A Method for Online Monitoring of Impedance Spectrum of Mobile Energy Storage Based on Load Current Fingerprinting

Abstract

This invention discloses an online monitoring method for the impedance spectrum of mobile energy storage based on load current fingerprinting, comprising: establishing a load current fingerprint database; performing current fingerprint identification on real-time acquired load current to determine operating conditions; extracting effective excitation components with causal correlation in the load current and voltage response through coherence analysis and constructing the battery impedance spectrum; detecting small-amplitude load fluctuation events and extracting low-frequency impedance characteristics through small-signal time-domain response analysis; employing a module-level isolated sampling architecture to measure the impedance of multiple modules; performing signal-to-noise ratio evaluation, consistency verification, and weighted fusion on the impedance data; extracting health feature parameters based on the weighted fused impedance spectrum data, establishing a health status assessment model, and realizing battery health status assessment and fault early warning. This invention utilizes the load current of the mobile energy storage system during normal operation as a natural excitation source, and achieves online measurement of the battery impedance spectrum through current fingerprinting and spectrum decomposition technology.

Description

Technical Field

[0001] This invention relates to the field of battery health status monitoring technology, and in particular to an online monitoring method for the impedance spectrum of mobile energy storage based on load current fingerprinting. Background Technology

[0002] Mobile energy storage systems, as a new type of energy infrastructure, are widely used in scenarios such as emergency power supply, peak shaving and valley filling, and charging pile replenishment. Mobile energy storage systems typically use lithium iron phosphate or ternary lithium batteries, with capacities ranging from hundreds of kilowatt-hours to several megawatt-hours. Because mobile energy storage systems require frequent charging and discharging and experience different operating conditions and environmental conditions, real-time monitoring of battery health is crucial.

[0003] Electrochemical impedance spectroscopy (EIS) is an effective means of assessing the health status of batteries, reflecting key parameters such as ohmic impedance, charge transfer impedance, and diffusion impedance. Traditional EIS measurement methods require a dedicated frequency response analyzer, applying a small-amplitude sinusoidal excitation signal to the battery to scan a wide frequency band from millihertz to kilohertz. Measurements can take several hours and require the battery to be in a static or lightly loaded state. This offline measurement method cannot meet the online monitoring needs of mobile energy storage systems.

[0004] Existing online EIS measurement technologies mainly suffer from the following problems: 1. Conflict between excitation signal and normal operation: Traditional methods require the injection of a dedicated excitation signal, which interferes with the normal charging and discharging operation of the mobile energy storage system and makes continuous monitoring impossible during actual operation. Some methods achieve excitation by superimposing a small disturbance signal in the controller, but if the disturbance amplitude is too small, the signal-to-noise ratio will be low; if the disturbance amplitude is too large, it will affect the system stability.

[0005] 2. Limited frequency coverage: In order to shorten the measurement time, existing online methods usually only measure part of the frequency band (such as the mid-to-high frequency band above 10Hz), which cannot obtain complete impedance spectrum information, especially lacking low-frequency data that reflects the battery diffusion process, affecting the accuracy of health status assessment.

[0006] 3. Difficulty in measuring low-frequency impedance: The impedance characteristics corresponding to the diffusion process of the battery appear in the extremely low frequency band below 10mHz. Traditional methods require at least 100 seconds of sinusoidal excitation period to measure a frequency point. A complete low-frequency scan takes tens of minutes or even hours, resulting in poor real-time performance.

[0007] 4. Low signal-to-noise ratio: Mobile energy storage systems are subject to various noise interferences during actual operation, including load fluctuations, grid harmonics, and converter switching noise, which result in a low signal-to-noise ratio for impedance measurement. In particular, when the amplitude of the excitation signal is limited, the impedance calculation error increases significantly.

[0008] 5. Complex monitoring architecture for multiple battery modules: Mobile energy storage systems typically consist of multiple battery modules connected in series, with the voltages of each module superimposed to form a high-voltage system. Traditional step-by-step measurement methods suffer from error accumulation, and the processing of high-voltage common-mode signals requires a complex isolation sampling architecture.

[0009] Therefore, there is an urgent need for a technical solution that can use the load current of a mobile energy storage system during normal operation as an excitation source, without the need for an additional excitation signal, to achieve rapid online measurement of the impedance spectrum over a wide frequency band. Summary of the Invention

[0010] The purpose of this invention is to provide an online monitoring method for the impedance spectrum of mobile energy storage based on load current fingerprinting. This method utilizes the load current of the mobile energy storage system during normal operation as a natural excitation source. Through current fingerprinting and spectrum decomposition technology, it achieves online measurement of the battery impedance spectrum, solving the problems of traditional methods that require dedicated excitation signals, have long measurement times, and cannot be monitored online.

[0011] To achieve the above objectives, the present invention provides the following solution: A method for online monitoring of the impedance spectrum of mobile energy storage based on load current fingerprinting includes the following steps: S1, Load current fingerprint database establishment: Collect load current waveforms of mobile energy storage systems in typical application scenarios, perform time-frequency analysis on the current signals, extract current fingerprint features, and establish a load current fingerprint database. S2, Real-time Current Fingerprint Recognition: During the normal operation of the mobile energy storage system, the battery terminal voltage and current signals are collected in real time, and the current load current is fingerprinted to determine the current operating condition type and current spectrum characteristics. S3, Wideband Excitation Component Extraction: Based on the identified current fingerprint type, a coherence analysis method is used to extract the effective excitation components with causal correlation from the load current and the corresponding voltage response, and to remove non-causal noise. S4, Impedance spectrum calculation: Based on the extracted excitation current component and the corresponding voltage response component, Fourier transform is used to calculate the complex impedance at each frequency point to construct the battery impedance spectrum; S5, Low-frequency impedance extension measurement: To address the problem of insufficient excitation energy in the low-frequency band, small-amplitude load fluctuation events are detected. The low-frequency impedance characteristics are extracted using the small-signal time-domain response analysis method and stored separately from the frequency domain measurement results for different health assessment indicators. S6, Module-level Impedance Distributed Measurement: Adopts a module-level isolated sampling architecture, with each battery module configured with an independent isolated sampling front end, and data is aggregated through a digital bus to achieve parallel measurement of the impedance of multiple modules; S7, Impedance Data Quality Assessment and Fusion: The measured impedance data is evaluated for signal-to-noise ratio and checked for consistency. Invalid data points are removed, and multiple measurement data weighted fusion technology is used to improve the accuracy of impedance measurement. S8, Health Status Assessment and Early Warning: Extract health characteristic parameters based on impedance spectrum data, establish a health status assessment model, and realize battery health status assessment and fault early warning.

[0012] Furthermore, in S1, typical application scenarios include emergency power supply scenarios, charging pile replenishment scenarios, peak shaving and valley filling scenarios, and fast charging scenarios. For each scenario, no less than 10 sets of load current waveform data are collected, and the recording time for each set of data is no less than 300 seconds.

[0013] Furthermore, in S1, the current fingerprint features include time-domain features and frequency-domain features; the time-domain features include the mean current, standard deviation current, peak factor, waveform factor, and pulse factor; the frequency-domain features include the dominant frequency component, total harmonic distortion rate, spectral entropy, and frequency band energy distribution ratio.

[0014] Furthermore, in S2, the current fingerprint recognition uses a template matching method based on Euclidean distance or a support vector machine classifier, with a classification accuracy of not less than 90%; when the recognition confidence is less than 0.8, it is determined to be an unknown scenario and a general impedance measurement strategy is adopted.

