Lithium battery test and analysis method and system based on deep learning and feature analysis

Abstract

This application relates to the technical field of battery performance testing and analysis, and discloses a lithium battery testing and analysis method based on deep learning and feature analysis, including: S1. Establish a dynamic environment model to obtain the dynamic distribution of temperature and humidity in the battery usage environment, where the dynamic distribution of temperature and humidity is based on the combined effects of diffusion, convection, and external disturbances; S2. Extract dynamic environment features, and use the main amplitudes and frequencies of temperature and humidity as environmental feature vectors; S3. Based on the lithium ion diffusion model, establish a coupling relationship between the lithium ion concentration distribution and the environmental features, where the lithium ion concentration distribution is affected by the changes in temperature and humidity in the dynamic environment. By adopting the technical solution based on dynamic environment modeling and feature extraction, the dynamic distribution of temperature and humidity is described by partial differential equations, and environmental feature vectors are extracted, achieving the technical effect of accurately simulating and quantifying the impact of complex environmental changes on battery performance.

Description

Technical Field

[0001] The present invention relates to the technical field of battery performance testing and analysis, and particularly to a lithium battery testing and analysis method and system based on deep learning and feature analysis. Background Art

[0002] As a high-energy-density and high-safety energy storage device, polymer lithium-ion batteries have been widely used in fields such as electric vehicles, portable devices, and energy storage systems. However, the performance of batteries is significantly affected by environmental factors. Especially in actual use, the dynamic changes in temperature and humidity may lead to significant fluctuations in the internal reaction rate, lithium-ion diffusion process, and overall discharge performance of the battery. Therefore, how to accurately evaluate the battery performance under dynamic environments has become an important issue in current research.

[0003] Generally, the existing technology for testing the performance of lithium-ion batteries is usually carried out under constant temperature and humidity conditions. This technical solution can indeed obtain stable test results in the laboratory environment, but it ignores the complexity of environmental conditions in real working conditions. For example, during the actual driving of an electric vehicle, the environmental temperature may fluctuate significantly with seasons and regions, and the humidity may also be affected by various external conditions. There is a lack of in-depth research on modeling and simulating dynamic environments in the existing methods, resulting in poor applicability of the test results in actual scenarios.

[0004] In the existing technology, some testing methods attempt to introduce changes in environmental factors into the evaluation of battery performance. However, these methods often rely on simple point measurements and usually only test a few fixed discrete values of temperature and humidity. Although this method can initially reflect the impact of environmental changes on battery performance, since it fails to simulate continuous dynamic environmental changes, the test results are difficult to accurately depict the performance change process under real working conditions. This defect greatly reduces the representativeness of the test results and the refinement degree of analysis.

[0005] In terms of the physical modeling of lithium-ion batteries, most existing research focuses on static diffusion and reaction processes, and the diffusion coefficient and reaction rate are usually assumed to be fixed values. Although this assumption simplifies the modeling process, in a dynamic environment, this method cannot reflect the regulatory effects of temperature and humidity on the internal diffusion process and reaction kinetics of the battery. For example, when the environmental temperature increases, the diffusion rate of lithium ions will change significantly, and the static model cannot capture this change process. The lack of a coupling mechanism between the dynamic environment and the internal reaction process makes the existing models less adaptable to complex working conditions. Summary of the Invention

[0006] In view of the deficiencies of the prior art, the present invention provides a lithium battery test and analysis method and system based on deep learning and feature analysis, which solves the technical problem in the prior art that it is impossible to accurately evaluate the influence of temperature and humidity changes on the discharge performance of polymer lithium-ion batteries in a dynamic environment.

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions: A lithium battery test and analysis method based on deep learning and feature analysis, including:

[0008] S1. Establish a dynamic environment model to obtain the dynamic distribution of temperature and humidity in the battery usage environment, and the dynamic distribution of temperature and humidity is based on the combined influence of diffusion, convection, and external disturbances;

[0009] S2. Extract dynamic environment features, and use the main amplitudes and frequencies of temperature and humidity as environmental feature vectors;

[0010] S3. Based on the lithium ion diffusion model, establish a coupling relationship between the lithium ion concentration distribution and environmental features, and the lithium ion concentration distribution is affected by temperature and humidity changes in the dynamic environment;

[0011] S4. Collect the discharge signals of the polymer lithium-ion battery, and extract voltage, current, and power signals as discharge feature vectors;

[0012] S5. Perform interactive analysis on the environmental feature vectors and discharge feature vectors to generate interactive feature vectors;

[0013] S6. Based on the interactive feature vectors, judge whether the discharge performance of the polymer lithium-ion battery is normal.

[0014] Preferably, in step S1, the dynamic environment model describes the dynamic changes of temperature and humidity through the following partial differential equations:

[0015] The change in temperature is determined by the diffusion term, convection term, and external disturbance term;

[0016] The change in humidity is determined by the diffusion term, convection term, and external disturbance term.

[0017] Preferably, in step S2, extracting the dynamic environment features includes: performing time-frequency domain analysis on the dynamic distribution of temperature and humidity to obtain their main amplitudes and frequencies, and forming dynamic environment feature vectors.

[0018] Preferably, in step S3, the lithium ion diffusion model describes that the diffusion coefficient of lithium ions is affected by temperature and humidity, and adjusts the diffusion rate and reaction rate through the dynamic environment feature vectors.

