An intelligent switching method for BMS balancing strategy for second-use battery packs

Abstract

The present invention discloses a method for intelligent switching of BMS balancing strategies for second-life battery packs, comprising: collecting sensor data from second-life battery packs, performing real-time parameter identification based on a three-branch electrochemical equivalent circuit model after preprocessing and noise filtering to obtain an electrochemical impedance parameter set and a clean data set; using a mutation detection mechanism to monitor the electrochemical impedance parameter set, performing threshold matching judgment in combination with a preconfigured fuzzy partition threshold table, and outputting internal resistance mutation detection results; performing mutation level classification and strategy mapping based on the internal resistance mutation detection results, generating specific execution parameters, and outputting strategy switching instructions. The present invention solves the technical problems of low accuracy in detecting internal resistance mutations in second-life batteries and slow response to balancing strategy switching, realizes intelligent dynamic switching of balancing strategies, and improves the safety and reliability of second-life battery packs.

Description

Technical Field

[0001] The present invention belongs to the field of second-use batteries, and in particular to a BMS balancing strategy intelligent switching method for second-use battery packs. Background Art

[0002] With the rapid development of the electric vehicle industry, a large number of power batteries are about to reach the end of their vehicle life cycle. These retired batteries still retain 70-80% of their initial capacity and have great potential for cascade utilization in energy storage systems. Cascade utilization can not only significantly reduce the cost of energy storage systems and reduce the demand for raw materials, but also effectively extend the value of the battery throughout its life cycle. However, due to the different usage histories and attenuation paths experienced by cascade-reused batteries, their internal electrochemical characteristics show significant heterogeneity and time-varying properties, which poses new challenges to the balancing strategy of the battery management system (BMS). Traditional fixed balancing strategies designed for new batteries are no longer able to adapt to the complex operating conditions of cascade-reused batteries. There is an urgent need to develop intelligent balancing strategy switching technology that can dynamically adjust according to the battery status to ensure the safety, reliability and economy of cascade-reused battery packs.

[0003] Current research on BMS balancing strategies for end-of-life batteries focuses on two main areas: passive balancing methods based on voltage or SOC differences, which consume excess energy from high-capacity batteries through resistive discharge; and active balancing methods based on inductance or capacitance, which enable energy transfer between batteries. In terms of detection, existing research primarily employs periodic impedance measurement techniques, which assess changes in battery internal resistance by injecting sinusoidal signals and analyzing the frequency domain response. The choice of balancing strategy is typically based on simple threshold judgments or neural network-based adaptive algorithms. When cell inconsistencies exceeding a preset threshold are detected, the system switches from passive to active balancing. Some research has also explored multi-objective optimization methods, attempting to find a balance between balancing effectiveness and energy loss. In terms of end-of-life battery characteristic modeling, scholars primarily use equivalent circuit models or electrochemical models to describe battery degradation behavior, and perform SOH estimation and remaining life prediction based on historical data.

[0004] However, there are two deep-seated technical problems in existing technologies when dealing with the special challenges of second-use batteries. The first is the nonlinear distortion problem of frequency-domain impedance analysis: the traditional AC impedance spectrum analysis method is based on the linear system assumption, and the time-domain signal is decomposed into different frequency components for analysis through Fourier transform. However, due to factors such as the uneven growth of the internal SEI film and the local attenuation of the electrode material, the electrochemical response of the second-use battery exhibits strong nonlinear characteristics, resulting in intermodulation interference between different excitation frequencies, which in turn leads to a decrease in the accuracy of internal resistance parameter identification and the inability to capture the instantaneous internal resistance mutation of the second-use battery in a timely manner. Secondly, there is a contradiction between the convergence and responsiveness of the adaptive threshold, which causes the existing adaptive methods to either respond with lag and miss the best switching opportunity, or be too sensitive to produce frequent strategy oscillations, making it impossible to achieve reliable intelligent switching. Summary of the Invention

[0005] The purpose of the invention is to provide a BMS balancing strategy intelligent switching method for cascade utilization battery packs, in order to solve at least one technical problem existing in the prior art.

[0006] The technical solution is an intelligent switching method for BMS balancing strategies for cascade utilization battery packs, including:

[0007] The current sensor data of the battery pack is collected and, after preprocessing and noise filtering, real-time parameter identification is performed based on the three-branch electrochemical equivalent circuit model to obtain the electrochemical impedance parameter set and clean data set;

[0008] The electrochemical impedance parameter set is monitored using a mutation detection mechanism, and threshold matching judgment is performed in combination with a pre-configured fuzzy partition threshold table to output the internal resistance mutation detection result.

[0009] According to the internal resistance mutation detection results, mutation level classification and strategy mapping are performed, specific execution parameters are generated, and strategy switching instructions are output.

[0010] Beneficial effects: The present invention avoids the problem of nonlinear intermodulation interference in the frequency domain decomposition process, solves the technical problems of low accuracy in detecting internal resistance mutations of used batteries and slow response in switching balancing strategies, realizes intelligent dynamic switching of balancing strategies, and improves the safety and reliability of used battery packs. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A flowchart of the steps of a BMS balancing strategy intelligent switching method for a second-use battery pack provided in an embodiment of the present application.

[0012] Figure 2 Flowchart of the steps for obtaining an electrochemical impedance parameter set and a clean data set provided in an embodiment of the present application.

[0013] Figure 3A flowchart of the steps for obtaining an equivalent circuit model provided in an embodiment of the present application.

[0014] Figure 4 A flowchart of the steps for extracting an electrochemical impedance parameter set provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0016] It should be noted that to clearly illustrate the steps of this application, serial numbers are assigned to each step in the specification. These serial numbers are for illustrative purposes only and do not limit the order in which the steps must be executed. In actual operation, depending on the technical requirements of the specific implementation scenario, the steps may be executed in a different order than shown in the specification, and in some cases, parallel processing between steps may be implemented.

