A battery health state estimation method and system based on time-frequency characteristics

Abstract

The present invention provides a method and system for estimating the health status of a battery based on time-frequency characteristics in the field of battery detection technology. The method includes the following steps: step S10, obtaining a large amount of battery charging and discharging data; step S20, extracting the time-frequency characteristics of each of the charging and discharging data, and then converting each of the charging and discharging data into a time-frequency spectrum diagram; step S30, creating a battery health status estimation model, and using each of the time-frequency spectrum diagrams to train the battery health status estimation model; step S40, using the trained battery health status estimation model to estimate the battery health status. The advantages of the present invention are: greatly improving the accuracy and generalization ability of battery health status estimation.

Description

Technical Field

[0001] The present invention relates to the field of battery detection technology, and in particular to a battery health state estimation method and system based on time-frequency characteristics. Background Art

[0002] Batteries have been widely used in electronic devices, new energy vehicles, energy storage and other fields. Since the mechanism of internal battery aging reaction is complex and is greatly affected by the external environment and operating conditions, accurate estimation of the battery's state of health (SOH) is a difficult problem in battery management.

[0003] The health status of a battery is generally characterized by capacity decay. There are two methods for estimating the health status:

[0004] 1. Model-based prediction method. This method not only requires a lot of expert knowledge, but also has a very high model complexity. In actual application, it is affected by the actual outdoor temperature and outdoor environment, resulting in insufficient model accuracy and poor generalization ability. In addition, this method only considers time domain features, which further affects the model accuracy. 2. Prediction method based on big data. Traditionally, the features of big data are manually extracted, which cannot guarantee the final accuracy of the model. The prediction results for battery data from different sources will be uneven, and the generalization ability is poor. In addition, this method only considers time domain features, which further affects the model accuracy.

[0005] Therefore, how to provide a battery health state estimation method and system based on time-frequency characteristics to improve the accuracy and generalization ability of battery health state estimation has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for estimating the state of health of a battery based on time-frequency characteristics, so as to improve the accuracy and generalization capability of the estimation of the state of health of the battery.

[0007] In a first aspect, the present invention provides a method for estimating a battery health state based on time-frequency characteristics, comprising the following steps:

[0008] Step S10, obtaining a large amount of battery charge and discharge data;

[0009] Step S20, extracting the time-frequency characteristics of each of the charge-discharge data, and then converting each of the charge-discharge data into a time-frequency spectrum;

[0010] Step S30, creating a battery health state estimation model, and using each of the time-frequency spectrum graphs to train the battery health state estimation model;

[0011] Step S40: Estimating the battery health state using the trained battery health state estimation model.

[0012] Furthermore, in the step S10, the charge and discharge data includes a voltage timing curve, a current timing curve, and a temperature timing curve of the battery when charging and discharging from a preset first SOC to a preset second SOC.

[0013] Furthermore, in step S20, the time-frequency features are extracted through Mel spectrum.

[0014] Furthermore, in step S30, the battery health state estimation model is created based on a residual convolutional neural network.

[0015] Furthermore, in step S30, the loss function of the battery health state estimation model adopts a regression loss function.

[0016] In a second aspect, the present invention provides a battery health status estimation system based on time-frequency characteristics, comprising the following modules:

[0017] A charge and discharge data acquisition module is used to acquire a large amount of battery charge and discharge data;

[0018] A time-frequency feature extraction module, used to extract the time-frequency features of each of the charge-discharge data, and then convert each of the charge-discharge data into a time-frequency spectrum diagram;

[0019] A model training module, used to create a battery health state estimation model, and train the battery health state estimation model using the time-frequency spectrum diagrams;

[0020] The battery health state estimation module is used to estimate the battery health state using the trained battery health state estimation model.

[0021] Furthermore, in the charge and discharge data acquisition module, the charge and discharge data includes a voltage timing curve, a current timing curve, and a temperature timing curve of the battery when charging and discharging from a preset first SOC to a preset second SOC.

[0022] Furthermore, in the time-frequency feature extraction module, the time-frequency feature is extracted through Mel spectrum.

[0023] Furthermore, in the model training module, the battery health state estimation model is created based on a residual convolutional neural network.

[0024] Furthermore, in the model training module, the loss function of the battery health state estimation model adopts a regression loss function.

[0025] The advantages of the present invention are:

[0026] By extracting the time-frequency features of the charging and discharging data and converting them into time-frequency spectrograms, the time-frequency spectrograms are used to train the battery health state estimation model created based on the residual convolutional neural network. It is best to use the trained battery health state estimation model to estimate the battery health state, that is, converting one-dimensional time series signals such as voltage, current, and temperature into time-frequency feature domain spectrograms, combining residual convolutional neural networks for time-frequency feature extraction, which is conducive to obtaining the characterization characteristics of the battery inside. The residual convolutional neural network has a simple structure. Feature extraction is performed through the residual convolutional neural network to replace traditional manual extraction, avoiding errors caused by manual extraction, and ultimately greatly improving the accuracy and generalization ability of battery health state estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0028] Figure 1 It is a flow chart of a battery health status estimation method based on time-frequency characteristics of the present invention.