[0019] Preferably, in step S4, the collected discharge signals include voltage, discharge current, and power values in the time series, and the discharge signals are preprocessed to obtain discharge feature vectors.

[0020] Preferably, in step S5, a feature interaction analysis method is adopted to fuse the environmental feature vector and the discharge feature vector through an attention mechanism to generate an interaction feature vector.

[0021] Preferably, the attention mechanism is implemented through the following steps:

[0022] Map the environmental feature vector and the discharge feature vector into query vector, key vector and value vector respectively;

[0023] Calculate the similarity score between the query vector and the key vector;

[0024] Use the similarity score to perform weighted average on the value vector to generate an interaction feature vector.

[0025] Preferably, in step S6, a classification model based on the interaction feature vector is used to judge whether the discharge performance of the polymer lithium-ion battery is normal, and the classification model is a deep neural network classifier.

[0026] Preferably, the classification model determines the battery performance through the probability distribution of the interaction feature vector and outputs the classification result of normal state or abnormal state.

[0027] A lithium battery test and analysis system based on deep learning and feature analysis includes:

[0028] A signal acquisition module for acquiring the discharge signal of the polymer lithium-ion battery and the temperature and humidity data of the battery working environment;

[0029] A dynamic environment modeling module for establishing a dynamic distribution model of temperature and humidity;

[0030] A feature extraction module for extracting the dynamic features of temperature and humidity and the features of the discharge signal;

[0031] A feature interaction analysis module for performing interaction analysis on the dynamic environment features and the discharge features to generate an interaction feature vector;

[0032] A performance evaluation module for judging whether the discharge performance of the polymer lithium-ion battery is normal according to the interaction feature vector and outputting a performance evaluation result.

[0033] The present invention provides a lithium battery test and analysis method and system based on deep learning and feature analysis. It has the following beneficial effects:

[0034] 1. The present invention adopts a technical solution based on dynamic environment modeling and feature extraction. By using partial differential equations to describe the dynamic distribution of temperature and humidity and extracting environmental feature vectors, it achieves the technical effect of accurately simulating and quantifying the impact of complex environmental changes on battery performance. Compared with the prior art that simply relies on constant temperature and humidity testing, the present invention solves the problem that traditional methods cannot adapt to the actual dynamic environment, and significantly improves the authenticity and applicability of test results.

[0035] 2. The present invention couples the lithium-ion diffusion model with dynamic environmental features and adopts an expression method of environment-sensitive diffusion coefficients and reaction rates to achieve an accurate description of the internal physical processes of the battery under the influence of environmental variables. This technical solution can dynamically reflect the regulatory effect of temperature and humidity changes on the lithium-ion concentration distribution. Compared with the static diffusion models in the prior art that ignore environmental factors, it effectively solves the problem of insufficient analysis accuracy of the internal physical processes of the battery under complex working conditions.

[0036] 3. The present invention adopts a feature interaction analysis method based on the attention mechanism, combines dynamic environmental features with battery discharge features to generate interaction feature vectors, thereby fully exploring the non-linear relationship between environmental factors and battery performance. Compared with the prior art that relies on simple linear analysis, the present invention overcomes the defect that the complex coupling between environmental features and battery performance is difficult to characterize, and provides a more reliable and comprehensive data basis for performance classification.

[0037] 4. The present invention constructs a deep learning classification model to analyze and judge the interaction feature vectors and outputs the classification results of the battery discharge performance, thereby realizing a fast and accurate assessment of the battery performance state under the influence of dynamic environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the 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 of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] Embodiment:

[0041] Please refer to the attached Figure 1 , the embodiment of the present invention provides a lithium battery test and analysis method based on deep learning and feature analysis, including:

[0042] S1. Establish a dynamic environment model to obtain the dynamic distribution of temperature and humidity in the battery usage environment, where the dynamic distribution of temperature and humidity is based on the combined effects of diffusion, convection, and external disturbances;

[0043] During the actual use of lithium-ion polymer batteries, their working environment is usually affected by the combined influence of dynamic temperature and humidity. Such changes in the dynamic environment may have a significant impact on the performance and lifespan of the batteries. Therefore, in order to more comprehensively evaluate the battery performance, it is first necessary to establish a dynamic environment model to simulate the spatio-temporal distribution law of temperature and humidity. As mentioned in the foregoing technical content, the dynamic distribution of temperature and humidity is mainly affected by the combined effects of diffusion, convection, and external disturbances. In this step, by constructing a mathematical model based on partial differential equations, the change characteristics of temperature and humidity are described in detail, providing the necessary basic data and analysis support for the subsequent steps.

[0044] In this embodiment, the dynamic environment model is established based on two-dimensional partial differential equations to describe the distribution and change law of temperature and humidity in time and space.

[0045] Specifically, the dynamic change of temperature can be expressed as the following partial differential equation:

[0046] ;

[0047] The dynamic change of humidity can be expressed as:

[0048] ;

[0049] Where: represents the temperature (unit: at the coordinate at a certain moment K ).

[0050] represents the humidity (unit: at the coordinate at a certain moment % ).

[0051] is the temperature diffusion coefficient (unit: / s), representing the diffusion rate of temperature in space.

[0052] is the humidity diffusion coefficient (unit: / s), representing the diffusion rate of humidity in space.

[0053] is the temperature convection coefficient (unit: m / s), representing the influence of convection on temperature.