[0017] The study found that when test signals of multiple frequencies are injected simultaneously, frequency aliasing and intermodulation component pollution will occur, making it impossible for the traditional FFT decomposition method to accurately separate the frequency responses, which in turn leads to a decrease in the accuracy of internal resistance parameter identification and an inability to timely capture the instantaneous internal resistance mutation of the cascade battery. In addition, the state parameters of the cascade battery are high in dimension and complex in variation, including SOC, SOH, temperature, internal resistance, number of cycles, storage time, and more than 10 other dimensions. The traditional adaptive threshold algorithm needs to find the optimal threshold in the high-dimensional parameter space. To ensure stable convergence of the algorithm, a long learning time and a large amount of historical data are usually required. However, the internal resistance mutation of the cascade battery often occurs in a short period of time (10-30 seconds), requiring the system to have a fast response capability. This dual requirement for long-term stability and short-term responsiveness constitutes a fundamental contradiction.

[0018] like Figure 1 As shown in FIG, a BMS balancing strategy intelligent switching method for cascade utilization battery packs is proposed, including the following steps:

[0019] S1. Collect the current sensor data of the used battery pack, perform real-time parameter identification based on a three-branch electrochemical equivalent circuit model that considers the nonlinear electrochemical characteristics of the used battery after preprocessing and noise filtering, and obtain an electrochemical impedance parameter set and a clean data set;

[0020] S2. Construct a three-dimensional state space based on the battery SOH, temperature change rate, and current rate. Design a fuzzy partition threshold table through the Gaussian membership function. Use a mutation detection mechanism to monitor the electrochemical impedance parameter set. Combined with the pre-configured fuzzy partition threshold table, perform threshold matching judgment and output the internal resistance mutation detection result.

[0021] S3. Perform mutation level classification and strategy mapping based on the internal resistance mutation detection results, use progressive switching to avoid system impact caused by strategy jumps, generate specific execution parameters, and output strategy switching instructions.

[0022] Specifically, sensor data includes each cell's voltage, injected excitation current, battery temperature, and estimated SOC. Based on the electrochemical impedance parameter set and the clean data set, battery state parameters are extracted to construct a three-dimensional physical state space. Electrochemical activity is partitioned, fuzzy boundary processing is designed, and a fuzzy partition threshold table is established.

[0023] like Figure 2 As shown, according to one aspect of the present application, an electrochemical impedance parameter set and a clean data set are obtained, including:

[0024] S11. Collect sensor data from the battery pack during cascade utilization, including the voltage of each single cell, injected excitation current, battery temperature, and SOC estimation; perform data synchronization, timestamp standardization, noise filtering, and outlier detection, and eliminate sensor failure and communication error data to obtain a clean data set;

[0025] S12. Based on the voltage and current data in the clean data set, a three-branch electrochemical equivalent circuit model including a double-layer capacitance branch, a charge transfer impedance branch, and a Warburg impedance branch is constructed to obtain an equivalent circuit model;

[0026] S13. Perform real-time parameter identification on the equivalent circuit model (e.g., using recursive least squares method) to extract the electrochemical impedance parameter set, including double layer impedance, charge transfer impedance, and Warburg impedance parameters.

[0027] like Figure 3 As shown, according to one aspect of the present application, an equivalent circuit model is obtained, including:

[0028] A topological structure of a three-branch parallel equivalent circuit was constructed: the first branch consisted of a double-layer capacitor and a double-layer resistor in parallel, corresponding to a high-frequency electrochemical response; the second branch consisted of a charge transfer resistor and a constant-phase element in parallel, corresponding to a medium-frequency electrochemical response; and the third branch consisted of a Warburg impedance, corresponding to a low-frequency diffusion response. Each branch had a clear electrochemical physical meaning and frequency response characteristics.

[0029] Set the physical boundary conditions of parameters based on the electrochemical characteristics of the cascade battery, including the value ranges of solution resistance, double-layer capacitance resistance, charge transfer resistance, and the impedance magnitude relationship constraints, to ensure the electrochemical rationality of the parameters and obtain the physical constraint parameter matrix;

[0030] Integrate the topological structure and the physical constraint parameter matrix to construct an equivalent circuit model applicable to the nonlinear electrochemical characteristics of the cascade battery.

[0031] Specifically, read the excitation current i(t) and voltage response v(t) data in the clean dataset, and establish a mathematical description of the electrochemical behavior of the cascade utilization battery: v(t)=i(t)Z(ω), where Z(ω) is the complex impedance, to obtain the electrochemical response equation. Construct a three-branch parallel equivalent circuit topological structure: the first branch is the parallel connection of the double-layer capacitance C_dl and the resistance R_dl, the second branch is the parallel connection of the charge transfer resistance R_ct and the constant phase element CPE, and the third branch is the Warburg impedance W. Establish the total impedance expression: Z_total = R_dl / (1 + jωR_dlC_dl)+R_ct / (1+(jωτ)

[0035] )+W / sqrt(jω), to obtain a three-branch equivalent circuit model. Here, j is the imaginary unit, ω is the angular frequency, τ is the time constant, and n is the constant phase element. Set the parameter boundaries based on the electrochemical constraint conditions of the cascade battery: R_dl>0.001Ω, R_ct>R_dl, W>0, C_dl>0.***F, 0.5 < n < 1, and establish the parameter physical constraint matrix: P_constraint=[R_dl_min, R_dl_max; R_ct_min, R_ct_max; W_min, W_max; C_dl_min, C_dl_max; n_min, n_max], to obtain the physical constraint parameter matrix. Here, _min represents the minimum value, and _max represents the maximum value. Integrate the electrochemical response equation, the three-branch equivalent circuit model, and the physical constraint parameter matrix to construct a complete mathematical model applicable to the nonlinear characteristics of the cascade battery and obtain the equivalent circuit model.

[0032] As Figure 4 shown, according to one aspect of the present application, extract the electrochemical impedance parameter set, including:

[0033] Based on the equivalent circuit model and the clean dataset, establish a linear parameter identification equation set and solve it to obtain the initial parameter estimation values;

[0034] Perform physical constraint verification on the initial parameter estimation values, and when violating the boundary conditions in the physical constraint parameter matrix, correct the parameters back to the physically feasible region to ensure the electrochemical rationality of the identification results and obtain the constraint-corrected parameters;

[0035] It should be noted that there seems to be an unclear value "0.***F" in the original text. You may want to check and correct it for a more accurate translation.The change rate of the constraint correction parameters at adjacent moments is calculated, and when the change rate exceeds the continuity threshold, filtering is performed to obtain the electrochemical impedance parameter set.