[0015] Furthermore, in S3, the coherence analysis uses the coherence function method to determine the causal correlation between current and voltage; frequency components with coherence function values ​​greater than a set threshold are determined to be effective excitation components and retained; frequency components with coherence function values ​​lower than the threshold are determined to be non-causal noise and suppressed; the set threshold value ranges from 0.7 to 0.9.

[0016] Furthermore, in S4, the frequency range for impedance spectrum calculation is from 0.1 Hz to 1000 Hz; the frequency resolution is determined by the measurement time window length, and the frequency resolution corresponding to the window length T is 1 / T Hz; the complex impedance at each frequency point is calculated by the ratio of the voltage Fourier transform value to the current Fourier transform value at that frequency.

[0017] Furthermore, in S5, a small-amplitude load fluctuation event is defined as an event in which the current change amplitude is within the range of 1% to 5% of the rated current and the change time is within the range of 0.1 seconds to 1 second; by fitting the time constant of the voltage response curve after the fluctuation, the equivalent low-frequency impedance characteristic parameter is calculated; the low-frequency impedance characteristic parameter is used as an independent health assessment indicator and is not directly fused with the frequency domain EIS data.

[0018] Furthermore, in S6, the module-level isolated sampling architecture includes: each battery module is configured with an isolated sampling front-end module, the isolated sampling front-end module includes an isolated differential amplifier and a high-resolution ADC to measure the terminal voltage of the module; each isolated sampling front-end module is connected to the main controller through an isolated digital interface; the main controller realizes the synchronous sampling of the voltage of each module through a synchronous trigger signal, and the sampling synchronization error does not exceed 1 microsecond.

[0019] Furthermore, in S7, the signal-to-noise ratio evaluation adopts the coherence function method, and frequency points with a coherence function value greater than 0.8 are determined as valid measurement points; the consistency test adopts the three-standard-deviation criterion, and measurement results with a deviation exceeding three standard deviations are determined as abnormal data and removed; the fusion of multiple measurement data adopts the weighted average method, and the weight coefficient is proportional to the coherence function value of the measurement.

[0020] Furthermore, in S8, the health characteristic parameters include ohmic internal resistance, charge transfer impedance, low-frequency impedance characteristic parameters, and impedance magnitudes at multiple characteristic frequency points; the ohmic internal resistance is extracted from the high-frequency real intercept of the impedance spectrum, the charge transfer impedance is extracted from the semicircle diameter of the Nyquist plot, and the low-frequency impedance characteristic parameters are extracted from time-domain response analysis; the health status assessment model uses support vector regression or random forest regression methods to output the battery health status SOH value and remaining lifetime prediction value.

[0021] According to specific embodiments provided by the present invention, the online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting disclosed by the present invention has the following technical effects: (1) No dedicated excitation signal required: The load current of the mobile energy storage system during normal operation is used as a natural excitation source, which does not interfere with the normal operation of the system and realizes true online monitoring. This overcomes the shortcomings of traditional methods that require the injection of disturbance signals or the shutdown of the system for measurement.

[0022] (2) Screening of excitation components based on coherence: The coherence function is used to evaluate the causal correlation between current and voltage, accurately distinguishing effective excitation components and non-causal interference noise, avoiding the problem that traditional filtering methods may mistakenly delete effective excitation sources.

[0023] (3) Separation of frequency domain and time domain features: Frequency domain EIS measurement and time domain response feature extraction are performed separately, without forcibly merging data with different physical meanings, to ensure the physical accuracy of various health assessment indicators.

[0024] (4) Module-level isolated sampling architecture: The independent isolated sampling front end is used to measure the voltage of each module, which solves the problem of high voltage common mode signal measurement and avoids the error accumulation problem of step-by-step calculation method.

[0025] (5) High measurement accuracy: The 24-bit high-resolution ADC, combined with the isolated differential amplifier, provides a voltage measurement resolution better than 10 microvolts, meeting the requirements for milliohm-level impedance measurement; the synchronous sampling error is controlled within 1 microsecond, ensuring the accuracy of phase measurement.

[0026] (6) Intelligent health assessment: By integrating frequency domain impedance characteristics and time domain response characteristics, a machine learning health status assessment model is established to achieve accurate assessment of battery health status and fault warning, ensuring the safe and reliable operation of mobile energy storage system. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is an overall flowchart of the online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting of the present invention; Figure 2 This is a schematic diagram of load current fingerprint feature extraction and scene recognition according to the present invention; Figure 3 This is a schematic diagram of the excitation component extraction based on coherence analysis according to the present invention; Figure 4 This is a schematic diagram of the small-signal time-domain response waveform of the present invention, wherein (a) is a small-amplitude load current fluctuation waveform, and (b) is the corresponding voltage response and fitting curve; Figure 5 This is a diagram of the module-level isolated sampling architecture of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] This invention provides an online monitoring method for the impedance spectrum of mobile energy storage batteries based on load current fingerprint excitation. It utilizes the load current of the mobile energy storage system during normal operation as a natural excitation source, and achieves online measurement of the battery impedance spectrum through current fingerprint recognition and spectrum decomposition technology. This solves the problems of traditional methods, such as the need for dedicated excitation signals, long measurement time, and inability to monitor online.

[0031] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] like Figures 1 to 5 As shown, this invention provides an online monitoring method for the impedance spectrum of mobile energy storage based on load current fingerprinting, specifically including the following steps: Step 1: Establishing the load current fingerprint database The load current waveform of the mobile energy storage system in typical application scenarios is collected, the current signal is analyzed in time and frequency, the current fingerprint features are extracted, and a load current fingerprint database is established.

[0033] The core idea of ​​this invention is to treat the load current of a mobile energy storage system during normal operation as a natural impedance excitation source. Unlike traditional methods that inject dedicated excitation signals, the load current itself contains rich frequency components and can be used as a broadband excitation signal.

[0034] Load currents in different application scenarios have different spectral characteristics, which can be described by "current fingerprints". The purpose of establishing a current fingerprint database is to identify the current operating condition and thus select the optimal excitation component extraction strategy.

[0035] 1.1 Typical Application Scenarios Classification Emergency power supply scenario: Power is supplied to emergency loads, including lighting, communication equipment, and medical equipment. The current waveform is relatively stable, mainly consisting of a fundamental 50Hz component and a small amount of low-order harmonics. Load fluctuations primarily arise from equipment start-up and shutdown, resulting in step-like changes. The current in this scenario is characterized by the 50Hz fundamental component as the primary excitation source, making it suitable for measuring impedance characteristics near 50Hz.

[0036] Charging station replenishment scenario: Powering electric vehicle charging stations, the AC / DC converter inside the charging station generates high-frequency switching noise, with switching frequencies typically ranging from 10kHz to 50kHz. Even after filtering, ripple components of several hundred Hz to several kHz remain. The current in this scenario is characterized by abundant high-frequency components, making it suitable for measuring high-frequency impedance and identifying changes in ohmic internal resistance.

[0037] Peak shaving and valley filling scenario: Charging and discharging are performed according to the grid load curve, and the power dispatch cycle is typically 15 minutes to 1 hour. The current changes slowly and mainly contains extremely low-frequency components, ranging from 0.001 Hz (1 millihertz) to 0.1 Hz. The current in this scenario is characterized by abundant low-frequency components, but the excitation energy may be insufficient.

[0038] Fast charging scenario: The mobile energy storage system itself accepts fast charging, with a large and stable charging current amplitude and low high-frequency ripple generated by the charger. This scenario features a large current change rate (at the start and end of charging), which can be used for small-amplitude step response analysis.