[0054] is the humidity convection coefficient (unit: m / s), representing the influence of humidity by convection.

[0055] is the temperature change rate caused by an external heat source (unit: K / s), mainly caused by external disturbances, such as rapid fluctuations in the surrounding environmental temperature.

[0056] is the humidity change rate caused by an external humidity source (unit: % / s), such as periodic fluctuations in air humidity.

[0057] In the above partial differential equation, the diffusion terms and describe the diffusion of temperature and humidity in space, and are usually used to characterize the diffusion of heat and humidity in a uniform environment. The convection terms and are used to describe the transfer process of temperature and humidity in a specific direction in the presence of wind speed, convection, etc. in the environment. The external disturbance terms and can be set according to specific environmental conditions, such as the influence of rapid fluctuations in outdoor temperature and humidity on the local environment of the battery.

[0058] As an option, the dynamic environment model in this embodiment can be implemented by numerical solution techniques, such as the finite difference method or the finite element method, to obtain the distribution of temperature and humidity in space and time.

[0059] Specifically, in some embodiments, the two-dimensional environmental field can be divided into a number of finite grid nodes. By discretizing the partial differential equation, the following numerical update formula is obtained:

[0060] ;

[0061] ;

[0062] Where: and respectively represent the temperature and humidity at the node ( ) at the -th time step.

[0063] is the time step, representing the time interval for each calculation (unit: s)

[0064] and are the second-order spatial partial derivatives of temperature and humidity respectively, calculated using a discrete format.

[0065] In some specific implementation manners, the accuracy of the environment simulation can also be controlled by setting appropriate initial conditions and boundary conditions. For example, the initial conditions can be set as the environmental temperature and humidity distribution when the battery starts to operate:

[0066] ;

[0067] The boundary conditions can be set as adiabatic boundaries, that is, there is no heat or humidity outflow at the boundaries:

[0068] ;

[0069] where represents the boundary normal vector.

[0070] Specifically, the external disturbance terms and can be fitted through the actually measured data or set according to specific application scenarios. For example:

[0071] During the operation of an electric vehicle, the external disturbance may be caused by the engine heat or the temperature difference of the external environment;

[0072] In an energy storage system, the humidity disturbance may be caused by seasonal climate changes.

[0073] In some embodiments, this kind of disturbance can be simulated by introducing a periodic function. For example:

[0074] ;

[0075] where , are the amplitudes of the external disturbance respectively, , is the frequency of the disturbance.

[0076] In this embodiment, by constructing the above dynamic environment model, an accurate distribution of the temperature and humidity around the battery changing with time and space is obtained, providing a reliable basis for the subsequent extraction of dynamic environment characteristics. This model can flexibly adapt to different usage scenarios and has high applicability and generality.

[0077] S2. Extract dynamic environment characteristics, and use the main amplitudes and frequencies of temperature and humidity as the environmental characteristic vectors;

[0078] After establishing the dynamic environment model in step S1 and obtaining the dynamic distribution of temperature and humidity, to further quantify the impact of the dynamic environment on the discharge performance of polymer lithium-ion batteries, it is necessary to extract dynamic environment characteristics. The extraction of dynamic environment characteristics is based on the main change characteristics of temperature and humidity, and through time-frequency domain analysis methods, the main amplitudes and frequencies of temperature and humidity changing with time are obtained. The extracted feature vectors will be used as important input data for feature interaction analysis in subsequent steps.

[0079] Generally, the temperature and humidity signals in the dynamic environment simultaneously contain stable components (such as long-term change trends) and fluctuating components (such as periodic changes). To more effectively characterize the features of these signals, in this step, the Fourier transform is used to transform the time series signal into the frequency domain and extract the main frequency and amplitude information of the signal. In this way, not only can the interference of noise be eliminated, but also the data volume can be greatly compressed, improving the efficiency of subsequent analysis.

[0080] In this embodiment, the dynamic environment feature vectors are extracted through the time-frequency domain analysis of the dynamic distribution of temperature and humidity, and the specific implementation method is as follows:

[0081] Specifically, the temperature signal and the humidity signal The dynamic changes of are first transformed from the time domain to the frequency domain representation through the Fourier transform. The mathematical expression of the Fourier transform is as follows:

[0082] ;

[0083] Where:

[0084] represents the amplitude of the signal in the frequency domain;

[0085] is the frequency (unit: Hz);

[0086] is the time (unit: s).

[0087] In the process of the Fourier transform of the temperature and humidity signals, in actual operation, the temperature and humidity time series will be discretized into sample data of finite length, and the fast Fourier transform (FFT) algorithm is used to improve the calculation efficiency.

[0088] In some embodiments, the Fourier transform of the temperature and humidity signals can further decompose its main frequency , and the main amplitude , . Where:

[0089] , respectively represent the frequencies of the main periodic components of the temperature signal and the humidity signal;

[0090] and respectively represent the amplitudes corresponding to the main frequency components, reflecting the intensity of the temperature and humidity changes.

[0091] For example, in a possible implementation, the main frequency of the temperature signal in the low-frequency band can be used to describe the periodic influence of the day-night temperature difference on the battery environment, and the high-frequency amplitude of the humidity signal may reflect the potential interference of the rapid fluctuations of the local air humidity on the battery performance.

