Lithium battery full-band electrochemical impedance spectroscopy rapid detection method and system

Abstract

The invention discloses a rapid detection method and system for a full-band electrochemical impedance spectroscopy of a lithium battery, and the method comprises the steps: firstly obtaining a Bode diagram of the electrochemical impedance spectroscopy of the lithium battery, and recognizing and extracting characteristic frequency points representing the transformation of different electrochemical processes; then, on the basis of the characteristic frequency points, an EK-DIBS excitation sequence with spectrum energy intensively distributed on the characteristic frequency points is constructed by adopting a discrete interval binary sequence method; after the excitation is applied to the battery, voltage and current response signals are synchronously acquired, and impedance values corresponding to the characteristic frequency points are extracted through fast Fourier transform; and finally, inputting the impedance values and the working condition parameters of the battery into a pre-trained sequence attention nucleus SeqKAN network model, and reconstructing a complete full-band electrochemical impedance spectrum. According to the method, the number of excitation signal frequency points and the sampling frequency requirement are remarkably reduced, the detection time is greatly shortened, the sampling and calculation load is reduced, and the problem that the detection speed, the implementation complexity and the measurement precision are difficult to consider into account in an embedded system is solved.

Description

Technical Field

[0001] This invention belongs to the field of lithium battery testing technology, and particularly relates to a rapid detection method and system for full-band electrochemical impedance spectroscopy of lithium batteries. Background Technology

[0002] The widespread application of large-scale lithium-ion battery energy storage power stations in new energy power generation and power grid systems has significantly improved the power system's ability to absorb renewable energy and its operational resilience. However, a series of thermal runaway safety accidents at energy storage power stations worldwide has triggered an urgent need for precise monitoring and proactive safety early warning technologies for the operating status of electrochemical energy storage systems. Electrochemical impedance spectroscopy (EIS) detection technology, by analyzing the frequency response characteristics of batteries to small-amplitude external AC excitations, can non-invasively obtain dynamic information on key electrochemical processes inside the battery. It has shown great potential in areas such as early diagnosis of battery faults, health status assessment, and state of charge estimation, and has become a research hotspot in the field of battery safety management.

[0003] Currently, research on rapid EIS detection technology mainly focuses on two directions: excitation signal optimization and efficient impedance data extraction. In terms of excitation design, mainstream methods aim to shorten measurement time by synthesizing multi-frequency signals to cover a wide frequency range with a single excitation. Specific technical approaches include: optimizing the amplitude, frequency, and phase of multiple sinusoidal signals to suppress peak factor; using step signals for excitation and employing time-frequency transformation techniques to analyze impedance; using pseudo-random binary sequences and other two-level signals to reduce the complexity of the excitation generation circuit; and designing discrete-interval binary sequences to concentrate signal energy at specific key frequency points to improve the signal-to-noise ratio. Regarding impedance extraction, for response processing of aggregated excitations, methods mainly include frequency domain analysis based on Fast Fourier Transform, filtering algorithms combining excitation power spectrum characteristics (such as 3D cloud filtering), and reconstructing the complete impedance spectrum from partial frequency points or high-frequency impedance data using neural network models.

[0004] However, existing technologies still face the following core contradictions when applied to engineering applications of embedded online monitoring: the excitation design and impedance extraction stages are independent, making it difficult to simultaneously achieve high detection speed, complex system implementation, and high measurement accuracy. Specifically, to achieve rapid detection, aggregated excitation covering a large number of frequencies is often used, but this leads to complex excitation signal amplitude control, large sample data volume, and high frequency requirements. Simultaneously, to ensure the extraction of full-band impedance from low signal-to-noise ratio responses, it is necessary to rely on computationally complex algorithms or large amounts of training data. This disconnect between "front-end excitation" and "back-end processing" makes it difficult for existing methods to simultaneously meet the stringent requirements of embedded systems for real-time detection, hardware resource constraints, and adaptability to dynamic operating conditions without sacrificing accuracy, severely restricting the large-scale engineering application of EIS technology in battery online monitoring scenarios. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a rapid detection method and system for full-band electrochemical impedance spectroscopy of lithium batteries. By synergistically optimizing excitation design and impedance extraction, the method ensures the accuracy and cross-condition stability of full-band impedance detection while meeting the real-time requirements of embedded systems. This solves the core contradiction of the difficulty in balancing detection speed, implementation difficulty, and measurement accuracy.

[0006] Technical solution: The rapid detection method for full-band electrochemical impedance spectroscopy of lithium batteries according to the present invention includes the following steps:

[0007] S1: Obtain the electrochemical impedance spectroscopy Bode plot of the lithium battery. Based on the characteristics of the impedance amplitude and phase angle changing with frequency in the Bode plot, identify and extract the characteristic frequency points that characterize the transition of different electrochemical processes.

[0008] S2: Based on the characteristic frequency point, an excitation sequence is constructed using the discrete interval binary sequence method, so that the spectral energy of the excitation sequence is concentrated at the characteristic frequency point, thus obtaining the characteristic frequency point-guided discrete interval binary sequence EK-DIBS excitation sequence;

[0009] S3: Apply the EK-DIBS excitation sequence to the lithium battery, collect the voltage response signal and current response signal generated by the lithium battery under the excitation, and extract the impedance value corresponding to the characteristic frequency point through fast Fourier transform;

[0010] S4: Input the impedance value corresponding to the characteristic frequency point and the operating condition parameters related to the lithium battery's operating state into the pre-trained Sequence Attention Kernel (SeqKAN) network model, and output and reconstruct the full-band electrochemical impedance spectrum of the lithium battery.

[0011] This invention utilizes a frequency-guided excitation sequence design to concentrate spectral energy at key frequencies characterizing different electrochemical processes, achieving precise matching between the excitation signal and the impedance characteristics of the lithium battery. Based on this, a Fast Fourier Transform (FFT) is employed to directly extract impedance at characteristic frequencies, significantly reducing data acquisition and processing complexity and dramatically improving detection speed. Furthermore, a sequence attention kernel network model is combined to reconstruct the full-band impedance spectrum with high precision using finite characteristic frequency impedance and operating parameters. This method, through the synergistic design of excitation optimization and impedance extraction, effectively balances the accuracy and reliability of full-band impedance detection with cross-operating-condition adaptability while ensuring the real-time performance and ease of engineering implementation of the embedded system. It fundamentally solves the core contradiction in traditional methods where it is difficult to balance detection speed, implementation difficulty, and measurement accuracy.