[0039] 1.2 Extraction of Current Fingerprint Features like Figure 2 As shown, the collected load current waveform is first preprocessed: obvious DC bias is removed, abnormal sampling points are eliminated using the 3x standard deviation criterion, and the current amplitude is normalized.

[0040] Temporal feature extraction: The average current reflects the average load level, and the calculation formula is:

[0041] Where N is the number of sampling points, and i(n) is the current value of the nth sampling point.

[0042] The standard deviation of current reflects the degree of current fluctuation:

[0043] in, σ I The standard deviation of the current represents the degree of dispersion of the current series. σ I The larger the value, the more drastic the current fluctuation.

[0044] The peak factor is defined as the ratio of the peak current to the effective current value; the waveform factor is defined as the ratio of the effective current value to the average current value; and the pulse factor is defined as the ratio of the peak current to the average current value. These three dimensionless coefficients together describe the shape characteristics of the current waveform, and different application scenarios result in different combinations of shape characteristics in the current waveform.

[0045] Frequency domain feature extraction: Performing a Fast Fourier Transform (FFT) on the current signal yields the current amplitude spectrum. The dominant frequency component is defined as the frequency point with the largest amplitude other than the DC component. The total harmonic distortion (THD) is defined as the ratio of the root mean square value of all harmonic components to the fundamental frequency component.

[0046] Spectral entropy reflects the degree of concentration or dispersion of spectral energy distribution. First, the spectrum is normalized to a probability distribution, and then the information entropy is calculated. A larger spectral entropy indicates a more uniform energy distribution, meaning the current contains a wider range of frequency components, making it more suitable as a broadband excitation source.

[0047] The frequency band energy distribution divides the entire spectrum into four bands: low frequency (0.1Hz to 1Hz), mid-low frequency (1Hz to 10Hz), mid frequency (10Hz to 100Hz), and high frequency (100Hz to 1000Hz). The percentage of energy in each band relative to the total energy is calculated, forming a four-dimensional vector. This vector directly reflects the excitation capability of the current in each frequency band.

[0048] 1.3 Fingerprint Database Establishment and Update For each application scenario, collect no fewer than 10 sets of typical current waveforms, with each set of data recorded for no less than 300 seconds and a sampling frequency of no less than 10kHz. Extract the aforementioned features from each set of data, calculate the mean and standard deviation of each feature, and form a current fingerprint template for that scenario.

[0049] The fingerprint template storage format is: scene identifier, feature mean vector, feature standard deviation vector, and validity marker. Fingerprint templates for all scenes are stored in the fingerprint database.

[0050] The fingerprint database supports online updates. When new application scenarios are encountered during system operation, new fingerprint templates can be manually labeled and added. The fingerprint database is regularly optimized, removing redundant templates with excessive similarity and merging similar scenarios.

[0051] Step 2: Real-time Current Fingerprint Recognition like Figure 1 and Figure 2 As shown, during the normal operation of the mobile energy storage system, the battery terminal voltage and current signals are collected in real time, and the current load current is fingerprinted to determine the current operating condition type and current spectrum characteristics.

[0052] 2.1 Data Acquisition Configuration Voltage Acquisition: A module-level isolated sampling architecture is employed, with each module configured with an independent isolated sampling front-end. The isolated sampling front-end includes an isolated differential amplifier (such as AMC1311) and a 24-bit high-resolution ADC (such as ADS1262), with a sampling frequency of 10kHz. The voltage measurement range of a single module is 0 to 100V, with a resolution better than 10 microvolts.

[0053] Current acquisition: A closed-loop Hall effect current sensor or a high-precision shunt is used, with an accuracy of no less than 0.1%. The sampling frequency is synchronized with the voltage. The current measurement range is configured according to the system's rated current, with a typical value of 0 to 1000A.

[0054] Time synchronization: All sampling channels are triggered by the same clock source to ensure strict alignment of sampling times, with a time error not exceeding 1 microsecond. The main controller sends sampling trigger pulses to each isolated sampling front-end via a synchronization trigger signal line.

[0055] 2.2 Sliding Window Settings Set up a sliding data window for fingerprint recognition and impedance calculation. The window length is determined based on the target minimum measurement frequency: if the minimum frequency is 0.1Hz, then at least two complete cycles, or 20 seconds, of data are required; considering frequency resolution and measurement accuracy, a window length of 20 to 100 seconds is recommended.

[0056] The window sliding step size determines the impedance update frequency. The step size is set to 50% of the window length, meaning a 20-second window corresponds to a 10-second update cycle. This ensures both the continuity of adjacent measurement results and provides a high update frequency.

[0057] 2.3 Real-time fingerprint calculation For the current data within the current window, calculate the time-domain and frequency-domain features using the same method as when building the database, to form the fingerprint vector of the current current.

[0058] Considering the real-time requirements, an incremental calculation method can be adopted: the mean and standard deviation are updated using recursive formulas; the FFT uses the sliding DFT algorithm, which only requires a small amount of calculation for each new sampling point.

[0059] 2.4 Scene Recognition Methods Method 1: Template Matching. Calculate the distance between the current fingerprint vector and each template in the fingerprint database, and select the scene corresponding to the template with the smallest distance as the recognition result. Mahalanobis distance is used for distance calculation to consider the different dimensions and correlations of various features. A distance threshold is set; if the minimum distance exceeds the threshold, it is determined to be an unknown scene.

[0060] Method 2: Support Vector Machine (SVM) Classification. In the offline phase, a multi-class SVM model is trained using labeled current waveform data, with the radial basis function (RBF) chosen as the kernel function. In the online phase, the current fingerprint vector is input into the model, and the scene category and confidence score are output. A one-to-one voting strategy is employed to handle the multi-class classification problem.

[0061] Recognition result processing: If the recognition confidence is higher than 0.8, the dedicated incentive extraction strategy corresponding to the scenario is adopted; if the confidence is between 0.6 and 0.8, the weighted combination strategy is adopted; if the confidence is lower than 0.6, the general strategy is adopted.

[0062] Step 3: Extraction of wideband excitation components Based on the identified current fingerprint type, a coherence analysis method is used to extract the effective excitation components with causal correlation from the load current and the corresponding voltage response, while eliminating non-causal noise.

[0063] like Figure 3 As shown, load current contains various frequency components, but not all components are suitable as excitation sources for impedance measurement. This invention employs a coherence analysis method to identify effective excitation components from the load current that are causally correlated with the voltage response.

[0064] 3.1 Principle of Coherence Analysis The physical basis of impedance measurement is the causal relationship between current excitation and voltage response: changes in current cause a corresponding voltage response within the battery. However, real-world systems contain various non-causal interference sources: Common-mode noise: Converter switching noise is coupled to both current and voltage measurement channels, manifesting as components of the same frequency in both current and voltage, but not a true "current excitation → voltage response" relationship.

[0065] External interference: Noise introduced into voltage measurement by power grid harmonics, electromagnetic interference, etc., has no causal relationship with current.

[0066] The coherence function is a standard method for evaluating the causal correlation between two signals, and is defined as:

[0067] in, The cross-power spectral density of voltage and current. and These are the self-power spectral densities of voltage and current, respectively.

[0068] The coherence function value is between 0 and 1. A value close to 1 indicates that the voltage change at that frequency is mainly caused by the current passing through a linear system (impedance), and can be used as an effective excitation; a value close to 0 indicates that there is no causal relationship between voltage and current at that frequency, and it cannot be used for impedance calculation.

[0069] 3.2 Coherence-based excitation component selection Calculate the coherence function value for each frequency point and classify them according to the threshold: Effective excitation frequency: The frequency point where the coherence function value is greater than 0.8. The voltage response is mainly generated by current excitation and can be used for accurate impedance calculation.