[0092] Generally, in the spectrogram of the signal amplitude after Fourier transform, there are multiple peak points, and each peak point corresponds to a main frequency component of the temperature and humidity signals. To further simplify the feature extraction process, only the main frequency with the largest amplitude and its corresponding amplitude value can be retained, and other minor components can be ignored.

[0093] In a possible implementation, the main frequency and amplitude value can be extracted by the following formula:

[0094] ;

[0095] ;

[0096] where:

[0097] and are respectively the absolute values of the amplitudes of the temperature signal and the humidity signal in the frequency domain;

[0098] represents obtaining the frequency corresponding to the point with the largest amplitude.

[0099] Through the above extraction process, a dynamic environment feature vector can be obtained:

[0100] ;

[0101] The feature vector quantifies the main amplitude and frequency information of the temperature and humidity in the dynamic environment, facilitating the interactive analysis with the discharge characteristics of the battery in subsequent steps.

[0102] As an option, during the Fourier transform process, a filtering algorithm can also be combined to further improve the accuracy of feature extraction. For example, in some embodiments, a band-pass filter is used to limit the main frequency range of the temperature and humidity signals within the actually interested interval, thereby removing high-frequency noise or low-frequency interference. The transfer function of the band-pass filter can be expressed as:

[0103] ;

[0104] wherein and are the lower and upper limit frequencies of the filter respectively. For the signal after filtering, performing Fourier transform can significantly reduce interference and improve the extraction accuracy of the main frequency and amplitude.

[0105] In this embodiment, the feature vector can also be extended according to the actual application scenario.

[0106] Specifically, in some embodiments, in addition to the main frequency and main amplitude, statistical features of the temperature and humidity signals, such as mean value, standard deviation, etc., can also be extracted. For example:

[0107] The mean value of temperature and the mean value of humidity respectively represent the long-term trend;

[0108] The standard deviation σT of temperature and the standard deviation σH of humidity represent the volatility of the signal.

[0109] The addition of the above extended features can make the dynamic environment feature vector change from ;

[0110] ;

[0111] Through the dynamic environment feature extraction in this step, the complex temperature and humidity signals can be converted into a quantified feature vector form. These feature vectors can not only effectively reflect the intensity and periodicity of environmental changes, but also provide sufficient basic data support for the interactive analysis of the environment and battery discharge performance in the subsequent steps.

[0112] S3. Based on the lithium-ion diffusion model, establish the coupling relationship between the lithium-ion concentration distribution and the environmental characteristics, and the lithium-ion concentration distribution is affected by the temperature and humidity changes in the dynamic environment;

[0113] After completing the dynamic environment feature extraction in step S2, the main feature vectors of the environmental temperature and humidity, such as amplitude and frequency, have been obtained. The changes in the dynamic environment features directly affect the diffusion rate of lithium ions and the electrode reaction rate inside the lithium polymer battery. Therefore, it is necessary to couple the dynamic environment features with the lithium-ion concentration distribution through the lithium-ion diffusion model to form an accurate description of the internal physical process of the battery. The goal of this step is to establish the coupling relationship between the dynamic environment and the internal electrochemical behavior of the battery, and provide theoretical support and data basis for the subsequent interactive analysis of the discharge signal and the dynamic environment.

[0114] Generally, the lithium-ion diffusion model is the main tool for describing the diffusion behavior of lithium ions in the battery electrode material, and its core lies in the diffusion coefficient and reaction rate The environmental dependence of the diffusion coefficient and reaction rate is characterized to describe the modulation effect of environmental factors on the lithium-ion diffusion process. As an option, the diffusion model in the present invention is based on the Newman model. By correcting the dynamic characteristic dependence of the diffusion coefficient and reaction rate, the dynamic changes of temperature and humidity are further coupled to obtain an environmentally sensitive form of the diffusion equation.

[0115] In this embodiment, the lithium-ion diffusion model is described by the following partial differential equation:

[0116] ;

[0117] Where:

[0118] represents the lithium-ion concentration, with the unit of 𝑚𝑜l ;

[0119] is the time, with the unit of ;

[0120] is the spatial coordinate in the electrode thickness direction, with the unit of ;

[0121] is the lithium-ion diffusion coefficient, with the unit of s;

[0122] is the current density, with the unit of ;

[0123] is the Faraday constant, with a value of 96485 .

[0124] In the diffusion equation, the diffusion coefficient is a function that varies with the dynamic environmental characteristics and is defined as:

[0125] Where:

[0126] is the reference diffusion coefficient of lithium ions, with the unit of s;

[0127] is the diffusion activation energy, with the unit of J / ;

[0128] is the gas constant, with a value of 8.314 J ;

[0129] is the temperature, with the unit of K ;

[0130] is the humidity, with the unit of %;

[0131] is the humidity sensitivity coefficient, which is used to quantify the influence of humidity on the diffusion coefficient, with the unit of dimensionless.

[0132] Specifically, the diffusion coefficient

[0133] The first term of

[0134] describes the exponential influence of temperature on the diffusion rate, and the second term describes the linear correction of humidity on the diffusion coefficient. In one possible implementation, the correction degree of humidity on the diffusion coefficient can be calibrated by the numerical range of, for example, obtained by fitting the measurement of the diffusion rate in an environment with humidity ranging from 20% to 80%.