[0036] Specifically, the equivalent circuit model and clean data set are read, and the complex impedance equation is linearized: the real and imaginary parts are separated by Euler's formula, and a linear parameter identification equation set Aθ=B is established, where θ=[R_dl, R_ct, W, C_dl, n] T is the parameter vector to be identified, T Denotes transposition, A is the parameter coefficient matrix, B is the measurement response vector, and the linearized parameter equation system is obtained. The recursive least squares method with forgetting factor λ is used to solve the linearized parameter equation system: θ(k)=θ(k-1)+K(k)[y(k)-φ T (k)θ(k-1)], where K(k)=P(k-1)φ(k) / [λ+φ T (k)P(k-1)φ(k)] is the gain vector, and the forgetting factor λ=0.95 is used to obtain the real-time parameter estimation value. The real-time parameter estimation value is subjected to physical constraint check: check whether the boundary conditions in the physical constraint parameter matrix are met. If the constraint is violated, the projection algorithm is used to project the parameters back to the feasible domain: θ_projected=argmin||θ-θ_estimated|| 2 Subject to (constraint condition) P_constraint, obtain the constraint correction parameters. Calculate the rate of change of the constraint correction parameters between adjacent moments: Δθ(k) = |θ(k) - θ(k-1) | / θ(k-1). When Δθ(k) > 0.1, a parameter continuity alarm is triggered. A sliding average filter is applied: θ_smooth(k) = αθ(k) + (1-α)θ_smooth(k-1), with α = 0.7. The final output is the electrochemical impedance parameter set containing the R_dl, R_ct, and W parameters of each battery. k is the current time step, y(k) is the measured value (observation data) at the current moment, φ(k) is the regression vector, P(k-1) is the covariance matrix at the previous moment, θ_projected is the parameter vector after constraint correction, P_constraint is the physical constraint parameter matrix, θ_estimated is the parameter estimate without constraint correction, and θ_smooth is the smoothed parameter estimate after sliding average filtering.

[0037] According to one aspect of the present application, the configuration of the fuzzy partition threshold table includes:

[0038] Based on the historical sensor data of the used battery pack, the historical electrochemical impedance parameter set and the clean data set are obtained. Based on this, the SOH, temperature change rate and current rate are calculated to obtain the activity parameter matrix.

[0039] Based on the activity parameter matrix, the electrochemical activity is used to set the boundaries of high, medium, and low activity zones. Different double layer impedance mutation detection thresholds are set for each zone, resulting in a three-dimensional zone boundary and fixed zone threshold table.

[0040] For the boundary area of ​​the three-dimensional partition boundary, the Euclidean distance to the center of each partition is calculated, and the interpolation threshold is obtained by weighted average and smoothed. According to the smoothed result, the fixed partition threshold table is updated to the fuzzy partition threshold table.

[0041] Specifically, three key state parameters (SOH, temperature change rate ΔT, and current rate I_rate) for each battery were extracted from the electrochemical impedance spectroscopy (EI) parameter set and the clean data set to construct a three-dimensional physical state space. Based on electrochemical activity theory, the three-dimensional physical state space was divided into three sub-regions: high activity, medium activity, and low activity. A fixed R_ct mutation detection threshold was set for each sub-region to generate a partition threshold mapping table. A fuzzy membership function was designed for threshold interpolation at the partition boundary to avoid threshold jumps caused by hard switching. The partition threshold mapping table was then updated to a fuzzy partition threshold table.

[0042] For example, the fuzzy partition threshold table adopts a three-dimensional index structure, with SOH, temperature change rate ΔT and current rate I_rate as index dimensions. Specifically, the threshold table data structure is [SOH_index][ΔT_index][I_rate_index], where the SOH_index range is 0-100 corresponding to the SOH value 0%-100%, the ΔT_index range is 0-50 corresponding to the temperature change rate 0-5°C / min, and the I_rate_index range is 0-20 corresponding to the current rate 0-2C. Each index position stores a threshold structure, which contains the detection threshold threshold _value , membership weight membership _weight and boundary smoothing coefficient boundary _smooth For example, when the battery state is SOH=85%, ΔT=1.2°C / min, and I_rate=0.8C, the threshold value is stored at the corresponding index position

[85]

[12] [8]. _value =0.095, membership _weight =0.78, boundary _smooth =0.15.

[0043] According to one aspect of the present application, a fixed partition threshold table is obtained, including:

[0044] Calculate quantitative activity indicators based on the activity parameter matrix, including SOH, temperature change rate and current multiplier;

[0045] A three-dimensional physical state space coordinate system is constructed based on the quantitative activity index. The high activity area is the area where the quantitative activity index is higher than the first threshold, the low activity area is the area where the quantitative activity index is lower than the second threshold, and the medium activity area is the area between the two, thus obtaining the three-dimensional partition boundaries.

[0046] Based on the inverse relationship between electrochemical activity and internal resistance mutation sensitivity, corresponding mutation detection thresholds are set for high-activity areas, low-activity areas, and medium-activity areas within the three-dimensional partition boundaries, and a fixed partition threshold table is obtained.

[0047] Specifically, the R_ct parameter in the electrochemical impedance parameter set and the temperature and current data in the clean data set are read to calculate the electrochemical activity quantitative indicators: battery health status SOH=(R_ct_initial-R_ct_current) / R_ct_initial, temperature change rate ΔT=(T(k)-T(k-10)) / 10, current rate I_rate=|I_current| / I_nominal, and the activity parameter matrix is ​​obtained.

[0048] Among them, SOH is defined as the degree of battery health attenuation, the temperature change rate ΔT reflects the thermodynamic dynamic characteristics, and the current rate I_rate characterizes the intensity of the electrochemical reaction. These three physical quantities jointly determine the electrochemical activity of the cascade battery.