[0029] Figure 2 It is a structural schematic diagram of a battery health status estimation system based on time-frequency characteristics of the present invention.

[0030] Figure 3 It is a schematic diagram of a time spectrum diagram of the present invention.

[0031] Figure 4 It is a schematic diagram of the structure of the residual convolutional neural network of the present invention. DETAILED DESCRIPTION

[0032] The technical solution in the embodiments of the present application has the following overall idea: converting one-dimensional time series signals such as voltage, current, and temperature into a time-frequency feature domain spectrogram, and combining a residual convolutional neural network to automatically extract time-frequency features instead of traditional manual extraction, which is beneficial to obtaining the characterization features inside the battery. The residual convolutional neural network has a simple structure, which reduces the impact of outdoor temperature and outdoor environment on the model, thereby improving the accuracy and generalization ability of battery health status estimation.

[0033] Please refer to Figures 1 to 4 As shown, a preferred embodiment of a battery health state estimation method based on time-frequency characteristics of the present invention includes the following steps:

[0034] Step S10, obtaining a large amount of battery charge and discharge data;

[0035] Step S20, extracting the time-frequency features of each of the charge and discharge data, and then converting each of the charge and discharge data into a time-frequency spectrum diagram; since the time series signals of the voltage, current, and temperature during battery charge and discharge are non-stationary signals, the time-frequency features are extracted to obtain the characterization features inside the battery;

[0036] Step S30, creating a battery health state estimation model, and using each of the time-frequency spectrum graphs to train the battery health state estimation model;

[0037] Step S40: Estimating the battery health state using the trained battery health state estimation model.

[0038] In step S10, the charge and discharge data includes a voltage timing curve, a current timing curve, and a temperature timing curve of the battery from a preset first SOC to a preset second SOC, for example, charging the battery from a 0 SOC state to a 100 SOC state.

[0039] The battery charging process is as follows: the battery is charged at 24 degrees Celsius in a constant current mode of 1.5A until the battery voltage reaches 4.2V, and then charged in a constant voltage mode until the current drops to 20mA.

[0040] In step S20, the time-frequency feature is extracted through the Mel spectrum, and the formula is as follows:

[0041] mel(f)=2595*log 10 (1+f / 700);

[0042] Among them, mel(f) represents the Mel frequency scale; f represents the frequency of the charge and discharge data; due to the log relationship, when the frequency is small, mel(f) changes faster with Hz; when the frequency is large, mel(f) rises very slowly and the slope of the curve is very small; the benefits of using time-frequency feature extraction are, first, better mining of intrinsic information in the time-frequency domain, and second, converting it into an image-like pattern, which can be used for subsequent model building and training using a deep convolutional neural network.

[0043] In step S30, the battery health state estimation model is created based on a residual convolutional neural network. The residual convolutional neural network introduces an identity mapping, which reduces possible gradient diffusion and gradient explosion during model training, making it easier to converge.

[0044] In step S30, the loss function of the battery health state estimation model adopts a regression loss function (Log-Cosh). Log-Cosh is another loss function applied to regression tasks. It is smoother than L2 loss and is the logarithm of the hyperbolic cosine of the prediction error. A warm-up learning rate strategy is adopted and an Adam optimizer is used for model training. The formula is as follows:

[0045]

[0046] Among them, L(y,y p ) represents the loss value of the regression loss function; cosh() represents the hyperbolic function; represents the predicted SOH value of the i-th sample; y i Represents the true SOH value of the i-th sample.

[0047] A preferred embodiment of a battery health status estimation system based on time-frequency characteristics of the present invention includes the following modules:

[0048] A charge and discharge data acquisition module is used to acquire a large amount of battery charge and discharge data;

[0049] A time-frequency feature extraction module is used to extract the time-frequency features of each of the charging and discharging data, and then convert each of the charging and discharging data into a time-frequency spectrum diagram; since the time series signals of the voltage, current, and temperature during battery charging and discharging are non-stationary signals, the time-frequency features are extracted to obtain the characterization features inside the battery;

[0050] A model training module, used to create a battery health state estimation model, and train the battery health state estimation model using the time-frequency spectrum diagrams;

[0051] The battery health state estimation module is used to estimate the battery health state using the trained battery health state estimation model.