[0070] Available excitation frequencies: frequencies with coherence function values ​​between 0.5 and 0.8, where the voltage response is generated by current excitation, and the impedance calculation results need to be labeled with confidence levels.

[0071] Invalid frequencies: Frequency points where the coherence function value is less than 0.5, where there is no obvious causal relationship between voltage and current, and the impedance value at this frequency is unreliable and should be eliminated.

[0072] 3.3 Processing of the 50Hz power frequency component In emergency power supply scenarios, 50Hz power frequency current is a major component of the load current. Unlike traditional filtering methods, this invention does not consider 50Hz as "interference," but rather as an important excitation source.

[0073] First, calculate the coherence function value at 50Hz. If the coherence function value is higher than the threshold (e.g., 0.8), it indicates a good causal correlation between the current and voltage responses at 50Hz, and can be used to calculate the impedance at 50Hz. The 50Hz impedance mainly reflects the battery's internal ohmic resistance and high-frequency characteristics.

[0074] If the coherence function value is below the threshold, it indicates significant non-causal interference at 50Hz, making impedance measurements at this frequency unreliable. In this case, impedance values ​​at other frequencies can be referenced, or measurements can be repeated after the operating conditions change.

[0075] 3.4 Voltage Signal Synchronization Processing Perform the same spectral analysis on the voltage signal as on the current signal. The voltage signal needs to be separated into open-circuit voltage and impedance voltage drop. The open-circuit voltage changes slowly and can be extracted using an ultra-low-pass filter with a cutoff frequency set to 0.01Hz. The impedance voltage drop equals the total voltage minus the open-circuit voltage.

[0076] 3.5 Evaluation of Incentive Effectiveness Based on the coherence function screening, the sufficiency of excitation energy in each frequency band is further evaluated. The excitation effectiveness index is defined as follows:

[0077] In the formula, η band As an excitation effectiveness index, it is used to measure whether the current excitation signal in a certain frequency band is strong enough and has sufficient correlation with the voltage response. The higher the value, the more reliable the impedance measurement results in that frequency band. P signal,band This represents the signal power of the current component within that frequency band, reflecting the energy level of the excitation current in that frequency band. P noise,band The noise power of the current measurement channel in this frequency band is usually estimated by measurement under no-excitation or known static conditions. This represents the average coherence function value within this frequency band. Describe the voltage response versus current excitation at frequency f The degree of linear causal correlation on the surface The average value of the square of the coherence function over the entire frequency band is taken, with a range of [0,1]. The closer the value is to 1, the stronger the correlation between the excitation and the response.

[0078] The signal power is the power of the current component in that frequency band, the noise power is estimated by analyzing the background noise of the current measurement channel, and the coherence function is the average coherence function value of that frequency band.

[0079] The validity threshold is set to 5 (approximately 7 dB). Frequency bands below the threshold are marked as insufficient excitation, and the confidence level of impedance measurement results in that frequency band is reduced.

[0080] Step 4: Impedance spectrum calculation Based on the extracted excitation current component and the corresponding voltage response component, the complex impedance at each frequency point is calculated using Fourier transform to construct the battery impedance spectrum.

[0081] 4.1 Fourier Transform Perform Discrete Fourier Transform on the current and voltage time-domain signals within the window. Assume the sampling frequency is... If the window length is N sampling points, then the frequency resolution is .

[0082] For a window length of 20 seconds and a sampling frequency of 10kHz, N=200000, the frequency resolution is 0.05Hz, and the highest resolvable frequency is 5kHz. In actual calculations, for low-frequency bands, downsampling can be performed before FFT to reduce the computational load.

[0083] The time-domain signal is windowed using a Hanning window or a Blackman window to reduce spectral leakage. After windowing, an FFT transformation is performed to obtain the current spectrum I(f) and the voltage spectrum U(f), both of which are complex numbers.

[0084] 4.2 Complex Impedance Calculation frequency f The complex impedance at a given point is defined as the ratio of the voltage spectrum to the current spectrum.

[0085] The impedance magnitude and phase angle are respectively:

[0086]

[0087] The real and imaginary parts of the impedance are respectively:

[0088]

[0089] For battery impedance, the real part is usually positive, while the imaginary part is negative at high frequencies (inductive effect) and negative at low frequencies (capacitive effect).

[0090] 4.3 Frequency Point Selection and Quality Marking Not all frequency points are output. Based on the impedance analysis requirements, select frequency points with a logarithmically uniform distribution. From 0.1Hz to 1000Hz, select 10 to 15 points every 10 octaves, for a total of approximately 40 to 60 frequency points.

[0091] For each output frequency point, a quality label is attached: High confidence level: coherence function greater than 0.8, excitation validity greater than 10, and data reliability.

[0092] Medium confidence level: coherence function 0.5 to 0.8, or stimulus validity 5 to 10. Data can be used as a reference but should be used with caution.

[0093] Low confidence level: coherence function less than 0.5, or excitation validity less than 5, the data is unreliable and for reference only.

[0094] 4.4 Nyquist plot and Bode plot Plot the complex impedance as a Nyquist plot: the horizontal axis represents the real part of the impedance. The vertical axis represents the negative value of the imaginary part of the impedance. A typical battery impedance Nyquist plot exhibits a high-frequency intercept plus a semi-circular arc shape.

[0095] Bode plots include amplitude-frequency and phase-frequency curves: the horizontal axis of the amplitude-frequency curve represents the logarithm of frequency, and the vertical axis represents the impedance magnitude; the vertical axis of the phase-frequency curve represents the phase angle. Bode plots facilitate observation of how impedance changes with frequency.

[0096] Step 5: Low-frequency impedance spread measurement To address the issue of insufficient excitation energy in the low-frequency band, small-amplitude load fluctuation events are detected. The low-frequency impedance characteristics are extracted using a small-signal time-domain response analysis method and stored separately from the frequency domain measurement results for use in different health assessment indicators.

[0097] like Figure 4 As shown, the characteristic frequencies corresponding to the diffusion process of the battery are typically in the range of 0.001Hz to 0.1Hz. In the load current excitation method, the low-frequency range often results in insufficient excitation energy, low coherence function, and inability to obtain reliable EIS data.

[0098] This invention proposes to extract low-frequency impedance features using small-amplitude load fluctuation events. It is important to note that this method extracts time-domain response characteristic parameters, not equivalent frequency-domain EIS data. The two are physically different and are not directly fused; instead, they are used as independent health assessment indicators.

[0099] 5.1 Importance of Small Signal Conditions Electrochemical impedance spectroscopy (EIS) is based on linear small-signal theory, requiring a sufficiently small excitation amplitude to keep the battery operating in the quasi-linear region. For lithium-ion batteries, typical small-signal conditions are current perturbations less than 0.05C or voltage perturbations less than 10mV. If the excitation amplitude is too large (e.g., a large current step), the Butler-Volmer equation of the battery exhibits significant nonlinearity, and the measured value is the DC internal resistance (DCR) rather than the AC impedance. The DCR includes concentration polarization and nonlinear electrochemical reaction effects, and its numerical and physical meanings differ from those of EIS. Therefore, this invention strictly limits the load fluctuation amplitude used for low-frequency feature extraction to ensure operation within the small-signal linear region.

[0100] 5.2 Detection of Small Load Fluctuation Events Real-time monitoring of load current variation characteristics. Low-frequency feature extraction is triggered when a load fluctuation event meeting the following conditions is detected: Current variation range: within 1% to 5% of the rated current. For example, for a system with a rated current of 500A, the variation range is 5A to 25A.