[0012] Preferably, step S1, which involves identifying and extracting characteristic frequency points representing the transitions in different electrochemical processes, includes:

[0013] S1.1: Obtain electrochemical impedance spectroscopy test data of lithium batteries under different states of charge, and plot Bode plots of impedance amplitude |Z| and phase angle θ as a function of frequency f;

[0014] S1.2: Calculate the derivative dθ / d(logf) of the phase angle θ with respect to the frequency log(f) in the Bode plot, identify the extreme points and inflection points on the derivative curve, and the frequencies corresponding to the extreme points and inflection points are the critical frequencies that characterize the change of dominance of different electrochemical processes inside the battery.

[0015] S1.3: Based on the electrochemical mechanism, multiple candidate characteristic frequency points are selected from the critical frequency;

[0016] S1.4: Based on the SHAP interpretability analysis method, analyze and evaluate the contribution of each candidate characteristic frequency point to impedance reconstruction, and select the optimal frequency point combination by combining the mean absolute error index, and determine the optimal frequency point combination as the final characteristic frequency point set.

[0017] By calculating the phase derivative curve and combining it with knowledge of electrochemical mechanisms, the critical frequency points characterizing the transitions in key kinetic behaviors such as internal solid-liquid diffusion, charge transfer, and Ohmic processes are accurately identified from the Bode plot. Based on this, SHAP interpretability analysis is further introduced to objectively quantify and screen the set of core characteristic frequencies that contribute most to impedance reconstruction across the entire frequency band. This step ensures that the extracted characteristic frequency set has clear electrochemical and physical significance and optimal data representation capabilities. This not only significantly enhances the targeting and spectral efficiency of subsequent excitation signal design but also provides deep learning models with strong feature inputs that reliably reflect the essence of the battery state, thus providing a solid guarantee for improving the accuracy and generalization ability of the overall detection method from the source.

[0018] Preferably, the candidate feature frequency points mentioned in step S1.3 include:

[0019] The characteristic frequency point in the low-frequency region is located at the critical position where the phase angle transitions from being dominated by charge transfer to being dominated by the diffusion impedance process (Warburg), marking the beginning of the solid-phase diffusion process dominating the system response;

[0020] The characteristic frequency points in the low-to-mid frequency region are located at the boundary of the coupling region where solid-phase diffusion and interfacial charge transfer compete with each other, characterizing the alternation of kinetic dominance between lithium-ion concentration relaxation and surface Faraday reaction within active particles.

[0021] The mid-frequency characteristic point is located at the boundary between the relaxation process and charge transfer reaction of the SEI film at the solid electrolyte interface. Above the mid-frequency characteristic point, the RC time constant response of the SEI film gradually weakens, and the Faraday process at the electrode / electrolyte interface begins to dominate the impedance characteristics. The RC time constant reflects the response speed of the SEI film to the AC excitation signal.

[0022] The mid-to-high frequency region is the boundary between bulk ion conduction and SEI membrane ion transport. After crossing the mid-to-high frequency region, the ion migration resistance of the SEI membrane separates from the bulk ohmic response and becomes an identifiable independent impedance contribution.

[0023] The candidate feature frequencies selected in this invention precisely cover the key kinetic transition regions from high-frequency bulk ohmic conduction, mid-to-high-frequency SEI membrane ion transport, the boundary between mid-frequency SEI membrane relaxation and charge transfer reactions, the mid-to-low-frequency solid-phase diffusion and charge transfer coupling boundary, to the low-frequency solid-phase diffusion-dominant transition. Each frequency point has a clear physical meaning, corresponding to the critical state where the dominance of electrochemical processes at different time scales within the battery alternates. This systematic selection of frequency points based on physical mechanisms not only ensures that the captured features can comprehensively and essentially reflect the dynamic behavior of the battery across multiple time scales, but also constructs a highly interpretable and information-density impedance characterization system, providing a solid physical foundation and reliable data support for subsequent high-fidelity reconstruction of impedance from limited key features to the entire frequency range.

[0024] Preferably, the method for constructing the EK-DIBS excitation sequence in step S2 includes:

[0025] S2.1: Define a period of T and an amplitude of binary sequence And calculate its Fourier series. ,in The Fourier series represents the harmonic order. Represented as:

[0026]

[0027] in, The amplitude of the binary sequence is T, the sequence period is j, and the imaginary unit is j.

[0028] S2.2: Based on the voltage and current sampling noise levels and the preset maximum permissible measurement error, calculate the lower limit of the excitation signal harmonic amplitude that meets the minimum signal-to-noise ratio requirement at the target characteristic frequency. The lower limit of the amplitude The calculation expression is:

[0029]

[0030] in, The maximum permissible measurement error is... This represents the amplitude of the battery's ohmic internal resistance. and These represent the voltage sampling noise amplitude and the current sampling noise amplitude at the k-th harmonic frequency, respectively.

[0031] S2.3: Construct the expected signal Its Fourier series is represented as and to minimize the binary sequence With the expected signal Frequency domain deviation between To optimize the objective, a constrained optimization model is established:

[0032]

[0033]

[0034] S2.4: For the expected signal Discrete sequences are obtained by performing discrete sampling. According to the preset sign function, Quantized into binary sequence The quantization rule is as follows:

[0035]

[0036] According to the discrete sequence Calculate the amplitude estimate of the binary sequence. :

[0037]

[0038] Where N is the length of the discrete sequence;

[0039] S2.5: Set the expected signal The initial Fourier series form is ,in The initial phase angle is randomly generated;

[0040] S2.6: Update the expected signal using an iterative optimization algorithm. The phase angle is adjusted until the convergence condition is met or the preset maximum number of iterations is reached. A single iteration of the iterative optimization algorithm includes:

[0041] Based on the current phase angle Calculate the Discrete Fourier Transform of the intermediate sequence ,in for The discrete Fourier transform;

[0042] right The updated discrete sequence is obtained by performing an inverse discrete Fourier transform. ;

[0043] Will Quantized into binary sequence And calculate its discrete Fourier transform. ;

[0044] extract phase angle The input phase angle for the next iteration ;

[0045] S2.7: After the iteration is complete, select the candidate binary sequence from all the candidate binary sequences generated by the iterations that results in the frequency domain deviation. The smallest sequence is determined as the final EK-DIBS excitation sequence; the spectral energy of the final EK-DIBS excitation sequence is concentrated at the characteristic frequency points determined in step S1.