[0049] Based on the activity parameter matrix, a three-dimensional state space coordinate system was constructed. The partition boundaries were set: the high activity region boundary is defined as SOH > 0.8, |ΔT| > 2°C / min, and |I_rate| > 0.5°C; the low activity region boundary is defined as SOH < 0.6, |ΔT| < 0.5°C / min, and |I_rate| < 0.2°C; and the medium activity region is defined as the region between the two. The three-dimensional partition boundaries were obtained. R_ct mutation detection thresholds were set for each partition: Th_high = 0.08 (8% change rate) for the high activity region, Th_mid = 0.12 (12% change rate) for the medium activity region, and Th_low = 0.15 (15% change rate) for the low activity region. A partition-threshold mapping table was established, resulting in a fixed partition threshold table. Here, R_ct_initial is the initial charge transfer resistance of the battery, R_ct_current is the charge transfer resistance at the current moment, T(k) is the temperature at the current moment k, I_current is the current value at the current moment, and I_nominal is the nominal current value.

[0050] According to one aspect of the present application, updating the fixed partition threshold table to a fuzzy partition threshold table includes:

[0051] Based on the fixed partition threshold table and the three-dimensional partition boundaries, the Euclidean distance from the battery status point to the center of each partition is calculated using the Gaussian membership function, and the original interpolation threshold is obtained by weighted average. The difference between the maximum and minimum membership values ​​is calculated based on the original interpolation threshold. When the difference is less than the preset threshold, it is determined to be a fuzzy area and the fuzzy area identifier is obtained.

[0052] For the boundary area in the fuzzy area identification, the Sigmoid function is used to calculate the smoothing weight according to the distance from the state point to the center of the partition boundary to obtain the smoothing weight coefficient;

[0053] The original interpolation threshold is modified based on the smoothing weight coefficient, and the fixed partition threshold table is updated to a fuzzy partition threshold table according to the modified result.

[0054] Specifically, read the fixed partition threshold table and the three-dimensional partition boundary, and calculate the Euclidean distance from the area near the boundary to the center of each partition:

[0055] d_i=sqrt[(SOH-SOH_center_i) 2 +(ΔT-ΔT_center_i) 2 +(I_rate-I_rate_center_i) 2 ], using Gaussian membership function μ_i=exp(-d_i 2 / 2σ 2) calculates the membership, σ=0.1, and obtains the interpolation threshold Th_interpolated=Σ(μ_iTh_i) / Σμ_i through weighted averaging to obtain the partition threshold mapping table. Read the interpolation threshold calculation results in the partition threshold mapping table to identify the boundary fuzzy areas with multimodal phenomena in the membership distribution: when max(μ_i)-min(μ_i)<0.3, it is determined to be a fuzzy area, and the fuzzy area identification is obtained. For the boundary areas in the fuzzy area identification, a smooth transition mechanism is designed using the Sigmoid function: S(x)=1 / (1+exp(-k(x-x0))), where k=10 is the steepness parameter and x0 is the center point of the partition boundary. Calculate the smoothing weight w_smooth=S(distance_to_boundary) to obtain the smoothing weight coefficient. Apply a smoothing weight coefficient to modify the original interpolation threshold: The interpolation threshold is weighted and fused with the average of the adjacent partition thresholds to achieve a smooth transition of the threshold near the partition boundary: Th_final = w_smoothTh_interpolated + (1-w_smooth)Th_boundary_average, where Th_boundary_average is the average of the adjacent partition thresholds to avoid threshold jumps. The partition threshold mapping table is updated to a fuzzy partition threshold table. Where SOH_center_i is the center value of the battery state of health (SOH) for the i-th partition; ΔT_center_i is the center value of the temperature change rate for the i-th partition; I_rate_center_i is the center value of the current rate for the i-th partition; Th_i is the fixed threshold for the i-th partition; S(x) is the Sigmoid function; x is the state variable of the current calculation point; distance_to_boundary is the distance from the current calculation point to the partition boundary; and Th_final is the final modified partition threshold.

[0056] According to one aspect of the present application, outputting the internal resistance mutation detection result includes:

[0057] S21, monitoring the rate of change of the charge transfer impedance in the electrochemical impedance parameter set, obtaining an individualized detection threshold through three-dimensional interpolation in combination with the fuzzy partition threshold table, performing mutation judgment, and obtaining a main channel detection signal;

[0058] S22, calculating the abnormal change of the voltage response slope in the clean data set when the current pulse is injected, performing independent mutation detection, and obtaining the auxiliary channel detection signal;

[0059] S23. Perform differential confirmation analysis on the main channel detection signal and the auxiliary channel detection signal: when a mutation is detected by both channels simultaneously, it is confirmed as a true mutation; when only a single channel is detected, it is determined as measurement interference, thereby obtaining an internal resistance mutation detection result.

[0060] Specifically, the R_ct parameter time series data R_ct(k) from the electrochemical impedance parameter set is read, and the rate of change within the sliding window is calculated: using a sliding window of length N=5, the rate of change sequence ΔR_ct(k)=|R_ct(k)-R_ct(k-1)| / R_ct(k-1) is calculated to obtain the R_ct rate of change sequence. The fuzzy partition threshold table and the state space position coordinates of the current battery are read, and the individual detection threshold Th_individual of the current battery is obtained through three-dimensional interpolation. The trilinear interpolation method is used: Th_individual=TrilinearInterpolation(SOH_current, ΔT_current, I_rate_current, fuzzy partition threshold table) to obtain the individual threshold. The R_ct rate of change sequence and the individual threshold are compared and judged: when ΔR_ct(k)>Th_individual, the main channel detection signal Signal_main=1 is triggered; otherwise, Signal_main=0. The trigger time t_trigger and the amplitude of change amplitude_main are recorded to obtain the main channel detection signal. Where TrilinearInterpolation is the trilinear interpolation function, SOH_current is the battery health status at the current moment, ΔT_current is the temperature change rate at the current moment, and I_rate_current is the current rate at the current moment.