[0101] Change time: 0.1 seconds to 1 second, corresponding to slow load adjustment.

[0102] Type of change: monotonically increasing or monotonically decreasing, non-oscillating change.

[0103] For example, scenarios such as minor load adjustments at charging stations and the activation of lighting equipment can generate small load fluctuations that meet certain conditions.

[0104] 5.3 Time Domain Response Analysis After a qualified load fluctuation event is detected, record the current and voltage waveforms for 30 seconds before and after the fluctuation.

[0105] Determine the start time of fluctuation Calculate the amplitude of current change ,in, The mean steady-state current before the fluctuation. This represents the average steady-state current after fluctuations.

[0106] Extracting voltage response ,in The voltage is the instant before the fluctuation.

[0107] 5.4 Time Constant Fitting The simplified time-domain response of a battery can be described by a multi-exponential decay model. For small current changes... The voltage response is:

[0108] in, The ohmic resistance corresponding to the transient response. Let the resistance be the k-th RC element. , where is the time constant.

[0109] The voltage response curve is fitted using the nonlinear least squares method to identify... , and Parameters. Typically, two RC circuits are sufficient to fit the actual response well.

[0110] 5.5 Definition of Low-Frequency Characteristic Parameters The following feature parameters are extracted from the time-domain response analysis: transient resistance The ratio of the transient component of the voltage step to the change in current mainly reflects the ohmic internal resistance.

[0111] polarization resistor The sum of the resistances of all RC components reflects the polarization characteristics of the battery.

[0112] Master time constant Maximum time constant: Reflects the slowest dynamic process inside the battery.

[0113] Total response resistance : This reflects the quasi-steady-state internal resistance of the battery.

[0114] These parameters serve as independent health assessment indicators to track battery aging trends, rather than being directly integrated with frequency domain EIS data.

[0115] Step 6: Module-level distributed impedance measurement A module-level isolated sampling architecture is adopted, with each battery module configured with an independent isolated sampling front end. Data is aggregated through a digital bus to achieve parallel measurement of impedance of multiple modules.

[0116] like Figure 5 As shown, the mobile energy storage system comprises multiple battery modules connected in series, requiring simultaneous monitoring of the impedance status of each module. This invention employs a module-level isolated sampling architecture to address the issues of high-voltage common-mode signal measurement and error accumulation.

[0117] 6.1 System Topology Analysis Taking the 768V mobile energy storage system as an example, it uses lithium iron phosphate batteries and consists of 8 battery modules connected in series. Each module contains 30 individual batteries connected in series, with a single battery rated voltage of 3.2V and a module rated voltage of 96V.

[0118] Because the modules are connected in series, each module has a different reference potential. The negative terminal of the first module is grounded, and the positive terminal of the eighth module has a potential of 768V. Traditional common-ground measurement architectures cannot directly measure the voltage of each module.

[0119] 6.2 Module-level isolated sampling architecture Each battery module is configured with an independent isolated sampling front-end module, which includes the following components: Isolated differential amplifier: Employs AMC1311 or similar chips, with input terminals directly connected to the positive and negative terminals of the module to measure the module's terminal voltage. It features an isolation withstand voltage of over 1000Vrms and a common-mode rejection ratio greater than 100dB, ensuring that high-voltage common-mode signals do not affect the accuracy of differential measurements.

[0120] High-resolution ADC: Employs a 24-bit sigma-delta ADC (such as the ADS1262), with a range of 0 to 100V and a resolution better than 10 microvolts. The LSB of a 24-bit ADC is approximately 6 microvolts, much smaller than the impedance response signal (typically hundreds of microvolts to several millivolts).

[0121] Digital isolation interface: A digital isolator (such as Si8641) is used to realize data transmission and synchronous trigger signal transmission between the main controller and the isolated sampling front end.

[0122] 6.3 Synchronous Sampling Mechanism The main controller sends sampling trigger pulses to all isolated sampling front-ends via a synchronous trigger signal line. All front-ends simultaneously start ADC conversion upon receiving the trigger pulse.

[0123] The synchronous trigger signal uses differential transmission, which provides strong anti-interference capability. The trigger pulse width is 1 microsecond, and the response delay consistency of each front end is better than 100 nanoseconds. Considering the propagation delay differences of digital isolators (typically 10 to 50 nanoseconds), the overall sampling synchronization error is controlled within 1 microsecond.

[0124] A synchronization error of 1 microsecond corresponds to a phase error of 0.36 degrees for a 1MHz signal. For impedance measurement frequencies below 1kHz, the phase error is negligible.

[0125] 6.4 Current Measurement Since all modules are connected in series, the current flowing through each module is the same. A high-precision current sensor can be installed in the common branch of the battery pack (usually the negative terminal). The current sensor can be a closed-loop Hall sensor or a high-precision shunt, with an accuracy better than 0.1%.

[0126] 6.5 Independent Calculation of Module Impedance For the k-th module, the impedance calculation formula is:

[0127] in, The Fourier transform of the voltage of the k-th module is... For the Fourier transform of the common current.

[0128] The impedance of each module is calculated independently, eliminating the error amplification problem caused by "larger numbers reducing smaller numbers". The impedance measurement error of each module depends only on the voltage measurement accuracy and common current measurement accuracy of that module, and is independent of other modules.

[0129] 6.6 Limitations of In-Module Individual Monitoring The module-level isolation sampling architecture of this invention is primarily used for module-level impedance measurement. For impedance distribution at the individual unit level within a module, it is limited by the following factors: Individual cell voltage sampling rate: If a battery management AFE chip (such as LTC6811) is used for individual cell voltage sampling, due to the daisy-chain communication delay, it takes approximately 10 to 20 milliseconds to complete one round of sampling for 30 cells. According to the Nyquist theorem, it can only support individual cell impedance measurements below 25 to 50 Hz.

[0130] Phase synchronization accuracy: In the polling sampling mode of the AFE chip, the sampling time of each individual unit within the same module has a millisecond-level deviation. For low-frequency measurements (such as below 1Hz), this can be corrected through phase compensation; for high-frequency measurements, excessive phase error leads to inaccurate impedance calculation.

[0131] Therefore, this invention limits the individual-level impedance monitoring to the low-frequency range (below 1 Hz), primarily for identifying consistency differences among individual units within a module. The high-frequency range (above 1 Hz) only provides module-level impedance data.

[0132] Step 7: Impedance data quality assessment and fusion The measured impedance data are evaluated for signal-to-noise ratio and checked for consistency. Invalid data points are removed, and the impedance measurement accuracy is improved by using a weighted fusion technique based on multiple measurement data.

[0133] Due to the randomness of load current and the presence of measurement noise, single impedance measurements may contain errors. It is necessary to assess data quality, remove outliers, and fuse multiple measurement results to improve accuracy.

[0134] 7.1 Coherence Function Evaluation As mentioned earlier, the coherence function is a core indicator for evaluating the quality of impedance measurements. After the impedance calculation is completed, the coherence function value at each frequency point is checked: A value above 0.8 indicates high confidence and reliable data.

[0135] 0.5 to 0.8: Medium confidence level, data is usable but caution is advised.

[0136] Below 0.5: Low confidence level, data is unreliable, flag or remove.

[0137] 7.2 Consistency check of multiple measurements Theoretically, the results of multiple impedance measurements of the same battery module at different times should be consistent (assuming that the battery state does not change significantly).

[0138] For a given frequency point, record the measurement results multiple times. Calculate the mean and standard deviation .

[0139] Anomaly detection is performed using the 3-standard-deviation criterion: if the deviation of a measurement result from the mean exceeds [a certain threshold]... Data that is deemed abnormal will be removed.