[0046] By constructing a multi-objective optimization framework that balances frequency domain energy concentration, minimum quantization distortion, and system noise robustness, a feature frequency-guided binary excitation sequence was successfully designed. This design first dynamically determines the minimum energy threshold for each feature frequency based on preset accuracy and signal-to-noise ratio requirements. Then, through an iterative optimization algorithm, it achieves an optimal balance between accurately approximating the ideal spectral distribution and the physical realizability of the binary signal. The resulting EK-DIBS excitation sequence can accurately and efficiently inject the vast majority of energy into pre-identified key feature frequencies. This design ensures that, under finite amplitude constraints, the excitation signal can produce a strong response at the target frequency far exceeding the noise level. Simultaneously, its inherent binary characteristics significantly reduce the linearity requirements of the power amplifier and the hardware implementation cost, thus bridging the gap between engineering usability and measurement accuracy. This lays a solid foundation for achieving fast, high signal-to-noise ratio impedance feature extraction on embedded platforms.

[0047] Preferably, the method for extracting the impedance value in step S3 includes:

[0048] S3.1: The EK-DIBS excitation sequence is superimposed on the charge / discharge current of the lithium battery as a measurement excitation, and the total terminal voltage signal containing the excitation response component is acquired simultaneously. With total current signal ;

[0049] S3.2: Convert the total terminal voltage signal With total current signal The components are decomposed into corresponding operating condition components and excitation response components, and their decomposition relationship is expressed as follows:

[0050]

[0051]

[0052] in, and This represents the total terminal voltage and total current corresponding to the m-th sampling point. and Let represent the m-th voltage and current components generated during normal charging and discharging of the battery, respectively. and These represent the m-th voltage response component and the current response component generated by the measurement excitation, respectively, where m is the sampling point number, ranging from 1 to... , The length of the signal interval used for impedance calculation;

[0053] S3.3: Based on the total current signal Calculate the estimated value of the current operating condition component. The estimated value is obtained by averaging the total current signal within the signal interval, and the calculation expression is as follows:

[0054]

[0055] in, Represents the discrete sampled values ​​of the total battery current signal;

[0056] S3.4: Based on the total terminal voltage signal The estimated values ​​of the voltage operating condition components are obtained by curve fitting. The specific method is as follows: select the moment of stimulus injection as the center, and the time before and after it. The fitting interval consists of 2 sampling points, where Given the data sampling rate, a cubic polynomial fitting is performed on the voltage sampling data points within the fitting interval to obtain the fitting function. The estimated voltage condition component at the m-th sampling point within the impedance measurement interval is:

[0057]

[0058] S3.5: Based on the decomposition relationship described in step S3.2 and the estimated values ​​obtained in steps S3.3 and S3.4, calculate the excitation response current respectively. With excitation response voltage :

[0059]

[0060]

[0061] S3.6: Response current to the excitation and excitation response voltage Perform a discrete Fourier transform to obtain its frequency... Frequency domain complex representation and :

[0062]

[0063]

[0064] Where j is the imaginary unit, For harmonic order;

[0065] S3.7: According to Ohm's law, calculate the frequency... Complex impedance value at :

[0066]

[0067] Wherein, the frequency Corresponding to the characteristic frequency points determined in step S1.

[0068] By superimposing characteristic frequency excitations onto the normal operating conditions of the battery and effectively separating the steady-state components and dynamic excitation response components in the voltage and current using mean and curve fitting techniques, impedance measurement can be performed online without interrupting normal battery operation. This method, combined with discrete Fourier transform, can extract complex impedance values ​​at preset characteristic frequencies from the response signal with high precision. This strategy of simultaneous measurement and precise decomposition not only greatly enhances the adaptability and online feasibility of impedance detection, avoiding system interruptions and state disturbances caused by traditional offline measurements, but also ensures the extraction of weak excitation response signals from strong background noise, providing accurate, reliable, and physically meaningful frequency domain feature inputs for subsequent high-fidelity reconstruction of the full-band impedance spectrum.

[0069] Preferably, the training method for the SeqKAN network model in step S4 includes:

[0070] Training dataset construction: Under various ambient temperature conditions and various charge / discharge rates, dynamic electrochemical impedance spectroscopy tests are performed on lithium batteries to obtain complex impedance spectroscopy data covering the preset full frequency range as real labels; from the complex impedance spectroscopy data, complex impedance values ​​corresponding to the characteristic frequency points determined in step S1 are extracted as model input samples; the input samples are combined with the corresponding real labels and divided into training set and test set according to a preset ratio;

[0071] Network model training: Set training hyperparameters including batch size, learning rate, number of training epochs, temporal hidden dimension, and conditional hidden dimension; use the training set to perform supervised training on the SeqKAN network model; during training, use mean squared error as the loss function to calculate the error between the model's predicted impedance spectrum and the true impedance spectrum. The loss function expression is:

[0072]

[0073] Where N is the length of the discrete sequence, and For the real and imaginary parts of the predicted impedance, and These are the real and imaginary parts of the actual impedance;

[0074] The weight parameters of the SeqKAN network model are updated using a gradient descent-based optimization algorithm until the loss function converges or reaches a preset number of training rounds, thus obtaining a trained SeqKAN network model.

[0075] By constructing a dynamic impedance spectrum dataset covering complex operating conditions such as multiple temperatures and rates for model training, the constructed SeqKAN network is able to learn and capture the common patterns and evolution modes of lithium battery impedance spectra under different operating conditions. It employs a mean squared error loss function and gradient descent optimization to drive the model to accurately fit the complex nonlinear mapping relationship from impedance at finite feature frequencies to the full-band impedance spectrum. The model trained in this way can fully utilize the sequence attention mechanism to effectively integrate the impedance characteristics of time-series inputs with operating condition information, achieving high-precision, cross-condition stable reconstruction of the full-band electrochemical impedance spectrum. This not only significantly reduces the time and hardware cost of traditional full-frequency scanning but also endows the detection system with strong generalization ability and operating condition adaptability, ultimately realizing fast, convenient, and reliable embedded online impedance monitoring at the engineering level.