[0052] In the charging and discharging data acquisition module, the charging and discharging data includes a voltage timing curve, a current timing curve, and a temperature timing curve of the battery charging and discharging from a preset first SOC to a preset second SOC, for example, charging the battery from a state of 0 SOC to a state of 100 SOC.

[0053] The battery charging process is as follows: the battery is charged at 24 degrees Celsius in a constant current mode of 1.5A until the battery voltage reaches 4.2V, and then charged in a constant voltage mode until the current drops to 20mA.

[0054] In the time-frequency feature extraction module, the time-frequency feature is extracted through the Mel spectrum, and the formula is as follows:

[0055] mel(f)=2595*log 10 (1+f / 700);

[0056] Among them, mel(f) represents the Mel frequency scale; f represents the frequency of the charge and discharge data; due to the log relationship, when the frequency is small, mel(f) changes faster with Hz; when the frequency is large, mel(f) rises very slowly and the slope of the curve is very small; the benefits of using time-frequency feature extraction are, first, better mining of intrinsic information in the time-frequency domain, and second, converting it into an image-like pattern, which can be used for subsequent model building and training using a deep convolutional neural network.

[0057] In the model training module, the battery health state estimation model is created based on a residual convolutional neural network. The residual convolutional neural network introduces an identity mapping, which reduces possible gradient diffusion and gradient explosion during model training, making it easier to converge.

[0058] In the model training module, the loss function of the battery health state estimation model adopts a regression loss function (Log-Cosh). Log-Cosh is another loss function applied to regression tasks. It is smoother than L2 loss and is the logarithm of the hyperbolic cosine of the prediction error. A warm-up learning rate strategy is adopted and the Adam optimizer is used for model training. The formula is as follows:

[0059]

[0060] Among them, L(y,y p ) represents the loss value of the regression loss function; cosh() represents the hyperbolic function; represents the predicted SOH value of the i-th sample; y i Represents the true SOH value of the i-th sample.

[0061] In summary, the advantages of the present invention are:

[0062] By extracting the time-frequency features of the charging and discharging data and converting them into time-frequency spectrograms, the time-frequency spectrograms are used to train the battery health state estimation model created based on the residual convolutional neural network. It is best to use the trained battery health state estimation model to estimate the battery health state, that is, converting one-dimensional time series signals such as voltage, current, and temperature into time-frequency feature domain spectrograms, combining residual convolutional neural networks for time-frequency feature extraction, which is conducive to obtaining the characterization characteristics of the battery inside. The residual convolutional neural network has a simple structure. Feature extraction is performed through the residual convolutional neural network to replace traditional manual extraction, avoiding errors caused by manual extraction, and ultimately greatly improving the accuracy and generalization ability of battery health state estimation.

[0063] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A battery health status estimation method based on time-frequency characteristics, Features: The steps include: Step S10, acquiring a large amount of battery charge and discharge data; the charge and discharge data includes a voltage timing curve, a current timing curve, and a temperature timing curve of the battery charging and discharging from a preset first SOC to a preset second SOC; Step S20, extracting the time-frequency characteristics of each of the charge-discharge data, and then converting each of the charge-discharge data into a time-frequency spectrum; Step S30, creating a battery health state estimation model, and using each of the time-frequency spectrum graphs to train the battery health state estimation model; the loss function of the battery health state estimation model adopts a regression loss function; Step S40: Estimating the battery health state using the trained battery health state estimation model.

2. A method for estimating a battery health state based on time-frequency characteristics as claimed in claim 1, Features: In the step S20, the time-frequency features are extracted through Mel spectrum.

3. The method for estimating the battery health status based on time-frequency characteristics according to claim 1, Features: In step S30, the battery health state estimation model is created based on a residual convolutional neural network.

4. A battery health status estimation system based on time-frequency characteristics, Features: Includes the following modules: A charge and discharge data acquisition module, used to acquire a large amount of charge and discharge data of the battery; the charge and discharge data includes a voltage timing curve, a current timing curve, and a temperature timing curve of the battery charging and discharging from a preset first SOC to a preset second SOC; A time-frequency feature extraction module, used to extract the time-frequency features of each of the charge-discharge data, and then convert each of the charge-discharge data into a time-frequency spectrum diagram; A model training module, used to create a battery health state estimation model, and train the battery health state estimation model using each of the time-frequency spectrum graphs; the loss function of the battery health state estimation model adopts a regression loss function; The battery health state estimation module is used to estimate the battery health state using the trained battery health state estimation model.

5. A battery health status estimation system based on time-frequency characteristics as claimed in claim 4, Features: In the time-frequency feature extraction module, the time-frequency features are extracted through Mel spectrum.

6. A battery health status estimation system based on time-frequency characteristics as claimed in claim 4, Features: In the model training module, the battery health state estimation model is created based on a residual convolutional neural network.

Images