[0140] Outlier data may be caused by transient interference, acquisition channel failure, or sudden load changes. After removing outlier data, the mean and standard deviation are recalculated until all retained data are within three standard deviations.

[0141] 7.3 Fusion of Multiple Measurement Data For the valid measurement data after removing outliers, a weighted average method is used for fusion:

[0142] The weighting coefficient is proportional to the coherence function value of that measurement:

[0143] Measurements with high coherence functions have greater weight and contribute more to the fusion result; measurements with low coherence functions have less weight and their influence is weakened.

[0144] Compared to a single measurement, the fused impedance data exhibits reduced random error and improved accuracy. Theoretically, if the noise from m measurements is independent, the error after fusion is reduced to one-third of the original. .

[0145] Step 8: Health Status Assessment and Early Warning Health characteristic parameters are extracted from impedance spectrum data to establish a health status assessment model, thereby enabling battery health status assessment and fault early warning.

[0146] Impedance spectrum and low-frequency characteristic parameters contain rich information about the internal state of the battery. By extracting key characteristic parameters and establishing an evaluation model, a quantitative assessment of the battery's health status can be achieved.

[0147] 8.1 Extraction of Health Feature Parameters Frequency domain impedance characteristics: Ohmic resistance The ohmic internal resistance is read from the high-frequency real intercept of the Nyquist plot or from the high-frequency impedance magnitude of the Bode plot. The ohmic internal resistance is mainly composed of the resistance of the electrode materials, current collector, and electrolyte. As the battery ages, the loss of active material, electrolyte decomposition, and increased contact resistance lead to a gradual increase in ohmic internal resistance.

[0148] Charge transfer impedance The charge transfer impedance is read from the mid-frequency semicircle diameter of the Nyquist plot. Charge transfer impedance reflects the kinetics of electrochemical reactions on the electrode surface. In aged cells, the SEI film thickens and the number of active sites decreases, leading to… Increase.

[0149] Impedance magnitude at characteristic frequencies: Select several characteristic frequencies, such as 0.1Hz, 1Hz, 10Hz, and 100Hz, and read the impedance magnitude at these frequency points. The impedance at characteristic frequency points constitutes a simplified representation of the impedance spectrum.

[0150] Time-domain response characteristics: transient resistance It reflects the ohmic characteristics of the battery and has a similar physical meaning to the frequency domain ohmic internal resistance, so they can be mutually verified.

[0151] polarization resistor This reflects the polarization characteristics of the battery and is related to charge transfer impedance.

[0152] Master time constant This reflects the slowest dynamic process inside the battery and is related to diffusion characteristics.

[0153] 8.2 Health Status Assessment Model State of Health (SOH) is defined as the ratio of current capacity to initial rated capacity, ranging from 0% to 100%.

[0154] A machine learning approach was used to establish a mapping model from impedance characteristics to state of impedance (SOH).

[0155] Training data acquisition: Capacity tests and impedance measurements are performed on batteries at different aging stages to obtain paired data. Capacity tests are conducted under standard operating conditions, and impedance measurements are performed using the method of this invention or an offline frequency response analyzer. The dataset should cover a SOH range from 100% to 60%, with a sample size of no less than 100.

[0156] Feature vector construction: , , , The eigenvectors are composed of impedance magnitudes at multiple characteristic frequency points, with dimensions of approximately 8 to 12. The eigenvectors are then normalized to eliminate the influence of dimensions.

[0157] Model training: Support Vector Regression (SVR) or Random Forest Regression (RFR) are used. SVR employs a radial basis function kernel and selects the optimal hyperparameters through cross-validation. RFR uses 100 to 500 decision trees with a maximum depth of 10 to 20 layers.

[0158] Model Evaluation: The model accuracy is evaluated using the hold-out method or k-fold cross-validation. Evaluation metrics include mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2 The goal is for MAE to be less than 3%, R 2 Greater than 0.9.

[0159] Online application: Input the measured impedance characteristic parameters into the trained model, and output the SOH estimate. It can also output a remaining lifetime prediction, extrapolating the time to SOH=80% based on the SOH change trend.

[0160] 8.3 Fault Early Warning Mechanism Set multi-level early warning thresholds: Health warning: Triggered when SOH drops below 85%, indicating that the battery is starting to age. It is recommended to increase the monitoring frequency.

[0161] Maintenance warning: Triggered when SOH drops below 80%, indicating that the battery is nearing the end of its lifespan and suggesting maintenance or replacement.

[0162] Anomaly Warning: Triggered when impedance characteristics change abnormally, such as a sudden increase in ohmic internal resistance exceeding 20%, a sudden increase in charge transfer impedance exceeding 50%, or an impedance difference of more than 30% between a module and other modules. These anomalies may indicate faults such as internal short circuits, lithium plating, or loose connections.

[0163] Change rate warning: Triggered when SOH decreases rapidly in a short period of time, such as a decrease of more than 5% within one month. Rapid aging may indicate abnormal operating conditions or battery defects.

[0164] The warning information is recorded and uploaded to the monitoring platform to notify the operation and maintenance personnel to handle it. Specific implementation examples: This embodiment provides a specific application of a method for online monitoring of the impedance of a mobile energy storage battery based on load current fingerprint excitation.

[0166] A mobile energy storage system uses lithium iron phosphate batteries, with a system capacity of 500 kWh, a rated voltage of 768 V, and a rated current of 650 A. The battery system consists of 8 battery modules connected in series, with each module containing 30 individual cells connected in series. The module's rated voltage is 96 V. Each individual cell has a rated voltage of 3.2 V and a rated capacity of approximately 217 Ah.

[0167] The system is used for replenishing power to charging piles and providing power support to fast charging stations. The load current fluctuates frequently, ranging from 50A to 600A, including high-frequency ripple generated by the charging pile converter.

[0168] Step 1: Establishing the load current fingerprint database like Figure 2 As shown, typical load current waveforms under charging pile charging scenarios were collected before the system was put into operation. Ten representative sets of data were selected, with each set recorded for 300 seconds and a sampling frequency of 10kHz.

[0169] Fingerprint features were extracted for each group of current waveforms. Taking the first group of data as an example, the time-domain feature calculation results are as follows: mean current 285A, standard deviation 78A, peak factor 1.82, waveform factor 1.14, and pulse factor 2.08.

[0170] A 4096-point FFT analysis was performed on this set of data, with a frequency resolution of approximately 2.44Hz. The dominant frequency component was identified as 200Hz, corresponding to twice the PWM switching frequency of the charging pile. Significant harmonic peaks were observed at 100Hz, 200Hz, and 300Hz.

[0171] The calculated spectral entropy is 4.35, indicating a relatively dispersed spectral distribution containing a rich variety of frequency components. The frequency band energy distribution is as follows: low frequency band (0.1Hz-1Hz) accounts for 3.2%, mid-low frequency band (1Hz-10Hz) accounts for 6.8%, mid frequency band (10Hz-100Hz) accounts for 15.3%, and high frequency band (100Hz-1000Hz) accounts for 74.7%.

[0172] The features of the 10 sets of data are averaged to form a current fingerprint template for the charging pile replenishment scenario, which is then stored in the fingerprint database.

[0173] Step 2: Real-time Current Fingerprint Recognition When the system is running normally, it collects battery voltage and current in real time.

[0174] Voltage acquisition configuration: Each of the eight battery modules is equipped with an isolated sampling front-end module. Each module includes an AMC1311 isolated differential amplifier and an ADS1262 24-bit ADC, with a range of 0 to 100V and a resolution of approximately 6 microvolts. Each module is connected to the main controller via an isolated digital interface. The main controller sends a synchronous trigger signal to achieve synchronous sampling at a sampling frequency of 10kHz.