[0061] The main and auxiliary channel detection signals are read, and their respective trigger times t_main and t_aux, as well as their amplitudes amplitude_main and amplitude_aux, are extracted. The timing difference Δt = |t_main - t_aux| and the amplitude difference Δamp = |amplitude_main - amplitude_aux| / max(amplitude_main, amplitude_aux) are calculated to obtain the difference characteristic parameters. A confidence assessment model is established based on these difference characteristic parameters: when Δt < 3 seconds and Δamp < 0.2, the confidence level is 0.9 (high dual-channel consistency); when Δt > 10 seconds or Δamp > 0.5, the confidence level is 0.1 (likely an interference signal). For intermediate cases, linear interpolation is used to calculate the confidence level, resulting in the sudden change confidence level. Final mutation confirmation is performed based on the mutation confidence: the confidence threshold Conf_threshold is set to 0.7. When Confidence > Conf_threshold, a true mutation is confirmed (Detection_result = True); otherwise, it is determined to be measurement interference (Detection_result = False). The mutation battery ID, mutation time, mutation amplitude, and confidence are recorded to obtain the mutation confirmation result. The mutation confirmation results of all batteries are integrated to generate a comprehensive test report containing the mutation battery identification list (Battery_ID_list), mutation amplitude list (Amplitude_list), mutation time list (Time_list), and confidence list (Confidence_list). The internal resistance mutation test results are then output. Where t_main is the trigger time of the main channel detection signal, t_aux is the trigger time of the auxiliary channel detection signal, amplitude_main is the change amplitude of the main channel detection signal, and amplitude_aux is the change amplitude of the auxiliary channel detection signal.

[0062] According to one aspect of the present application, the internal resistance mutation detection result may be output as follows:

[0063] The multi-dimensional local features of the double-layer impedance in the electrochemical impedance parameter set are extracted to construct a battery pack topology association graph that includes physical, electrical, and thermal coupling. Based on the battery pack topology association graph, the state information of neighboring batteries is collected and the association weights are calculated to obtain a local anomaly score and a collection of neighboring state information.

[0064] Based on the neighboring state information set, state propagation is performed to identify and distinguish individual mutation anomalies and systemic interference. Combined with spatiotemporal correlation verification to confirm the abnormal propagation characteristics, the group detection confidence and spatiotemporal correlation verification results are obtained;

[0065] The local anomaly score, group detection confidence and spatiotemporal correlation verification results are integrated to perform persistence and feature consistency tests, as well as historical pattern matching. The final confirmation decision is made in combination with the fuzzy partition threshold table to obtain the internal resistance mutation detection result.

[0066] According to one aspect of the present application, obtaining group detection confidence and spatiotemporal correlation verification results includes:

[0067] Based on the neighboring state information set, the battery pack state vector is established and the state information is propagated. When the global convergence condition is met, the consistency center and state dispersion are calculated to obtain the consistent convergence state.

[0068] Based on the consistency convergence state, the abnormal deviation of individual batteries and the global abnormal index are calculated. When the individual deviation exceeds the threshold and the global abnormal index is normal, it is classified as individual mutation abnormality. When the global abnormal index exceeds the threshold, it is classified as systematic interference. The abnormal type classification result is obtained and the group detection confidence is calculated;

[0069] Based on the anomaly type classification results, the anomaly persistence and intensity change trends are analyzed, the propagation correlation coefficient is calculated, the anomaly propagation mode is identified, the spatiotemporal consistency score is calculated, and the spatiotemporal correlation verification results are obtained.

[0070] According to one aspect of the present application, obtaining an internal resistance mutation detection result includes:

[0071] A Bayesian information fusion framework is used to integrate local anomaly scores, group detection confidence, and spatiotemporal correlation verification results to establish an evidence set and calculate the likelihood probability of detection at each layer. Dynamic weight fusion is used to adaptively adjust the weight coefficient according to the historical accuracy of each detection layer to obtain the fused anomaly probability.

[0072] A multi-level time-series confirmation mechanism is implemented based on the fusion of anomaly probabilities. The first level is a persistence test that requires the anomaly probability to be maintained continuously within the time window. The second level is a feature consistency test that requires the three types of anomaly indicators, electrochemical characteristics, electrical characteristics, and thermodynamic characteristics, to exceed their respective thresholds. The third level is a historical pattern matching that uses dynamic time warping distance to calculate the match degree with historical anomaly patterns to obtain multiple confirmation results.

[0073] Integrate multiple confirmation results. When all three confirmations are passed, output confirmation abnormality and calculate the final confidence level. Otherwise, output rejection abnormality. At the same time, record the abnormal battery identification, abnormality type, detection confidence level and abnormal time to obtain the internal resistance mutation detection result.

[0074] Specifically, a three - layer collaborative detection architecture is adopted to conduct hierarchical intelligent mutation monitoring on the electrochemical impedance parameter set. Through the analysis of the state correlation between batteries and the distributed consensus algorithm, the group confirmation of single - cell anomalies and the intelligent filtering of systematic interference are realized. Combined with the fuzzy partition threshold table for multi - dimensional cross - verification, the internal resistance mutation detection result confirmed by swarm intelligence is output. The specific process is as follows:

[0075] Perform multi - dimensional local feature calculation and anomaly scoring. Read the time series of the R_ct parameter of the target battery in the electrochemical impedance parameter set and the voltage response data in the clean dataset, and calculate the local feature vector including three dimensions of electrochemistry, electricity, and thermodynamics. For the electrochemistry dimension, calculate the change rate of R_ct, ΔR_ct(k)=|R_ct(k)-R_ct(k - 1)| / R_ct(k - 1), and the change in impedance phase angle, Δφ(k)=|arg(Z(k))-arg(Z(k - 1))|; for the electricity dimension, calculate the abnormal change of the voltage response slope dV / dt=(V(t + Δt)-V(t)) / Δt when injecting pulsed current and the current distribution deviation rate; for the thermodynamics dimension, calculate the temperature change rate dT / dt and the thermal gradient distribution index. Use the weighted fusion method to calculate the local anomaly score: Local_anomaly_score = w1·f1(ΔR_ct)+w2·f2(Δφ)+w3·f3(dV / dt)+w4·f4(thermal gradient), where the weight coefficients w1 = 0.4, w2 = 0.2, w3 = 0.3, w4 = 0.1 are optimized and determined according to the characteristics of the hierarchical battery, to obtain the local feature vector and the local anomaly score. Here, arg(Z(k)) is the phase angle of the complex impedance Z(k), Δt is the time interval, and f1, f2, f3, f4 represent the mapping functions of different features.