[0135] As an option, in this embodiment, the influence of the dynamic environment on the battery electrode reaction rate is further considered, and the environment-sensitive reaction rate is defined as: Where: is the reference value of the reaction rate, with the unit of

[0136] The above reaction rate expression indicates that the reaction rate is mainly affected by temperature and increases exponentially with the increase of temperature. In some embodiments, the definition of can be extended by the influence of humidity on the reaction interface. For example, high humidity may reduce the interface impedance, thereby indirectly increasing the reaction rate.

[0137] In this embodiment, through the coupling of the dynamic environment characteristics and the diffusion model, the dynamic distribution of the lithium ion concentration can be accurately described.

[0138] Specifically, the main frequency in the dynamic environment characteristic vector , and the amplitude , will act on the diffusion coefficient and the reaction rate . In one possible implementation, the direct association between the dynamic characteristics and the diffusion parameters can be achieved in the following way:

[0139] The temperature amplitude by modifying the activation energy , indirectly affecting the diffusion rate;

[0140] Humidity frequency Characterizes the rapidity of humidity fluctuations, thus affecting the humidity correction coefficient of the dynamic response.

[0141] For example, when the main frequency

[0142] of temperature change is relatively high, the diffusion process will exhibit relatively high non-linear dynamic characteristics. At this time, a finer time step is required to solve the diffusion model to capture the transient behavior of the change in lithium-ion concentration.

[0143] In a possible implementation, the diffusion model can also be optimized by the variational method to minimize the error of the diffusion concentration distribution.

[0144] The optimization objective of the variational method can be expressed as: where:

[0145] represents the optimization objective function, which is the square integral of the concentration distribution error;

[0146] is the integration region in the electrode thickness direction.

[0147] By minimizing , the optimized diffusion coefficient and reaction rate parameters can be obtained, thereby further improving the simulation accuracy of the lithium-ion concentration distribution.

[0148] Through the coupling of the lithium-ion diffusion model in this step with the dynamic environmental characteristics, the accurate description of the dynamic environment on the internal physical processes of the battery is successfully achieved. This model can completely capture the dynamic regulation effects of temperature and humidity on the diffusion rate and reaction rate, laying a solid theoretical foundation for the interactive analysis of the discharge performance in the subsequent steps.

[0149] S4. Collect the discharge signals of the polymer lithium-ion battery, and extract the voltage, current, and power signals as the discharge feature vectors;

[0150] After completing the coupling of the dynamic environment and the lithium-ion diffusion model in step S3, to comprehensively evaluate the actual impact of the environment on the performance of the polymer lithium-ion battery, it is also necessary to collect the signal data generated during the discharge of the battery and extract the feature vectors representing the battery performance from it. The collection of the discharge signals mainly includes time series data such as voltage, current, and power, which can truly reflect the discharge behavior and its dynamic change characteristics of the battery under different environmental conditions.

[0151] Generally, the volatility and characteristic changes of the discharge signal are closely related to the electrochemical behavior inside the battery. By preprocessing these signals and extracting features, the battery performance characteristics can be effectively quantified, providing reliable input data for subsequent interaction analysis and performance classification.

[0152] In this embodiment, the discharge signals of the battery, including voltage, current, and power signals, are acquired in real time by a signal acquisition device. The specific implementation method is as follows:

[0153] Specifically, the discharge voltage of the battery

[0154] , the discharge current

[0155] and the power

[0156] are respectively acquired by high-precision sensors, and the sampling frequency and accuracy are set according to actual requirements. The power signal

[0157] has the following calculation formula: Where:

[0158] is the discharge power at time , with the unit of W;

[0159] is the battery terminal voltage at time , with the unit of ;

[0160] is the discharge current at time , with the unit of A.

[0161] Generally, the signal acquisition device needs to meet the requirements of high resolution and low noise to ensure the accuracy and authenticity of the acquired data. For example, in a possible implementation, the sampling frequency can be set to 10 kHz to ensure capturing the fast fluctuation characteristics of voltage and current.

[0162] As an option, the acquired discharge signals can be preprocessed to eliminate the influence of noise and abnormal data on feature extraction.

[0163] Specifically, in some embodiments, the signal preprocessing includes the following steps:

[0164] Filtering process: A low-pass filter is used to remove high-frequency noise, and the cut-off frequency of the filter is set according to the frequency range of the battery discharge signal. For example, the main frequency of the discharge signal is usually lower than 1 kHz, and the cut-off frequency of the filter can be set to 1.2 kHz.

[0165] Outlier Detection: Detect and remove outliers through statistical analysis methods. For example, the standard deviations of voltage and current can be calculated, and the outlier points beyond twice the standard deviation of the mean can be removed.

[0166] Normalization: Normalize the signal amplitude so that feature data with different dimensions can be uniformly processed. The normalization formula is: where:

[0167] is the normalized signal value;

[0168] is the original signal value;

[0169] , are the minimum and maximum values of the signal respectively.

[0170] Through the above preprocessing steps, clean and easy-to-analyze signal data can be obtained, providing a reliable basis for subsequent feature extraction.

[0171] In this embodiment, the extraction of discharge signal features is based on time series analysis methods, converting the time-domain and frequency-domain features of the signal into feature vectors.

[0172] Specifically, the time series features of voltage , current and power include:

[0173] Mean

[0174] : The average value of the signal, used to characterize the overall level of the signal: where is the total length of the signal time, represents the signal value.

[0175] Standard Deviation

[0176] : The volatility of the signal, used to quantify the stability of the signal: Main Frequency in Frequency Domain and Main Amplitude : Extract the main frequency and amplitude of the signal through Fourier transform: where is the amplitude of the signal in the frequency domain.