[0076] Preferably, step S4, which involves outputting and reconstructing the full-band electrochemical impedance spectroscopy of the lithium battery, includes:

[0077] S4.1: Construct network input data by arranging the impedance values ​​extracted in step S3 from low to high according to the frequency of the corresponding characteristic frequency points to form a time-series impedance sequence, wherein the impedance value of each characteristic frequency point contains the real part. and the virtual part The two components together constitute the input feature tensor. Where B is the batch size, T is the sequence period, and F is the number of input features at each time step; simultaneously, the operating condition parameters of the lithium battery during its current operation are obtained to form a conditional input vector. The operating condition parameters include battery ambient temperature, charge / discharge rate, and state of charge. The number of condition variables;

[0078] S4.2: Perform conditional encoding and fusion, converting the conditional input vector... The input conditional encoder performs feature encoding to obtain the conditional encoded vector. ,in The conditional hiding dimension is used; the conditional encoding vector is extended along the time dimension to obtain the conditional information. The condition information is aligned with the dimension of the time-series impedance sequence;

[0079] S4.3: Perform time-series recursive calculations and initialize the hidden state. For a zero vector, perform the following recursive calculation for each time step t from 1 to T: First, extract the input features of the current time step. The embedding vector is obtained by performing a nonlinear transformation through the feature embedding layer. Then the embedding vector Hidden state from the previous moment After concatenation, the result is input into the timing processing layer. And update the hidden state through residual connections. Finally, the updated hidden state is concatenated with the expanded conditional information to obtain the merged hidden state. In the SeqKAN network model, the activation functions of the feature embedding layer and the temporal processing layer... A weighted combination of learnable spline functions and fixed activation functions is used: ,in For a fixed activation function, For learnable B-spline functions, and These are trainable weight parameters;

[0080] S4.4: Perform full-band impedance output, stacking the fused hidden states of all time steps along the time dimension to form a complete hidden representation. ,in , To hide the time dimension; to represent the hidden dimension Input / output layer The output layer predicts and outputs the real and imaginary parts of the impedance within a preset frequency range, forming a complete full-band electrochemical impedance spectrum.

[0081] By constructing a fused input of time-series impedance sequences and multi-dimensional operating parameters, and utilizing a sequence attention kernel network based on learnable splines and residual connections for deep processing, the model can accurately capture the time-series dependence of impedance characteristics on frequency evolution and the complex nonlinear coupling relationship between impedance and dynamic operating conditions. This network structure, through conditional encoding and time-step fusion, ensures that the model output can adaptively reflect the precise changes in impedance spectra under different temperatures, rates, and charging states. Ultimately, the model achieves high-fidelity, strong generalization reconstruction from impedance at finite characteristic frequencies to impedance spectra across the entire frequency band. This enables the online generation of accurate, continuous, and physically meaningful electrochemical impedance spectra with extremely low computational cost, without the need for hardware scanning. This provides crucial and reliable technical support for real-time state diagnosis and lifetime prediction in embedded battery management systems.

[0082] Secondly, the lithium battery full-band electrochemical impedance spectroscopy rapid detection system of the present invention includes:

[0083] The feature frequency extraction module is used to obtain the electrochemical impedance spectrum Bode plot of lithium battery, and to identify and extract feature frequency points that characterize the transition of different electrochemical processes based on the characteristics of impedance amplitude and phase angle changes with frequency in the Bode plot.

[0084] The excitation sequence generation module is used to construct an excitation sequence based on the feature frequency points extracted by the feature frequency point extraction module using a discrete interval binary sequence method, so that the spectral energy of the excitation sequence is concentrated at the feature frequency points, thereby generating a feature frequency point-guided discrete interval binary sequence excitation sequence.

[0085] The excitation application and signal acquisition module is used to apply the excitation sequence generated by the excitation sequence generation module to the lithium battery, and simultaneously acquire the voltage response signal and current response signal generated by the lithium battery under the excitation action;

[0086] The local impedance calculation module is connected to the excitation application and signal acquisition module. It is used to process the acquired voltage response signal and current response signal, and calculate the impedance value corresponding to the characteristic frequency point through fast Fourier transform.

[0087] The full-band impedance reconstruction module, connected to the local impedance calculation module, is used to receive the impedance value corresponding to the characteristic frequency point, and combine it with the operating condition parameters related to the lithium battery's operating state, inputting them into the pre-trained sequence attention kernel network model to reconstruct and output the full-band electrochemical impedance spectrum of the lithium battery.

[0088] Thirdly, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed by the described rapid detection method for full-band electrochemical impedance spectroscopy of lithium batteries.

[0089] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned rapid detection method for full-band electrochemical impedance spectroscopy of lithium batteries.

[0090] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: 1. By co-optimizing excitation design and impedance extraction, this invention constructs an excitation sequence oriented towards key feature frequencies, significantly reducing the number of excitation signal frequencies and sampling frequency requirements. While ensuring impedance reconstruction accuracy across the entire frequency band, it greatly shortens the detection time and reduces sampling and computational load, solving the problem of balancing detection speed, implementation complexity, and measurement accuracy in embedded systems; 2. Through feature frequency-oriented excitation design and a lightweight sequence network model based on an attention mechanism, while maintaining high-precision reconstruction capabilities, it effectively compresses data volume and computational complexity, making the system suitable for… 3. By combining conditional coding and time-series fusion strategies, the impedance reconstruction model can adapt to changes in operating conditions such as temperature, rate, and state of charge, maintaining stable impedance reconstruction accuracy across the entire frequency band and enhancing the applicability and reliability of the method in actual dynamic operating environments. 4. By combining concentrated distribution of excitation energy with intelligent reconstruction, high-fidelity reconstruction of the impedance spectrum across the entire frequency band can still be achieved while reducing excitation frequencies and sampling rates. This systematically solves the conflict between excitation coverage, signal strength, real-time performance, and accuracy in traditional methods, enhancing the engineering practical value of electrochemical impedance spectroscopy detection. Attached Figure Description

[0091] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0092] Figure 2 This is a schematic diagram of the Bode plot and characteristic frequency extraction of the lithium battery electrochemical impedance spectroscopy of the present invention;

[0093] Figure 3 This is a comparison and verification diagram of the impedance spectra of EK-DIBS and DIBS of the present invention. Detailed Implementation

[0094] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0095] Example 1:

[0096] Figure 1 This is a flowchart of the efficient lithium battery full-band impedance extraction method according to Embodiment 1 of the present invention. This flowchart only illustrates the logical sequence of the method described in this embodiment. Provided there are no conflicts, different methods may be used in other possible embodiments of the present invention. Figure 1 Perform the steps shown or described in the order indicated. See also Figure 1 The method in this embodiment specifically includes the following steps:

[0097] S1: Based on the evolution of impedance amplitude and phase angle with frequency in the Bode plot of battery electrochemical impedance spectroscopy, characteristic frequency points marking the transition of different electrochemical processes are extracted.