[0175] Current acquisition configuration: A closed-loop Hall sensor is installed at the negative terminal of the battery pack, with a range of ±800A, an accuracy of 0.1%, and a sampling frequency of 10kHz.

[0176] The sliding data window length is set to 20 seconds, and the sliding step size is 10 seconds, meaning the impedance measurement results are updated every 10 seconds.

[0177] At a certain moment, fingerprint characteristics are calculated for the current data within the window: mean current 312A, standard deviation 92A, peak factor 1.78, main frequency 200Hz, and spectral entropy 4.28.

[0178] The Mahalanobis distance between this feature vector and the template in the fingerprint database is calculated. The distance is 0.12 for the charging pile replenishment scenario, 0.85 for the emergency power supply scenario, and 1.23 for the peak shaving and valley filling scenario.

[0179] The charging pile replenishment scenario has the shortest distance and a confidence level of 0.93, which is greater than the threshold of 0.8. Therefore, the scenario is identified as a charging pile replenishment scenario.

[0180] Step 3: Extraction of wideband excitation components like Figure 3 As shown, the coherence function is calculated for the current and voltage data within the window.

[0181] Taking module 1 as an example, the coherence function between the current and the voltage of module 1 is calculated. The results show: 0.1Hz to 1Hz frequency band: coherence function value 0.45 to 0.65, insufficient excitation energy, marked as low confidence.

[0182] 1Hz to 10Hz frequency band: coherence function value 0.72 to 0.85, medium excitation energy, suitable for impedance measurement.

[0183] 10Hz to 100Hz frequency band: coherence function value 0.88 to 0.95, sufficient excitation energy, high confidence.

[0184] 100Hz to 1000Hz frequency band: coherence function value 0.82 to 0.92, sufficient excitation energy, high confidence.

[0185] At 200Hz (multiplier of the charging pile switching frequency): the coherence function value is 0.91, indicating that the voltage response at this frequency is mainly generated by current excitation, and it can be used as an effective excitation frequency and should not be filtered out.

[0186] At 50Hz: coherence function value 0.78, medium confidence level, can be used for impedance measurement but needs to be marked.

[0187] Based on the coherence function screening results, frequency points with a coherence function greater than 0.5 were retained for impedance calculation, resulting in approximately 45 effective frequency points covering the range of 1Hz to 1000Hz.

[0188] Step 4: Impedance spectrum calculation Perform FFT transformation on the current and voltage of each module within the window. The window length is 20 seconds, the sampling frequency is 10kHz, and the frequency resolution is 0.05Hz.

[0189] Taking module 1 with a frequency of 10Hz as an example, the current Fourier transform amplitude is 8.2A and the phase is 32 degrees; the voltage Fourier transform amplitude of module 1 is 82mV and the phase is 47 degrees. Calculate the impedance: Modulus = 82mV / 8.2A = 10.0mΩ; Phase = 47° - 32° = 15°; Real part = 10.0 × cos(15°) = 9.66 mΩ; Imaginary part = 10.0 × sin(15°) = 2.59 mΩ; Calculate the impedance of all modules and all effective frequency points in sequence.

[0190] For module 1, plot the Nyquist plot. The plot shows typical characteristics: the real intercept in the high-frequency range is about 6.0 mΩ (ohmic internal resistance), and a semi-circular arc with a diameter of about 4.8 mΩ appears in the mid-frequency range (charge transfer).

[0191] The impedance spectrum distributions of the eight modules are similar, and the coefficient of variation of the ohmic internal resistance between modules is about 3.5%, which is within the normal range.

[0192] Step 5: Low-frequency impedance spread measurement like Figure 4 As shown, during the monitoring process, a qualified small-amplitude load fluctuation event was detected: the current slowly decreased from 310A to 295A, with a change range of 15A (accounting for 2.3% of the rated current), and the change time was about 0.8 seconds.

[0193] This event meets the small signal condition (the change amplitude is less than 5% of the rated current) and can be used for low-frequency feature extraction.

[0194] Record the voltage and current waveforms for 30 seconds before and after the fluctuation. Taking module 1 as an example, the steady-state voltage before the fluctuation is 96.15V, and the steady-state voltage after the fluctuation is 96.00V.

[0195] The voltage response curve Δu(t) is extracted and fitted using a double exponential model:

[0196] Fitting results: = 5.8mΩ, = 2.6mΩ, = 1.5s, = 1.8mΩ, = 8s.

[0197] Extracting low-frequency feature parameters: transient resistance = 5.8mΩ; polarization resistor = + = 4.4mΩ; Master time constant = 8s; Total response resistance = 10.2mΩ; These parameters are recorded as independent health assessment indicators and are not directly integrated with frequency domain EIS data.

[0198] transient resistance = 5.8mΩ, which is close to the high-frequency intercept of 6.0mΩ in the frequency domain, and the two can be verified against each other.

[0199] Step 6: Module-level distributed impedance measurement like Figure 5 As shown, the system adopts a module-level isolated sampling architecture. Each of the eight modules measures voltage independently, and sampling timing is aligned through synchronous triggering.

[0200] Synchronization accuracy verification: Test pulses were injected into the synchronization trigger signal line, and the response delay of the sampling front end of each module was measured. The measured difference in response delay between each module was less than 0.5 microseconds, meeting the design requirement of 1 microsecond.

[0201] The impedance of the 8 modules was calculated independently, and the results are as follows: Module 1: = 6.0mΩ, = 4.8mΩ; Module 2: = 5.8mΩ, = 4.6mΩ; Module 3: = 6.2mΩ, = 5.0mΩ; Module 4: = 5.9mΩ, = 4.7mΩ; Module 5: = 6.1mΩ, = 4.9mΩ; Module 6: = 5.7mΩ, = 4.5mΩ; Module 7: = 6.0mΩ, = 4.8mΩ; Module 8: = 6.3mΩ, = 5.1mΩ; The total system impedance is the sum of the impedances of each module: = 47.8mΩ, = 38.4mΩ.

[0202] The coefficient of variation of inter-module ohmic internal resistance is 3.5%, and the coefficient of variation of charge transfer impedance is 4.2%, both of which are within the excellent range (<5%).

[0203] Step 7: Impedance data quality assessment and fusion The quality of the measured impedance data is assessed.

[0204] Coherence function evaluation: In the frequency band from 1Hz to 1000Hz, the coherence function of all frequency points is greater than 0.7, of which 85% of the frequency points are greater than 0.8, indicating good data quality.

[0205] Accumulate 6 measurement data points across multiple measurement windows (once every 10 seconds). Perform a consistency check at each frequency point.

[0206] Taking a 10Hz frequency as an example, the impedance moduli of the six measurements were 10.0, 10.1, 9.9, 10.2, 10.0, and 9.8 mΩ, respectively. The mean was 10.0 mΩ, and the standard deviation was 0.14 mΩ. All data were within three times the standard deviation, with no outliers.

[0207] The six measurement data were weighted and fused. The coherence functions at 10Hz for each measurement were 0.91, 0.88, 0.93, 0.87, 0.90, and 0.89, respectively, with normalized weights of 0.170, 0.164, 0.174, 0.163, 0.168, and 0.166.

[0208] The weighted impedance at 10Hz is: modulus 10.01mΩ. The standard deviation decreased from 0.14mΩ in a single measurement to 0.06mΩ, improving accuracy by approximately 57%.