[0177] In a possible implementation, the extracted discharge features can be further combined into a feature vector as the input for subsequent analysis.

[0178] The definition of the discharge feature vector is: Wherein:

[0179] respectively represent the mean value and standard deviation of the voltage signal; respectively represent the main frequency and main amplitude of the voltage signal;

[0180] respectively represent the mean value and standard deviation of the current signal; respectively represent the main frequency and main amplitude of the current signal;

[0181] respectively represent the mean value and standard deviation of the power signal.

[0182] In some embodiments, the dimension of the feature vector can also be extended to add additional signal features, such as instantaneous maximum value, minimum value, etc.

[0183] Through the acquisition and feature extraction of the discharge signal in this step, complex time series signals can be converted into feature vectors that are easy to analyze, laying a data foundation for the interactive analysis of subsequent dynamic environment features and discharge features. This feature extraction method fully considers the time domain and frequency domain characteristics of the signal, improving the accuracy of analysis and the efficiency of data processing.

[0184] S5. Perform interactive analysis on the environmental feature vector and the discharge feature vector to generate an interactive feature vector;

[0185] After completing the acquisition and feature extraction of the discharge signal in step S4, the dynamic environmental feature vector and the discharge feature vector have been obtained. These two sets of features respectively describe the change rules of temperature and humidity in the dynamic environment and the battery discharge behavior. Since the dynamic environment directly affects the electrochemical reaction and diffusion process of lithium-ion batteries, it is necessary to perform interactive analysis on the environmental features and discharge features to form a unified interactive feature vector, providing complete data input for subsequent performance judgment.

[0186] Generally, the relationship between dynamic environmental features and discharge features is non-linear and time-series complex, and it is difficult to be characterized by a simple linear model. As an option, the present invention adopts the method of feature interaction analysis to fuse the environmental features and discharge features in the feature space, combines the attention mechanism to realize the weight assignment of key features, and generates an interactive feature vector.

[0187] In this embodiment, through the feature interaction analysis method, the dynamic environmental features and discharge features are associated, and the specific implementation method is as follows: the dynamic environmental feature vector and the discharge feature vector are respectively expressed as: Wherein:

[0188] The amplitude and frequency of temperature respectively ；

[0189] The amplitude and frequency of humidity respectively;

[0190] The mean and standard deviation of the voltage signal;

[0191] The mean and standard deviation of the current signal; , The main frequency and main amplitude of voltage and current respectively;

[0192] The mean and standard deviation of the power signal respectively.

[0193] In a possible implementation, feature interaction adopts a method based on the attention mechanism.

[0194] Specifically, the dynamic environment feature vector and the discharge feature vector are mapped into query, key, and value matrices to calculate the correlation between the two through attention weights. The calculation formula of the attention mechanism is as follows: Where:

[0195] Are the query matrix, key matrix, and value matrix respectively; ; is the weight matrix, used to map the feature vector to a high-dimensional feature space;

[0196] Is the dimension of the key vector, used to normalize the attention score.

[0197] In the above formula

[0198] The function is used to calculate the attention weight, and the weight value represents the correlation size between the environment feature and the discharge feature. By weighted averaging the value matrix Generate the interaction feature vector : As an option, the implementation of feature interaction can also be completed by simple vector concatenation or weighted fusion.

[0199] In some embodiments, directly concatenate the dynamic environment feature vector and the discharge feature vector to form the initial interaction feature: To further improve the expression ability of the interaction feature, a non-linear transformation can be performed on the concatenated feature vector, for example, implemented through a fully connected neural network: Where:

[0200] Is the weight matrix, is the bias term;

[0201] is the activation function, such as the ReLU function or the sigmoid function.

[0202] In this embodiment, the interaction feature vector can also be extended according to the actual application scenario.

[0203] Specifically, in some application scenarios, dynamic environmental features may have different effects on different dimensions of the discharge characteristics. For example, the frequency feature of temperature may be more sensitive to voltage fluctuations, while the amplitude feature of humidity may have a greater impact on current fluctuations. In this case, by introducing the multi-head attention mechanism, the influence weights of environmental features on different discharge characteristics can be calculated separately.

[0204] The calculation formula of the multi-head attention mechanism is:

[0205] Where:

[0206] represents the th attention head;

[0207] is the th weight matrix of the attention head;

[0208] is the output weight matrix, which is used to map the result of the multi-head attention back to the output space.

[0209] Through the multi-head attention mechanism, the expression ability of the interaction features can be significantly improved, and at the same time, the complex influence of the dynamic environment on the battery discharge performance can be captured more accurately.

[0210] Through the feature interaction analysis of this step, the generated interaction feature vector contains both the information of the dynamic environmental features and the variation law of the discharge characteristics.

[0211] This interaction feature vector provides a complete and rich data input for the subsequent performance classification, and its result can better reflect the dynamic characteristics of the battery performance changing with the environment. Whether through the attention mechanism or the direct splicing and fusion method, this step lays an important foundation for performance analysis and diagnosis.

[0212] S6. Based on the interaction feature vector, determine whether the discharge performance of the polymer lithium-ion battery is normal.