[0098] S2: Based on the characteristic frequency points, an EK-DIBS excitation sequence is designed using the discrete interval binary sequence method to concentrate the signal spectrum energy at the characteristic frequency points;

[0099] S3: Apply the EK-DIBS excitation sequence to the lithium battery, collect the battery's voltage and current response signals, and extract the impedance value at the characteristic frequency point through FFT transformation;

[0100] S4: Input the characteristic frequency impedance values ​​and operating condition parameters into the pre-trained SeqKAN network to reconstruct the full-band electrochemical impedance spectrum.

[0101] In practical applications, to achieve efficient detection of the full-band impedance of a battery, the efficient extraction method for the full-band impedance of a lithium battery according to this invention is implemented as follows: Taking a lithium-ion battery as the research object, firstly, characteristic frequency points are extracted and EK-DIBS excitation sequences are designed. Then, the SeqKAN network is used to reconstruct the impedance from 5 characteristic frequency points to 40 full-band impedances.

[0102] First, combined Figure 2 (Diagram of Bode plot and characteristic frequency extraction of lithium battery electrochemical impedance spectroscopy) Characteristic frequency extraction and analysis are performed. The electrochemical impedance spectroscopy of a lithium battery reflects the response characteristics of different electrochemical processes within the battery, with different electrochemical mechanisms dominating in different frequency ranges. By analyzing the derivative dθ / d(logf) of the phase angle with respect to the logarithm of frequency in the Bode plot, the extreme points and inflection points of the derivative curve are identified. These extreme points and inflection points correspond to the critical frequencies at which the dominance of different electrochemical processes within the battery alternates.

[0103] like Figure 2As shown, based on the physical meaning of the electrochemical process, this invention extracts the following five characteristic frequency points: Low-frequency characteristic frequency point (0.7Hz): Located at the critical position where the phase angle transitions from charge transfer dominance to Warburg diffusion dominance, marking the beginning of solid-phase diffusion process dominating the system response; Mid-low frequency characteristic frequency points (1.6Hz and 2.8Hz): Located at the boundary of the coupling region where solid-phase diffusion and interfacial charge transfer compete with each other, characterizing the alternation of kinetic dominance between lithium ion concentration relaxation and surface Faraday reaction within the active particles; Mid-frequency characteristic frequency point (38.4Hz): Located at the boundary between SEI film relaxation process and charge transfer reaction. Above this frequency, the RC time constant response of the SEI film gradually weakens, and the Faraday process at the electrode / electrolyte interface begins to dominate the impedance characteristics; Mid-high frequency characteristic frequency point (376.6Hz): Located at the boundary between bulk ion conduction and SEI film ion transport. After crossing this frequency, the ion migration resistance of the SEI film separates from the bulk ohmic response and becomes an identifiable independent impedance contribution. Based on the aforementioned characteristic frequencies, SHAP analysis was used to evaluate the contribution of each candidate frequency to impedance reconstruction. Combined with the MAE index, the optimal frequency combination was selected, and finally, five characteristic frequencies of 0.7Hz, 1.6Hz, 2.8Hz, 38.4Hz, and 376.6Hz were determined as the target frequencies of the EK-DIBS excitation sequence.

[0104] Based on the determination of characteristic frequencies, an EK-DIBS excitation sequence is designed. A periodic binary sequence is defined. Calculate the Fourier series of its k-th harmonic: Where T is the sequence period. The amplitude of the sequence is given. Through an iterative optimization algorithm, the signal spectral energy is concentrated on five characteristic frequency points, resulting in an EK-DIBS excitation sequence with the following parameters: transmission frequency of 1 kHz, measurement time of 4.3 seconds, and sequence length of 4287 bytes. Compared to traditional DIBS excitation sequences covering 40 frequency points, the EK-DIBS excitation sequence designed in this invention reduces the sampling frequency from 50 kHz to 2.5 kHz, a reduction of 95%; and compresses the data size from 100 kb to 4.287 kb, a compression ratio of 95.7%.

[0105] Based on the design of characteristic frequency points and EK-DIBS excitation sequences, impedance reconstruction across the entire frequency band is achieved using a SeqKAN network. For example... Figure 1 As shown, the SeqKAN network, as a core component of the overall methodology, incorporates a recursive structure into the KAN network, including a hidden state layer. and output layer Two KAN layers are used, and a conditional fusion strategy is employed to concatenate operating parameters such as temperature and charge / discharge rate into the hidden state, thereby decoupling the temporal features from the conditional features. The specific steps are as follows:

[0106] S3.1: Constructing network input data

[0107] The impedance values ​​at the five characteristic frequencies extracted in step S3 are arranged in ascending order of frequency to form a time-series impedance sequence. The impedance value at each frequency contains the real part. and the virtual part The two components form the input feature tensor. Where B is the batch size representing the number of samples processed simultaneously, T is the time dimension corresponding to 5 feature frequency points, and F is the number of input features at each time step. Simultaneously, the current measured operating condition parameters are used to construct a condition vector. The operating condition parameters include battery ambient temperature, charge / discharge rate, and state of charge (SOC). This represents the number of condition variables.

[0108] S3.2: Conditional Coding and Fusion

[0109] The conditional vector C is input into the conditional encoder for feature encoding to obtain the conditional encoded vector. ,in This is the conditional hiding dimension. The conditional encoding vector is then extended along the time dimension to obtain... This aligns conditional information with temporal feature dimensions. The conditional fusion strategy directly incorporates operating condition variables such as temperature and charge / discharge rate into the hidden state, avoiding redundant conditional modeling. This effectively decouples temporal and conditional features from the network, improving training efficiency and enhancing the model's cross-operating condition adaptability.

[0110] S3.3: Timing Recursive Calculation

[0111] Initialize hidden state For each time step t, from 1 to T, a recursive calculation is performed: first, the input features of the current time step are extracted. The embedding vector is obtained by nonlinear transformation through a KAN layer for feature embedding. Then the embedding vector is compared with the hidden state from the previous time step. After concatenation, the result is input into a time-series KAN layer. And update the hidden state through residual connections. To preserve the original feature information, the updated hidden state is finally concatenated with the expanded conditional information to obtain the fused hidden state. .

[0112] The activation function of each KAN layer in the SeqKAN network is a weighted combination of a learnable spline function and a fixed activation function. in Provides a stable base activation pattern for fixed activation functions. To achieve adaptive nonlinear fitting for learnable B-spline functions, and These are the trainable weight parameters.

