden of the BiLSTM.

[0134] Based on the fact that the CNN model has the characteristics of being able to capture local dependencies and key change trends in time series, the present invention effectively alleviates the instability of the CNN model caused by hyperparameter settings by adding an EFC weight distribution layer after the CNN model, enhances the ability to combine the original feature sequence and improves the generalization ability of the algorithm.

[0135] In another embodiment, the present invention proposes a lithium-ion battery health status estimation system based on physical domain features and CNN-EFC-BiLSTM, comprising:

[0136] The battery health status data acquisition module is used to measure the constant voltage charging health status data of each charge and discharge cycle of the lithium-ion battery to form a data sequence;

[0137] A physical domain feature extraction module is used to extract physical domain features from a data sequence as health indicators to form a physical domain feature sequence;

[0138] The health state estimation model building module is used to establish a health state estimation model based on CNN-EFC-BiLSTM based on the health state data and health indicators of lithium-ion batteries; the health state estimation model based on CNN-EFC-BiLSTM is as follows: an EFC weight distribution layer for capturing key change trends is added after the CNN model to form a CNN-EFC layer, and a BiLSTM layer for calculating the long-term dependency of the sequence is set after the CNN-EFC layer;

[0139] The battery health status determination module is used to extract physical domain features from the constant voltage charging stage data of the lithium-ion battery, and then calculate the health status results of the lithium-ion battery through the CNN-EFC-BiLSTM model.

[0140] In another embodiment, the present invention proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned lithium-ion battery health status estimation method based on physical domain features and CNN-EFC-BiLSTM is implemented.

[0141] In another embodiment, the present invention provides a computer-readable storage medium storing a computer program, which enables a computer to execute the above-mentioned lithium-ion battery health state estimation method based on physical domain features and CNN-EFC-BiLSTM.

[0142] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0143] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in any form. Any technician familiar with the profession, without departing from the scope of the technical solution of the present invention and based on the technical essence of the present invention, may make any simple modifications, equivalent substitutions and improvements to the above embodiments, which are still within the protection scope of the technical solution of the present invention.

Claims

1. A lithium-ion battery health state estimation method based on physical domain features and CNN-EFC-BiLSTM, characterized in that: The following steps are involved: Measure the constant voltage charging health status data of each charge and discharge cycle of the lithium-ion battery to form a data sequence; Extracting physical domain features from the data sequence as health indicators to form a physical domain feature sequence; Based on the health status data and health indicators of lithium-ion batteries, a health status estimation model based on CNN-EFC-BiLSTM is established; the health status estimation model based on CNN-EFC-BiLSTM is as follows: an EFC weight distribution layer for capturing key change trends is added after the CNN model to form a CNN-EFC layer, and a BiLSTM layer for calculating long-term dependencies of sequences is set after the CNN-EFC layer; After extracting the physical domain features from the data of the constant voltage charging stage of the lithium-ion battery, the health status of the lithium-ion battery is calculated through the CNN-EFC-BiLSTM model.

2. The method for estimating the health status of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM according to claim 1, characterized in that: The physical domain characteristics include: thermal characteristics, time characteristics and electrical characteristics; wherein the thermal characteristics refer to the average charging temperature of the lithium-ion battery, the time characteristics refer to the current trajectory length, and the electrical characteristics refer to the incremental energy.

3. The method for estimating the health status of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM according to claim 2, characterized in that: The thermal feature extraction method is as follows: The average charging temperature of lithium-ion batteries is extracted as a thermal feature, and the temperature data of the constant voltage charging stage of each charge and discharge cycle is obtained. The average value is calculated as follows: Among them, T avg Indicates the average charging temperature, T1 to T n Represents the temperature of each sampling point, and n represents the number of sampling points.

4. The method for estimating the health status of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM according to claim 2, characterized in that: The method for extracting the time feature is as follows: The length of the current trajectory is extracted as the time feature, and the current data and charging time of the constant voltage charging stage of each charge and discharge cycle are obtained. The length of the curve trajectory of the current changing with time is calculated as follows: Where L is the length of the current trajectory, f(x) is the expression of the current changing with time, and a and b represent the start and end time of the constant voltage charging stage, respectively.

5. The method for estimating the health status of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM according to claim 2, characterized in that: The method for extracting the electrical characteristics is as follows: The incremental energy is extracted as the electrical feature to obtain the energy data of the constant voltage charging stage of each charge and discharge cycle, and the difference between the end energy and the start energy is calculated using the following formula: Δ Where ΔE represents the incremental energy, E end and E start are the ending energy and the starting energy respectively.

6. The method for estimating the health status of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM according to claim 1, characterized in that: The algorithm of the CNN-EFC layer includes: Convolutional layer: For a one-dimensional input x(u) and a convolution kernel (filter) f(s), where u is the index in time or space and s is the index within the convolution kernel, the one-dimensional convolution operation is expressed as follows: Wherein, y(u) represents the output after the convolution operation, that is, the feature map; each filter in the convolution layer performs the one-dimensional convolution operation to generate a corresponding feature map; Activation function: Select Relu as the activation function. The formula of the Relu activation function is as follows: h(u)=max(0,(u)) Among them, h(u) represents the output after y(u) is activated. If the convolution output y(u) is greater than 0, the output of the activation function is y(u); if it is less than or equal to 0, the output is 0; Max pooling layer: The max pooling operation selects the maximum value in the set range as the representative of the range. For an activated feature map h(u) and pooling window size p, the max pooling operation is expressed as: Among them, m(u) is the output after the pooling operation, and the max operation is applied in the index s ranging from u to u+p-1, indicating that the maximum value is selected within each pooling window.

7. The method for estimating the health status of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM according to claim 6, characterized in that: The algorithm of the EFC weight distribution layer includes: To calculate the weight score, the feature vector m(u) of each time step u is converted into a weight score through a fully connected layer. The process is expressed as: z(u)=W·m(u)+b Among them, z(u) represents the weight score, W and b are the weight and bias of the fully connected layer respectively; Use the Softmax function to convert to actual weights: The actual weight for each time step is calculated as follows: Among them, a(u) represents the actual weight, U is the total number of time steps, and the sum of the weights of all time steps is ensured to be 1; By generating a weighted output, the original input m(u) is weighted using the obtained weight a(u) to obtain a weighted result. For each time step, the weighted result m′(u) is calculated as follows: m′(u)=a(u)·m(u).

8. A lithium-ion battery health status estimation system based on physical domain features and CNN-EFC-BiLSTM, characterized in that: include: The battery health status data acquisition module is used to measure the constant voltage charging health status data of each charge and discharge cycle of the lithium-ion battery to form a data sequence; A physical domain feature extraction module is used to extract physical domain features from a data sequence as health indicators to form a physical domain feature sequence; A health state estimation model building module is used to establish a health state estimation model based on CNN-EFC-BiLSTM based on the health state data and health indicators of the lithium-ion battery; the health state estimation model based on CNN-EFC-BiLSTM is: an EFC weight distribution layer for capturing key change trends is added after the CNN model to form a CNN-EFC layer, and a BiLSTM layer for calculating the long-term dependency of the sequence is set after the CNN-EFC layer; The battery health status determination module is used to extract physical domain features from the constant voltage charging stage data of the lithium-ion battery, and then calculate the health status results of the lithium-ion battery through the CNN-EFC-BiLSTM model.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for estimating the health state of a lithium-ion battery based on physical domain features and CNN-EFC-BiLSTM as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the lithium-ion battery health state estimation method based on physical domain features and CNN-EFC-BiLSTM as described in any one of claims 1 to 7.

Images