[0076] Construct the topology association graph of the battery pack and calculate the weights. Read the physical layout information of the battery pack and the state parameters of all batteries in the electrochemical impedance parameter set, and construct a dynamic topology association graph G=(V, E, W). The node set V contains all n batteries in the battery pack, and the edge set E is determined by the triple - coupling criterion: the physical coupling criterion is to establish a connection when the adjacent distance d_ij < d_threshold; the electrical coupling criterion is to establish a connection when the voltage difference |V_i - V_j| < V_threshold and the current difference |I_i - I_j| < I_threshold; the thermal coupling criterion is to establish a connection when the temperature difference |T_i - T_j| < T_threshold. Calculate the edge weight matrix W, where the connection weight w_ij between battery i and j is w_ij = α·exp(-d_ij 2 / 2σ_s 2 )+β·exp(-(V_i - V_j) 2 / 2σ_v 2)+γ·exp(-(T_i-T_j) 2 / 2σ_t 2 ), parameters α=0.5, β=0.3, and γ=0.2 represent the relative importance of spatial, electrical, and thermal coupling, respectively. σ_s=0.5m, σ_v=0.1V, and σ_t=5°C are the corresponding characteristic scale parameters. The battery pack topology association graph and association weight matrix are obtained. Among them, d_threshold is the physical coupling threshold, V_threshold is the electrical coupling threshold, I_threshold is the current coupling threshold, and T_threshold is the thermal coupling threshold.

[0077] Collect the state information of neighboring batteries and calculate the correlation. Based on the battery pack topology association diagram and the association weight matrix, collect the state information of the neighboring battery set N(i)={j|w_ij>w_min} of each battery i. For the neighboring battery j, extract its state vector State_j=[R_ct_j, ΔR_ct_j, SOC_j, T_j, V_j, Local_anomaly_score_j] and confidence Confidence_j. Calculate the state correlation Correlation_ij=Pearson_corr(State_history_i, State_history_j) based on the historical state sequence of the last 50 time steps, and calculate the dynamic similarity Similarity_ij(t)=exp(-||State_i(t)-State_j(t)||2 2 / 2σ_sim 2 ); where σ_sim is adaptively adjusted based on the state dispersion of the battery pack. A neighboring state information set Neighbor_info_i = {(j, State_j, Correlation_ij, Similarity_ij) | j∈N(i)} is established to obtain the neighboring state information set. w_min is the minimum weight threshold, Pearson_corr is the Pearson correlation coefficient, and State_i is the state vector of battery i.

[0078] A distributed consensus algorithm based on state propagation dynamics. Read the local anomaly scores and neighboring state information sets of all batteries and establish the battery pack state vector X(t) = [x1(t), x2(t), ..., x_n(t)] T, where the state vector of each battery x_i(t)=[R_ct_i(t), ΔR_ct_i(t), Local_anomaly_score_i(t), Confidence_i(t)]. The improved distributed average consensus algorithm is used for state information propagation: x_i(t+1)=x_i(t)+ε(t)·∑_{j∈N(i)}w_ij·[x_j(t)-x_i(t)], where the convergence step ε(t)=ε_base·exp(-λ·||X(t)-X(t-1)|| F ) uses an adaptive adjustment strategy, ε_base=0.1 as the base step size, and λ=2 as the convergence factor. When the global convergence condition max_i||x_i(t+1)-x_i(t)||2<Δ_conv=0.01 is met, the iteration is terminated and the consistency center μ_consensus=(1 / n)·∑ i x_i and state dispersion σ_consensus=sqrt((1 / n)·∑ i ||x_i-μ_consensus||2 2 ), and obtain the consistency convergence state and group state statistical characteristics. Where Δ_conv is the convergence threshold of the consensus algorithm.

[0079] Perform multi-scale anomaly pattern recognition and individual location. Based on the statistical characteristics of the consistency convergence state and group state, multi-scale anomaly detection is used to identify individual anomalies and filter out systematic interference. At the individual level, each battery's anomaly deviation is calculated as Deviation_i = ||x_i - μ_consensus||2 / σ_consensus. When Deviation_i > θ_individual = 2.5, it is marked as an individual anomaly candidate. At the group level, the global anomaly indicator Global_anomaly = σ_consensus / σ_historical is calculated, where σ_historical is the group dispersion baseline value under historical normal conditions. When Global_anomaly > θ_global = 1.8, it is determined to be a systematic anomaly (such as a sudden change in ambient temperature or load shock). Anomaly classification logic is used: if Global_anomaly ≤ θ_global and Deviation_i > θ_individual, the anomaly is classified as "individual mutation anomaly" and the outlier battery ID is recorded as Outlier_battery = argmax_iDeviation_i. If Global_anomaly > θ_global, the anomaly is classified as "systematic interference" and all deviating batteries are marked. If neither condition is met, the battery is considered "normal." Group detection confidence is calculated as Group_confidence = min(1.0, Convergence_speed·Neighbor_consistency·Historical_accuracy), where Convergence_speed reflects the convergence speed of the consensus algorithm and Neighbor_consistency reflects the consistency of neighboring battery states. This yields the anomaly classification result and group detection confidence. θ_individual is the individual anomaly detection threshold, θ_global is the global anomaly detection threshold, and Historical_accuracy is the historical detection accuracy.