[0209] Step 8: Health Status Assessment and Early Warning Extracting health feature parameters from the fused impedance data (taking the entire system as an example): Frequency domain characteristics: Ohmic resistance = 47.8mΩ (extracted from the high-frequency intercept); Charge transfer impedance = 38.4mΩ (extracted from the diameter of the semicircle); Impedance magnitude at characteristic frequencies: |Z(1Hz)| = 95mΩ, |Z(10Hz)| = 80mΩ, |Z(100Hz)| = 55mΩ; Time-domain characteristics (averaged from 8 modules): transient resistance = 48.2mΩ; polarization resistor = 35.8mΩ; Master time constant = 7.8s; The frequency domain ohmic internal resistance of 47.8mΩ and the time domain transient resistance of 48.2mΩ differ by only 0.8%, and the two methods can be used to verify each other.

[0210] These feature parameters were normalized and then input into a pre-trained random forest regression model. The model training data came from impedance measurements and capacity tests of 50 battery modules with different aging levels. The model's MAE on the test set was 2.3%, and R... 2 It is 0.92.

[0211] The model outputs a SOH estimate of 93.2% and a remaining lifetime prediction of approximately 720 cycles.

[0212] The SOH value of 93.2% is higher than the health warning threshold of 85%, indicating that the system is in good health.

[0213] Inter-module consistency assessment: The coefficient of variation of ohmic internal resistance is 3.5%, and the coefficient of variation of charge transfer impedance is 4.2%, both of which are excellent.

[0214] The measurement results were recorded in the database and compared with the data from one month ago. The internal resistance increased from 46.5mΩ to 47.8mΩ, an increase of 2.8%; the state of equilibrium (SOH) decreased from 94.5% to 93.2%, a decrease of 1.3%. Both were within the normal aging range and no abnormal warnings were triggered.

[0215] The system runs continuously, updating impedance data every 10 seconds and generating a health status report daily.

[0216] After three months of operation and verification, the SOH assessment results based on impedance data were compared with the quarterly capacity test results, and the error was less than 2.8%, verifying the accuracy of the method.

[0217] During operation, the system successfully issued an early warning for one module anomaly: the internal resistance of module 8 suddenly increased from 6.3mΩ to 8.1mΩ (an increase of 28.6%) within one week, triggering the anomaly warning. Inspection revealed that a loose connector inside module 8 caused the increased contact resistance; after timely tightening, the impedance returned to normal. This early warning prevented potential overheating failures and verified the practical value of the method.

[0218] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for online monitoring of the impedance spectrum of mobile energy storage based on load current fingerprinting, characterized in that, Includes the following steps: S1, Load current fingerprint database establishment: Collect load current waveforms of mobile energy storage systems in typical application scenarios, perform time-frequency analysis on the current signals, extract current fingerprint features, and establish a load current fingerprint database. S2, Real-time Current Fingerprint Recognition: During the normal operation of the mobile energy storage system, the battery terminal voltage and current signals are collected in real time. Based on the load current fingerprint database, the current load current is identified by current fingerprint recognition to determine the current operating condition type and current spectrum characteristics. S3, Wideband Excitation Component Extraction: Based on the identified current fingerprint type, a coherence analysis method is used to extract the effective excitation components with causal correlation from the load current and the corresponding voltage response, and to remove non-causal noise. S4, Impedance spectrum calculation: Based on the extracted excitation current component and the corresponding voltage response component, Fourier transform is used to calculate the complex impedance at each frequency point to construct the battery impedance spectrum; S5, Low-frequency impedance extension measurement: Detects small-amplitude load fluctuation events, extracts low-frequency impedance characteristic parameters through small-signal time-domain response analysis, and stores them separately from the frequency-domain measurement results for use in different health assessment indicators; S6, Module-level Impedance Distributed Measurement: Adopts a module-level isolated sampling architecture, with each battery module configured with an independent isolated sampling front end, and data is aggregated through a digital bus to achieve parallel measurement of the impedance of multiple modules; S7, Impedance Data Quality Assessment and Fusion: The measured impedance data is evaluated for signal-to-noise ratio and checked for consistency. Invalid data points are removed, and multiple measurement data are weighted and fused. S8, Health Status Assessment and Early Warning: Based on the weighted and fused impedance spectrum data, health characteristic parameters are extracted, a health status assessment model is established, and battery health status assessment and fault early warning are realized.

2. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In S1, typical application scenarios include emergency power supply scenarios, charging pile replenishment scenarios, peak shaving and valley filling scenarios, and fast charging scenarios. For each scenario, no less than 10 sets of load current waveform data are collected, and the recording time for each set of data is no less than 300 seconds.

3. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In S1, the current fingerprint features include time-domain features and frequency-domain features; Time-domain characteristics include mean current, standard deviation of current, peak factor, waveform factor, and pulse factor; Frequency domain characteristics include the dominant frequency component, total harmonic distortion rate, spectral entropy, and frequency band energy distribution ratio.

4. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In S2, the current fingerprint recognition uses a template matching method based on Euclidean distance or a support vector machine classifier; when the recognition confidence is lower than 0.8, it is determined to be an unknown scenario and a general impedance measurement strategy is adopted.

5. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In step S3, the coherence analysis method uses the coherence function method to determine the causal correlation between current and voltage; frequency components with coherence function values ​​greater than a set threshold are determined to be effective excitation components and retained; frequency components with coherence function values ​​lower than a set threshold are determined to be non-causal noise and eliminated; the set threshold value ranges from 0.7 to 0.

9.

6. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In step S4, the frequency range for impedance spectrum calculation is from 0.1 Hz to 1000 Hz; the frequency resolution is determined by the measurement time window length, and the frequency resolution corresponding to the window length T is 1 / T Hz; the complex impedance at each frequency point is calculated by the ratio of the voltage Fourier transform value to the current Fourier transform value at that frequency.

7. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In S5, a small-amplitude load fluctuation event is defined as an event in which the current change amplitude is within the range of 1% to 5% of the rated current and the change time is within the range of 0.1 seconds to 1 second. The equivalent low-frequency impedance characteristic parameters are calculated by fitting the time constant of the voltage response curve after fluctuation; these low-frequency impedance characteristic parameters are used as independent health assessment indicators and are not directly fused with frequency domain EIS data.

8. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In S6, the module-level isolated sampling architecture includes: each battery module is configured with an isolated sampling front-end module, the isolated sampling front-end module includes an isolated differential amplifier and a high-resolution ADC to measure the terminal voltage of the battery module; each isolated sampling front-end module is connected to the main controller through an isolated digital interface; the main controller realizes synchronous sampling of the voltage of each battery module through a synchronous trigger signal.

9. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In S7, the signal-to-noise ratio evaluation adopts the coherence function method, and frequency points with a coherence function value greater than 0.8 are determined as valid measurement points; the consistency test adopts the three-standard-deviation criterion, and measurement results with a deviation exceeding three standard deviations are determined as abnormal data and removed; the weighted fusion of multiple measurement data adopts the weighted average method, and the weight coefficient is proportional to the coherence function value of the measurement.

10. The online monitoring method for mobile energy storage impedance spectrum based on load current fingerprinting according to claim 1, characterized in that, In S8, the health characteristic parameters include ohmic internal resistance, charge transfer impedance, low-frequency impedance characteristic parameters, and impedance magnitudes at multiple characteristic frequency points; the ohmic internal resistance is extracted from the high-frequency real intercept of the impedance spectrum, the charge transfer impedance is extracted from the semicircle diameter of the Nyquist plot, and the low-frequency impedance characteristic parameters are extracted from time-domain response analysis. The health status assessment model uses support vector regression or random forest regression to output the battery health status (SOH) value and the predicted remaining life.

Images