[0113] S3.4: Full-band impedance output

[0114] The merged hidden states of all time steps are stacked along the time dimension to form a complete hidden representation. The complete hidden dimension , The time dimension is hidden. This will hide the input / output layer of layer H. The output layer predicts the full-band impedance value based on the timing and conditional features learned from the hidden state layer. The output is the predicted real and imaginary parts of the impedance at 40 target frequency points covering the frequency range of 0.1Hz to 5kHz, forming a complete full-band electrochemical impedance spectrum.

[0115] During network training, dynamic EIS testing was performed on the lithium battery under various ambient temperature conditions (15°C to 35°C) and various charge / discharge rate conditions (0.5C to 2C), with a measurement frequency range of 0.1Hz to 5kHz, encompassing 40 frequency points. The error between the predicted impedance and the true impedance was calculated using the mean square error loss function. The Adam optimizer is used to optimize the network weight parameters through backpropagation until the loss function converges or reaches the preset number of training rounds. The trained SeqKAN network model is then saved for online impedance reconstruction inference.

[0116] Based on the above analysis, this invention, through the synergistic combination of a low-aggregation excitation design based on electrochemical characteristic frequencies and an impedance reconstruction algorithm based on the SeqKAN network, achieves a reduction in detection time from over 10 seconds in traditional methods to 4.3 seconds, a reduction of 57%; the computational cost is only 717.6 kFLOPs, a 98% reduction compared to traditional deep learning methods; and the impedance reconstruction RMSE is less than 0.23 mΩ across the entire frequency band under temperature ranges from 15°C to 35°C and charge / discharge rates from 0.5C to 2C, significantly improving the engineering applicability of dynamic impedance detection.

[0117] Example 2:

[0118] like Figure 3 As shown, the present invention also provides a method for comparing and verifying the impedance spectra of EK-DIBS and traditional DIBS, which is used to verify the effectiveness of the low-polymerization excitation design and SeqKAN network impedance reconstruction method based on electrochemical key frequency points proposed in the present invention.

[0119] The verification experiments were conducted under the same experimental conditions, including four typical operating condition combinations: 15°C-0.5C, 25°C-1C, 25°C-1.5C, and 35°C-2C. For each operating condition, three key SOC points (20%, 50%, and 80%) were selected for comparative testing. The traditional 40-frequency DIBS impedance data from the test set was used as a benchmark to verify the full-band impedance spectrum detection accuracy of the method proposed in this invention. Figure 3 The figure presents a comparison between the full-band impedance spectrum obtained by reconstructing the EK-DIBS excitation sequence designed in this invention using a SeqKAN network and the impedance spectrum directly measured by traditional DIBS under the four operating conditions described above. The horizontal axis in the figure represents the real part of the impedance. The vertical axis represents the imaginary part of the impedance. Impedance spectrum comparison in the form of Nyquist plot.

[0120] The comparative verification results show that:

[0121] (1) Under all test conditions, the reconstructed impedance spectrum of the EK-DIBS of the present invention is highly consistent with the measured impedance spectrum of the traditional DIBS, and the RMSE is controlled within the range of 0.11-0.23mΩ across the entire frequency band;

[0122] (2) The maximum RMSE is 0.23 mΩ, which occurs under low temperature conditions of 15°C-0.5C. The real part error is 0.24 mΩ and the imaginary part error is 0.22 mΩ, mainly caused by the increase in battery internal resistance and the complexity of electrochemical characteristics at low temperature.

[0123] (3) The minimum RMSE is 0.11mΩ, the real part error is 0.11mΩ, and the imaginary part error is 0.10mΩ, which verifies the high-precision impedance reconstruction capability of the method of the present invention under normal temperature conditions.

[0124] (4) At different SOC points (20%, 50%, 80%), the EK-DIBS reconstructed impedance spectrum can accurately track the changing trend of the DIBS measured impedance spectrum, proving the good adaptability of the SeqKAN network to changes in SOC.

[0125] Through the above comparison and verification, the EK-DIBS excitation design proposed in this invention, combined with the SeqKAN network impedance reconstruction method, can still maintain impedance detection accuracy comparable to the traditional DIBS method under the conditions of reducing the detection time from 10 seconds to 4.3 seconds (a reduction of 57%), reducing the sampling frequency from 50kHz to 2.5kHz (a reduction of 95%), and compressing the data volume from 100kb to 4.287kb (a compression of 95.7%). This verifies the engineering application value of this invention in embedded online impedance detection scenarios.

[0126] Example 3:

[0127] Based on a similar inventive concept, this invention also provides a lithium battery full-band impedance high-efficiency extraction system corresponding to the aforementioned lithium battery full-band impedance high-efficiency extraction method, comprising:

[0128] The feature frequency extraction module is used to obtain the electrochemical impedance spectrum Bode plot of lithium battery, and to identify and extract feature frequency points that characterize the transition of different electrochemical processes based on the characteristics of impedance amplitude and phase angle changes with frequency in the Bode plot.

[0129] The excitation sequence generation module is used to construct an excitation sequence based on the feature frequency points extracted by the feature frequency point extraction module using a discrete interval binary sequence method, so that the spectral energy of the excitation sequence is concentrated at the feature frequency points, thereby generating a feature frequency point-guided discrete interval binary sequence excitation sequence.

[0130] The excitation application and signal acquisition module is used to apply the excitation sequence generated by the excitation sequence generation module to the lithium battery, and simultaneously acquire the voltage response signal and current response signal generated by the lithium battery under the excitation action;

[0131] The local impedance calculation module is connected to the excitation application and signal acquisition module. It is used to process the acquired voltage response signal and current response signal, and calculate the impedance value corresponding to the characteristic frequency point through fast Fourier transform.

[0132] The full-band impedance reconstruction module, connected to the local impedance calculation module, is used to receive the impedance value corresponding to the characteristic frequency point, and combine it with the operating condition parameters related to the lithium battery's operating state, inputting them into the pre-trained sequence attention kernel network model to reconstruct and output the full-band electrochemical impedance spectrum of the lithium battery.

[0133] Example 4:

[0134] The present invention also discloses an electronic device.

[0135] Specifically, the electronic device can be a desktop computer, laptop computer, handheld computer, or cloud server, etc. This computer device may include, but is not limited to, a processor and memory. The processor and memory can be connected via a bus or other means. The processor can be a Central Processing Unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, graphics processing units (GPUs), embedded neural network processing units (NPUs) or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0136] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor executes various functional applications and data processing by running non-transitory software programs, instructions, and modules stored in memory. Memory may include a program storage area and a data storage area. The program storage area may store the control unit and the application program required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, memory may include high-speed random access memory and non-transitory memory. In some embodiments, memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0137] Example 5:

[0138] The present invention also discloses a computer-readable storage medium.