[0080] Perform spatiotemporal correlation verification and propagation pattern analysis. Read anomaly type classification results and historical detection records to perform spatiotemporal correlation verification to confirm the authenticity and propagation characteristics of the anomaly. Temporal correlation verification uses a sliding window analysis: within the time window T_window = 30 seconds, calculate the anomaly persistence indicator Persistence = (anomaly duration) / (total time window) and the anomaly strength trend Trend = linear_regression_slope (Anomaly_strength_sequence). A persistent anomaly is confirmed when Persistence > 0.6 and |Trend| < 0.1. Spatial correlation verification is based on anomaly propagation path analysis. If an individual anomaly is detected, the state changes of its neighboring batteries within the time series [t-Δt, t+Δt] are analyzed. The propagation correlation coefficient (Propagation_corr) is calculated as max_{j∈N(i)}Cross_correlation(Anomaly_i(t), Anomaly_j(t+τ)), where τ∈[-Δt, +Δt] is the time delay search range. When Propagation_corr > 0.7, the anomaly is confirmed to have spatial propagation characteristics. Combining the battery pack's physical layout and electrical connections, the anomaly propagation modes are identified: thermal propagation (caused by adjacent high-temperature batteries), electrical propagation (due to uneven current distribution), and mechanical propagation (caused by vibration or pressure transfer). The spatiotemporal consistency score (Spatiotemporal_consistency) is calculated as w_t·Temporal_consistency+w_s·Spatial_consistency, where w_t = 0.6 and w_s = 0.4. This results in spatiotemporal correlation verification and anomaly propagation mode identification. Where linear_regression_slope is the linear regression slope, Anomaly_strength_sequence is the anomaly strength time series, Cross_correlation is the cross correlation coefficient, Anomaly_i(t) is the abnormal state of battery i at time t, Temporal_consistency is the temporal consistency score, and Spatial_consistency is the spatial consistency score.

[0081] Perform Bayesian multi-source information fusion and probability calculation. Read three layers of detection information: local anomaly score, group detection confidence, and spatiotemporal correlation verification results. Use the Bayesian information fusion framework to calculate the final anomaly probability. Define the evidence set Evidence = {E1: Local_anomaly_score > θ_local, E2: Anomaly type classification result = "Individual mutation abnormality", E3: Spatiotemporal_consistency > θ_consistency}, where the threshold parameters θ_local = 0.7 and θ_consistency = 0.8 are determined based on the current battery state in the fuzzy partition threshold table. Calculate the likelihood probability for each layer of detection: P(E1|Anomaly) = sigmoid(Local_anomaly_score / θ_local - 1), P(E2|Anomaly) = Group_confidence, and P(E3|Anomaly) = Spatiotemporal_consistency. Set the prior probability P(Anomaly) = Base_rate·Age_factor·Load_factor, where Base_rate = 0.01 is the base anomaly rate, Age_factor and Load_factor reflect the impact of battery aging and load intensity, respectively. Apply Bayesian theorem to calculate the posterior probability: P(Anomaly|Evidence) = P(Anomaly)·∏ i P(E i |Anomaly) / P(Evidence), and uses dynamic weight fusion: Final_anomaly_prob=w_local·P(E1|Anomaly)+w_group·P(E2|Anomaly)+w_spatial·P(E3|Anomaly). The weight coefficients are adaptively adjusted based on the historical accuracy of each detection layer: w_local=Accuracy_local / ∑Accuracy, w_group=Accuracy_group / ∑Accuracy, and w_spatial=Accuracy_spatial / ∑Accuracy. This yields the fused anomaly probability and dynamic weight coefficients. Final_anomaly_prob is the final anomaly probability, and Accuracy_local, Accuracy_group, and Accuracy_spatial are the historical accuracies of local anomaly detection, group detection, and spatiotemporal correlation detection, respectively.

[0082] Perform multiple time series confirmations and false positive filtering mechanisms. Implement multiple time series confirmation mechanisms based on fused anomaly probabilities to reduce false positive detections. The first level of confirmation is a continuity test: set the time window T_confirm = 15 seconds, and require Final_anomaly_prob>θ_primary=0.8 to be maintained continuously within this window. Calculate the continuity index Continuity=(length of time the high probability is maintained) / T_confirm. When Continuity>0.8, the first level of confirmation is passed. The second level of confirmation is a feature consistency test: within the confirmation window, the anomaly indicators of the three types of features, electrochemical features (R_ct changes), electrical features (voltage response), and thermodynamic features (temperature changes), are required to exceed their respective thresholds. Calculate the feature consistency Feature_consistency=(number of abnormal features) / (total number of features). When Feature_consistency>0.67, the second level of confirmation is passed. The third-level confirmation is a historical pattern match: The feature vector of the current anomaly, Current_pattern = [anomaly intensity sequence, propagation path, duration], is extracted and compared with the historical anomaly pattern library using DTW (Dynamic Time Warping) distance calculation. The minimum matching distance, Min_DTW_distance = min_kDTW(Current_pattern, Historical_pattern_k), is obtained. When Min_DTW_distance < θ_pattern = 0.3, the anomaly is confirmed as a known anomaly pattern. Comprehensive confirmation decision: If all three confirmations pass, Final_decision = "Confirmed anomaly" is output, with the final confidence level, Confidence_final = min(Final_anomaly_prob, Continuity, Feature_consistency). Otherwise, Final_decision = "Rejected anomaly" is output, resulting in the multiple confirmation results and final confidence level. θ_primary is the first-level confirmation threshold, Historical_pattern_k is the kth anomaly pattern in the historical anomaly pattern library, θ_pattern is the pattern matching threshold, and Final_decision is the final anomaly decision.

[0083] Perform detection result integration and propagation impact assessment. Read multiple confirmation results, final confidence levels, and anomaly propagation pattern identification to generate a complete swarm intelligence detection report. Summarized detection information includes: anomaly battery ID (Anomaly_battery_ID), anomaly type (Anomaly_type (individual mutation / system interference / normal), detection confidence (Confidence_final), anomaly time (Anomaly_timestamp), and duration (Duration). Analyze the propagation impact: Based on the anomaly propagation pattern identification results, predict the set of neighboring batteries potentially affected by the anomaly: Potential_affected = {j|The propagation path contains battery j and w_ij > w_propagation}, where w_propagation = 0.3 is the propagation threshold. Assess the propagation risk level: Calculate the number of affected batteries: Impact_count = |Potential_affected| and the maximum propagation probability: Max_propagation_prob = max_{j∈Potential_affected}Propagation_probability_j}. Determine the propagation risk level: Risk_level∈{low risk, medium risk, high risk} based on the risk matrix. Generate recommended detection actions: Based on different anomaly types and risk levels, corresponding handling suggestions (Recommended_actions) are output, including options such as enhanced monitoring, preventive maintenance, and emergency isolation. All information is integrated to form the internal resistance mutation detection result = {Anomaly_battery_ID, Anomaly_type, Confidence_final, Anomaly_timestamp, Duration, Potential_affected, Risk_level, Recommended_actions, and detailed detection information for each layer}, which serves as input data for the strategy switching step. Propagation_probability_j is the anomaly propagation probability of battery j.