[0213] After completing the interactive analysis of the environmental characteristics and the discharge characteristics in step S5, an interactive feature vector has been generated, which synthesizes the complex relationship between the dynamic environment and the discharge signal. To further achieve an accurate judgment of the battery discharge performance, in this step, based on the interactive feature vector, through a deep learning classification model or other appropriate analysis methods, the classification and determination of the discharge performance are completed.

[0214] Generally, the interactive feature vector contains rich dynamic information, such as the influence relationship between the temperature and humidity changes and the voltage and current fluctuations. These features can reflect the potential abnormal states of the battery performance, such as the ion migration obstacles inside the battery or the voltage abnormality caused by the electrolyte decomposition. Therefore, accurately identifying these states requires processing the feature vector through a classifier or an analysis model to determine whether the battery discharge performance is normal.

[0215] In this embodiment, based on the interactive feature vector

[0216] to judge the discharge performance of the polymer lithium-ion battery, the specific implementation method is as follows: Specifically, the interactive feature vector is the dynamic environmental characteristics and the discharge characteristics fusion result: Among them:

[0217] represents the main amplitude and frequency characteristics of the dynamic environment; represents the statistical and frequency domain characteristics of the discharge signal.

[0218] The interactive feature vector is input into the classification model to judge whether the battery discharge performance is normal.

[0219] In a possible implementation manner, the classification model adopts a deep neural network (DNN), and the specific structure is as follows:

[0220] Input layer: Receive the interactive feature vector , the feature dimension is , where Hidden layer: Consists of several fully connected layers, each layer contains neurons, and adopts an activation function , such as the ReLU function: Output layer: Output the classification result of the discharge performance , and normalize the probability of each category through the softmax function: Among them:

[0221] is the output weight matrix;

[0222] is the output of the last hidden layer;

[0223] is the output layer bias term.

[0224] The training data of the model can be generated from the discharge performance data collected in the laboratory, and the training objective is to minimize the cross-entropy loss: where:

[0225] is the number of samples;

[0226] is the number of classes (e.g., two classes of "normal" and "abnormal");

[0227] is the sample belongs to the class true label;

[0228] is the probability that the model predicts for the sample belongs to the class prediction probability.

[0229] As an option, a method of probability distribution analysis can also be used to judge the discharge performance.

[0230] Specifically, each dimension feature in the interaction feature vector is regarded as a high-dimensional random variable, and the feature distribution under normal conditions is estimated through a Gaussian distribution model: where:

[0231] is the feature mean vector;

[0232] is the feature covariance matrix;

[0233] is the feature dimension.

[0234] By calculating the probability density value of the interaction feature vector

[0235] , judge whether it significantly deviates from the normal distribution. If

[0236] is less than a certain threshold, it is determined to be an abnormal state.

[0237] In a possible implementation, the classification result can also be corrected by combining rule-based diagnostic logic.

[0238] For example, when the temperature amplitude and humidity amplitude When it exceeds a specific range simultaneously, the battery performance can be directly determined to be abnormal without relying on the results of the classifier. This method can effectively enhance the robustness of the model and is applicable to specific extreme environmental conditions.

[0239] In this embodiment, the classification result can be directly output in two states: "normal" or "abnormal", and is used to further optimize the battery design and usage conditions.

[0240] Based on the classification result, the specific environmental factors affecting the battery performance can also be analyzed. For example, if the determination result shows that the humidity frequency and the main frequency of voltage fluctuation are significantly correlated, it indicates that humidity fluctuation is the main cause of performance abnormality. In this case, the humidity adaptability can be improved by optimizing the battery sealing or adjusting the electrolyte formula.

[0241] Through the classification analysis of this step, the accurate judgment of the discharge performance of the polymer lithium-ion battery can be achieved, providing an important basis for the battery state monitoring and optimization. This step is closely combined with the aforementioned feature interaction analysis step, and the environmental and discharge features in the interaction feature vector are used to comprehensively analyze the battery performance changes, thus significantly improving the accuracy and efficiency of the diagnosis.

[0242] A test and analysis system for polymer lithium-ion batteries based on deep learning and feature analysis includes:

[0243] A signal acquisition module for acquiring the discharge signal of the polymer lithium-ion battery and the temperature and humidity data of the battery working environment;

[0244] This module is used to acquire the working state data of the polymer lithium-ion battery in real time, including the voltage, current, and power signals generated during the discharge process, as well as the temperature and humidity data in the battery working environment. The signal acquisition module acquires the original data through high-precision sensors and performs filtering and denoising processing on the data through a signal processing unit to ensure the accuracy and effectiveness of the data.

[0245] Implementation method

[0246] Use high-resolution voltage sensors and current sensors to acquire the discharge signal;

[0247] Obtain environmental data through temperature and humidity sensors, and the acquisition frequency is set according to dynamic change requirements, such as 10Hz or higher;

[0248] The data preprocessing part includes removing noise with a low-pass filter, as well as detecting and removing outliers.

[0249] A dynamic environment modeling module for establishing a dynamic distribution model of temperature and humidity;

[0250] Based on the temperature and humidity data provided by the signal acquisition module, the dynamic environment modeling module uses partial differential equations to model and simulate the dynamic distribution of environmental temperature and humidity. By simulating the diffusion, convection of temperature and humidity, and external disturbances, this module generates a dynamic model describing the changes of temperature and humidity over time and space.