[0139] Specifically, the computer-readable storage medium is used to store a computer program, which, when executed by a processor, implements the methods described in the above method implementation.

[0140] Those skilled in the art will understand that all or part of the processes in the methods described above can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

Claims

1. A rapid detection method for lithium batteries across the entire frequency band using electrochemical impedance spectroscopy, characterized in that, Includes the following steps: S1: Obtain the electrochemical impedance spectroscopy Bode plot of the lithium battery. Based on the characteristics of the impedance amplitude and phase angle changing with frequency in the Bode plot, identify and extract the characteristic frequency points that characterize the transition of different electrochemical processes. S2: Based on the characteristic frequency point, an excitation sequence is constructed using the discrete interval binary sequence method, so that the spectral energy of the excitation sequence is concentrated at the characteristic frequency point, thus obtaining the characteristic frequency point guided discrete interval binary sequence EK-DIBS excitation sequence; S3: Apply the EK-DIBS excitation sequence to the lithium battery, collect the voltage response signal and current response signal generated by the lithium battery under the excitation, and extract the impedance value corresponding to the characteristic frequency point through fast Fourier transform; S4: Input the impedance value corresponding to the characteristic frequency point and the operating condition parameters related to the lithium battery's operating state into the pre-trained Sequence Attention Kernel (SeqKAN) network model, and output and reconstruct the full-band electrochemical impedance spectrum of the lithium battery.

2. The method according to claim 1, characterized in that, Step S1 involves identifying and extracting characteristic frequency points that characterize the transitions in different electrochemical processes, including: S1.1: Obtain electrochemical impedance spectroscopy test data of lithium batteries under different states of charge, and plot Bode plots of impedance amplitude |Z| and phase angle θ as a function of frequency f; S1.2: Calculate the derivative dθ / d(logf) of the phase angle θ with respect to the frequency log(f) in the Bode plot, identify the extreme points and inflection points on the derivative curve, and the frequencies corresponding to the extreme points and inflection points are the critical frequencies that characterize the change of dominance of different electrochemical processes inside the battery. S1.3: Based on the electrochemical mechanism, multiple candidate characteristic frequency points are selected from the critical frequency; S1.4: Based on the SHAP interpretability analysis method, analyze and evaluate the contribution of each candidate characteristic frequency point to impedance reconstruction, and select the optimal frequency point combination by combining the mean absolute error index, and determine the optimal frequency point combination as the final characteristic frequency point set.

3. The method according to claim 2, characterized in that, The candidate feature frequency points mentioned in step S1.3 include: The characteristic frequency point in the low-frequency region is located at the critical position where the phase angle transitions from being dominated by charge transfer to being dominated by the diffusion impedance process (Warburg), marking the beginning of the solid-phase diffusion process dominating the system response; The characteristic frequency points in the low-to-mid frequency region are located at the boundary of the coupling region where solid-phase diffusion and interfacial charge transfer compete with each other, characterizing the alternation of kinetic dominance between lithium-ion concentration relaxation and surface Faraday reaction within active particles. The mid-frequency characteristic point is located at the boundary between the relaxation process and charge transfer reaction of the SEI film at the solid electrolyte interface. Above the mid-frequency characteristic point, the RC time constant response of the SEI film gradually weakens, and the Faraday process at the electrode / electrolyte interface begins to dominate the impedance characteristics. The RC time constant reflects the response speed of the SEI film to the AC excitation signal. The mid-to-high frequency region is the boundary between bulk ion conduction and SEI membrane ion transport. After crossing the mid-to-high frequency region, the ion migration resistance of the SEI membrane separates from the bulk ohmic response and becomes an identifiable independent impedance contribution.

4. The method according to claim 1, characterized in that, The method for constructing the EK-DIBS excitation sequence in step S2 includes: S2.1: Define the period as T and the amplitude as ± binary sequence And calculate its Fourier series. ,in For the harmonic order, the Fourier series Cq(k) is expressed as: ;in, The amplitude of the binary sequence is T, the sequence period is j, and the imaginary unit is j. S2.2: Based on the voltage and current sampling noise levels and the preset maximum permissible measurement error, calculate the lower limit of the excitation signal harmonic amplitude that meets the minimum signal-to-noise ratio requirement at the target characteristic frequency. The lower limit of the amplitude The calculation expression is: ;in, The maximum permissible measurement error is... This represents the amplitude of the battery's ohmic internal resistance. and These represent the voltage sampling noise amplitude and the current sampling noise amplitude at the k-th harmonic frequency, respectively. S2.3: Construct the expected signal Its Fourier series representation is and to minimize the binary sequence With the expected signal Frequency domain deviation between To optimize the objective, a constrained optimization model is established: ; S2.4: For the expected signal Discrete sequences are obtained by performing discrete sampling. According to the preset sign function, Quantized into binary sequence The quantization rule is as follows: According to the discrete sequence Calculate the amplitude estimate of the binary sequence. : Where N is the length of the discrete sequence; S2.5: Set the expected signal The initial Fourier series form is ,in The initial phase angle is randomly generated; S2.6: Update the expected signal using an iterative optimization algorithm. The phase angle is adjusted until the convergence condition is met or the preset maximum number of iterations is reached. A single iteration of the iterative optimization algorithm includes: Based on the current phase angle Calculate the Discrete Fourier Transform of the intermediate sequence ,in for The discrete Fourier transform; right The updated discrete sequence is obtained by performing an inverse discrete Fourier transform. ; Will Quantized into binary sequence And calculate its discrete Fourier transform. ; extract phase angle The input phase angle for the next iteration ; S2.7: After the iteration is complete, select the candidate binary sequence from all the candidate binary sequences generated by the iterations that results in the frequency domain deviation. The smallest sequence is determined as the final EK-DIBS excitation sequence; the spectral energy of the final EK-DIBS excitation sequence is concentrated at the characteristic frequency points determined in step S1.