[0251] A feature extraction module, which is used to extract the dynamic features of temperature and humidity and the features of the discharge signal;

[0252] Based on the data provided by the dynamic environment modeling module and the signal acquisition module, the feature extraction module extracts the main amplitude, frequency and other dynamic features of temperature and humidity, as well as the mean value, standard deviation and frequency domain features of the discharge signal, and forms a feature vector.

[0253] A feature interaction analysis module, which is used to perform interaction analysis on the dynamic environment features and the discharge features to generate an interaction feature vector;

[0254] Through the feature interaction analysis method, this module fuses the dynamic environment features and the discharge features to generate an interaction feature vector. The interaction analysis can capture the complex non-linear relationship between the dynamic environment and the discharge performance.

[0255] A performance evaluation module, which is used to judge whether the discharge performance of the polymer lithium-ion battery is normal according to the interaction feature vector and output the performance evaluation result.

[0256] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A lithium battery testing and analysis method based on deep learning and feature analysis, characterized in that: include: S1. Establish a dynamic environment model to obtain the dynamic distribution of temperature and humidity in the battery use environment, where the dynamic distribution of temperature and humidity is based on the combined influence of diffusion, convection and external disturbance; S2, extract dynamic environmental features, and use the main amplitude and frequency of temperature and humidity as environmental feature vectors; S3. Based on the lithium ion diffusion model, a coupling relationship between lithium ion concentration distribution and environmental characteristics is established, wherein the lithium ion concentration distribution is affected by changes in temperature and humidity in a dynamic environment; S4, collecting the discharge signal of the polymer lithium-ion battery, and extracting the voltage, current and power signals as the discharge feature vector; S5, interactively analyzing the environment feature vector and the discharge feature vector to generate an interactive feature vector; S6. judging whether the discharge performance of the polymer lithium-ion battery is normal based on the interactive feature vector; The dynamic environment model in step S1 is established based on a two-dimensional partial differential equation to describe the distribution and variation of temperature and humidity in time and space; Specifically, the dynamic change of temperature is expressed as the following partial differential equation: The dynamic change of humidity is expressed as: Where: T(t, x, y) represents the temperature at the coordinate (x, y) at a certain time t, in K; H(t, x, y) represents the humidity at the coordinate (x, y) at a certain time t, in %RH; α T is the temperature diffusion coefficient, in m 2 / s, represents the diffusion rate of temperature in space; α H is the humidity diffusion coefficient, in m 2 / s, represents the diffusion rate of humidity in space; β T is the temperature convection coefficient, in m / s, indicating the influence of convection on temperature; β H is the humidity convection coefficient, in m / s, indicating the influence of humidity on convection; Q T (t, x, y) is the temperature change rate caused by the external heat source, in K / s, caused by external disturbances, and the rapid fluctuation of the ambient temperature; Q H (t, x, y) is the rate of change of humidity caused by the external humidity source, in %RH / s; Extracting the dynamic environment features in step S2 includes: performing time-frequency domain analysis on the dynamic distribution of temperature and humidity, obtaining their main amplitudes and frequencies, and forming a dynamic environment feature vector.

2. The lithium battery testing and analysis method based on deep learning and feature analysis according to claim 1, characterized in that: The lithium ion diffusion model in step S3 describes the influence of temperature and humidity on the diffusion coefficient of lithium ions, and adjusts the diffusion rate and reaction rate through the dynamic environment characteristic vector.

3. The lithium battery testing and analysis method based on deep learning and feature analysis according to claim 1, characterized in that: The discharge signal collected in step S4 includes voltage, discharge current and power value in a time series, and the discharge signal is preprocessed to obtain a discharge feature vector.

4. The lithium battery testing and analysis method based on deep learning and feature analysis according to claim 1, characterized in that: In step S5, a feature interaction analysis method is adopted to fuse the environment feature vector and the discharge feature vector through an attention mechanism to generate an interaction feature vector.

5. The lithium battery testing and analysis method based on deep learning and feature analysis according to claim 4, characterized in that: The attention mechanism is implemented by the following steps: Mapping the environment feature vector and the discharge feature vector into a query vector, a key vector and a value vector respectively; Calculate the similarity score between the query vector and the key vector; The value vectors are weighted averaged using the similarity scores to generate an interaction feature vector.

6. The lithium battery testing and analysis method based on deep learning and feature analysis according to claim 1, characterized in that: In the step S6, whether the discharge performance of the polymer lithium-ion battery is normal is judged by a classification model based on the interactive feature vector, and the classification model is a deep neural network classifier.

7. The lithium battery testing and analysis method based on deep learning and feature analysis according to claim 6, characterized in that: The classification model determines the battery performance through the probability distribution of the interactive feature vector and outputs the classification result of the normal state or the abnormal state.

8. A lithium battery test and analysis system based on deep learning and feature analysis, based on the lithium battery test and analysis method based on deep learning and feature analysis according to any one of claims 1 to 7, characterized in that: include: Signal acquisition module, used to collect discharge signals of polymer lithium-ion batteries and temperature and humidity data of the battery working environment; Dynamic environment modeling module, used to establish dynamic distribution models of temperature and humidity; A feature extraction module, used to extract the dynamic features of temperature and humidity and the features of the discharge signal; A feature interaction analysis module is used to interactively analyze dynamic environment features and discharge features to generate interactive feature vectors; The performance evaluation module is used to determine whether the discharge performance of the polymer lithium-ion battery is normal according to the interactive feature vector and output the performance evaluation result.

Images