5. The method according to claim 1, characterized in that, The method for extracting the impedance value described in step S3 includes: S3.1: The EK-DIBS excitation sequence is superimposed on the charge / discharge current of the lithium battery as a measurement excitation, and the total terminal voltage signal containing the excitation response component is acquired simultaneously. With total current signal ; S3.2: Convert the total terminal voltage signal With total current signal The components are decomposed into corresponding operating condition components and excitation response components, and their decomposition relationship is expressed as follows: ; ;in, and This represents the total terminal voltage and total current corresponding to the m-th sampling point. and Let represent the m-th voltage and current components generated during normal charging and discharging of the battery, respectively. and These represent the m-th voltage response component and the current response component generated by the measurement excitation, respectively, where m is the sampling point number, ranging from 1 to... , The length of the signal interval used for impedance calculation; S3.3: Based on the total current signal Calculate the estimated value of the current operating condition component. The estimated value is obtained by averaging the total current signal within the signal interval, and the calculation expression is as follows: ;in, Represents the discrete sampled values ​​of the total battery current signal; S3.4: Based on the total terminal voltage signal The estimated values ​​of the voltage operating condition components are obtained by curve fitting. The specific method is as follows: select the moment of stimulus injection as the center, and the time before and after it. The fitting interval consists of sampling points, where Given the data sampling rate, a cubic polynomial fitting is performed on the voltage sampling data points within the fitting interval to obtain the fitting function. The estimated voltage condition component at the m-th sampling point within the impedance measurement interval is: S3.5: Based on the decomposition relationship described in step S3.2 and the estimated values ​​obtained in steps S3.3 and S3.4, calculate the excitation response current respectively. With excitation response voltage : ; S3.6: Response current to the excitation and excitation response voltage Perform a discrete Fourier transform to obtain its frequency... Frequency domain complex representation and : ; Where j is the imaginary unit, For harmonic order; S3.7: According to Ohm's law, calculate the frequency... The complex impedance value Z(fk) at point: Wherein, the frequency Corresponding to the characteristic frequency points determined in step S1.

6. The method according to claim 1, characterized in that, The training method for the SeqKAN network model described in step S4 includes: Training dataset construction: Under various ambient temperature conditions and various charge / discharge rates, dynamic electrochemical impedance spectroscopy tests are performed on lithium batteries to obtain complex impedance spectroscopy data covering the preset full frequency range as real labels; from the complex impedance spectroscopy data, complex impedance values ​​corresponding to the characteristic frequency points determined in step S1 are extracted as model input samples; the input samples are combined with the corresponding real labels and divided into training set and test set according to a preset ratio; Network model training: Set training hyperparameters including batch size, learning rate, number of training epochs, temporal hidden dimension, and conditional hidden dimension; use the training set to perform supervised training on the SeqKAN network model; during training, use mean squared error as the loss function to calculate the error between the model's predicted impedance spectrum and the true impedance spectrum. The loss function expression is: Where N is the length of the discrete sequence; and For the real and imaginary parts of the predicted impedance, and These are the real and imaginary parts of the actual impedance; The weight parameters of the SeqKAN network model are updated using a gradient descent-based optimization algorithm until the loss function converges or reaches a preset number of training rounds, thus obtaining a trained SeqKAN network model.

7. The method according to claim 1, characterized in that, Step S4, which involves outputting and reconstructing the full-band electrochemical impedance spectroscopy of the lithium battery, includes: S4.1: Construct network input data by arranging the impedance values ​​extracted in step S3 from low to high according to the frequency of the corresponding characteristic frequency points to form a time-series impedance sequence, wherein the impedance value of each characteristic frequency point contains the real part. and the virtual part The two components together constitute the input feature tensor. Where B is the batch size, T is the sequence period, and F is the number of input features at each time step; simultaneously, the operating condition parameters of the lithium battery during its current operation are obtained to form a conditional input vector. The operating condition parameters include battery ambient temperature, charge / discharge rate, and state of charge. The number of condition variables; S4.2: Perform conditional encoding and fusion, converting the conditional input vector... The input conditional encoder performs feature encoding to obtain the conditional encoded vector. ,in The conditional hiding dimension is used; the conditional encoding vector is extended along the time dimension to obtain the conditional information. The condition information is aligned with the dimension of the time-series impedance sequence; S4.3: Perform time-series recursive calculations and initialize the hidden state. For a zero vector, perform the following recursive calculation for each time step t from 1 to T: First, extract the input features of the current time step. The embedding vector is obtained by performing a nonlinear transformation through the feature embedding layer. Then the embedding vector Hidden state from the previous moment After concatenation, the result is input into the timing processing layer. And update the hidden state through residual connections. Finally, the updated hidden state is concatenated with the expanded conditional information to obtain the merged hidden state. In the SeqKAN network model, the activation functions of the feature embedding layer and the temporal processing layer... A weighted combination of learnable spline functions and fixed activation functions is used: ,in For a fixed activation function, For learnable B-spline functions, and These are trainable weight parameters; S4.4: Perform full-band impedance output, stacking the fused hidden states of all time steps along the time dimension to form a complete hidden representation. ,in , To hide the time dimension; to represent the hidden dimension Input / output layer The output layer predicts and outputs the real and imaginary parts of the impedance within a preset frequency range, forming a complete full-band electrochemical impedance spectrum.

8. A rapid detection system for full-band electrochemical impedance spectroscopy of lithium batteries based on artificial intelligence, characterized in that, include: The feature frequency extraction module is used to obtain the electrochemical impedance spectrum Bode plot of lithium battery, and to identify and extract feature frequency points that characterize the transition of different electrochemical processes based on the characteristics of impedance amplitude and phase angle changes with frequency in the Bode plot. The excitation sequence generation module is used to construct an excitation sequence based on the feature frequency points extracted by the feature frequency point extraction module using a discrete interval binary sequence method, so that the spectral energy of the excitation sequence is concentrated at the feature frequency points, thereby generating a feature frequency point-guided discrete interval binary sequence excitation sequence. The excitation application and signal acquisition module is used to apply the excitation sequence generated by the excitation sequence generation module to the lithium battery, and simultaneously acquire the voltage response signal and current response signal generated by the lithium battery under the excitation action; The local impedance calculation module is connected to the excitation application and signal acquisition module. It is used to process the acquired voltage response signal and current response signal, and calculate the impedance value corresponding to the characteristic frequency point through fast Fourier transform. The full-band impedance reconstruction module, connected to the local impedance calculation module, is used to receive the impedance value corresponding to the characteristic frequency point, and combine it with the operating condition parameters related to the lithium battery's operating state, inputting them into the pre-trained sequence attention kernel network model to reconstruct and output the full-band electrochemical impedance spectrum of the lithium battery.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the rapid detection method for full-band electrochemical impedance spectroscopy of lithium batteries as described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the rapid detection method for full-band electrochemical impedance spectroscopy of lithium batteries according to any one of claims 1 to 7.

Images