[0084] According to one aspect of the present application, outputting a policy switching instruction includes:

[0085] S31. Classify the internal resistance mutation according to the mutation amplitude in the internal resistance mutation detection result: the internal resistance mutation is divided into three levels: mild mutation corresponding to a change rate of 8-15%, moderate mutation corresponding to a change rate of 15-25%, and severe mutation corresponding to a change rate greater than 25%, thereby obtaining a mutation level classification;

[0086] S32. Query the preset policy mapping table based on the mutation level classification and the current equalization state of the battery pack. For mild mutations, switch to the active equalization strategy; for moderate mutations, start the fast equalization mode and increase the equalization current; for severe mutations, perform battery isolation and re-plan the equalization topology to obtain the target equalization strategy.

[0087] S33. Generate the specific execution parameters corresponding to the target equalization strategy, including the setting of the equalization current magnitude, the calculation of the equalization duration, and the configuration of the topology reconstruction instruction. Use progressive switching to avoid system shocks caused by strategy jumps and obtain the strategy switching instruction.

[0088] In another embodiment of the present application, the process of obtaining the equivalent circuit model and the electrochemical impedance parameter set is as follows: Read the excitation current i(t) and voltage response v(t) data in the clean dataset, and establish a mathematical description of the electrochemical behavior of the second-life battery: v(t) = i(t)Z(ω); where Z(ω) is the complex impedance, including the frequency-dependent real part and imaginary part: Z(ω) = Z_real(ω) + jZ_imag(ω), to obtain the electrochemical response equation. Construct a three-branch parallel equivalent circuit topology structure, where each branch has clear electrochemical physical meanings and frequency response characteristics: The first branch (double-layer capacitance branch, high-frequency response): The double-layer capacitance C_dl is in parallel with the double-layer resistance R_dl; impedance expression: Z_dl(jω) = R_dl / (1 + jωR_dlC_dl); physical meaning: The charge-discharge process of the double layer at the electrode / electrolyte interface; frequency characteristic: Dominates the high-frequency region response (f > 1 kHz). The second branch (charge transfer branch, medium-frequency response): The charge transfer resistance R_ct is in parallel with the constant phase element CPE; impedance expression: Z_ct(jω) = R_ct / (1 + R_ctY_CPE(jω) n )); physical meaning: The charge transfer dynamics process at the electrode interface; frequency characteristic: Dominates the medium-frequency region response (1 Hz < f < 1 kHz); CPE parameter: Y_CPE is the admittance parameter, and n is the exponent (0.7 ≤ n ≤ 0.9). The third branch (Warburg diffusion branch, low-frequency response): The Warburg impedance element; impedance expression: Z_w(jω) = σ_w / sqrt(jω) = σ_w(1 - j) / (sqrt(2)sqrt(ω)); physical meaning: The semi-infinite diffusion process of lithium ions in the electrode material; frequency characteristic: Dominates the low-frequency region response (f < 1 Hz); Warburg coefficient: σ_w = RT / (n 2 F 2Asqrt(2))(1 / sqrt(D_anode)+1 / sqrt(D_cathode)). Total impedance expression: Z_total(jω)=R_s+Z_dl(jω)+Z_ct(jω)+Z_w(jω)=R_s+R_dl / (1+jωR_dlC_dl)+R_ct / (1+R_ctY_CPE(jω) n )+σ_w / sqrt(jω). R_s is the solution resistance (a frequency-independent constant), resulting in a three-branch equivalent circuit model. Z_real is the real part of the complex impedance, Z_imag is the imaginary part of the complex impedance, R is the resistance, T is the temperature, F is the Faraday constant, A is the electrode surface area, D_anode is the lithium ion diffusion coefficient of the anode material, and D_cathode is the lithium ion diffusion coefficient of the cathode material.

[0089] Based on the electrochemical constraints and physical limitations of the cascade battery, precise parameter boundary conditions are set: Main constraints: P_constraint=[R_s:[0.001Ω, 0.1Ω]%, solution resistance range; R_dl:[0.001Ω, 0.05Ω]%, double layer resistance range; C_dl:[0.1F, 10F]%, double layer capacitance range; R_ct:[0.01Ω, 1.0Ω]%, charge transfer resistance range; Y_CPE:[0.001, 1.0]%, CPE admittance parameter range; n:[0.7, 0.9]%, CPE index range; σ_w:[0.001, 0.1]Ω·s -1 / 2 %, Warburg coefficient range]. Physical relationship constraints: Impedance size relationship: Usually R_s≤R_dl≤R_ct; Time constant relationship: τ_dl=R_dlC_dl<τ_ct=R_ctY_CPEω n-1 ; Temperature dependence: σ_w∝T -1 , R_ct ∝ exp(Ea / RT); SOC dependence: R_ct varies with SOC, and impedance increases at low SOC; the physical constraint parameter matrix is ​​obtained. Where τ_dl is the double layer time constant, τ_ct is the charge transfer time constant, and Ea is the activation energy.

[0090] The electrochemical response equation, three-branch equivalent circuit model and physical constraint parameter matrix are integrated to construct a complete mathematical model suitable for the nonlinear characteristics of cascade batteries. Frequency domain complex form: Z_real(ω)=R_s+R_dl / (1+(ωR_dlC_dl) 2 )+R_ctRe[1 / (1+R_ctY_CPE(jω) n )]+σ_w / sqrt(2ω); Z_imag(ω)=-ωR_dl 2C_dl / (1+(ωR_dlC_dl) 2 )-R_ctIm[1 / (1+R_ctY_CPE(jω) n )]-σ_w / sqrt(2ω); Time-domain differential equation form (for time-domain simulation): The differential equation group of the time-domain equivalent circuit is obtained through the inverse Laplace transform, which is suitable for real-time control algorithms to obtain the equivalent circuit model.

[0091] Establish a linearized parametric equation group. Read the equivalent circuit model and clean data set, separate the real and imaginary parts of the complex impedance equation and perform linearization processing: Frequency domain data processing: For the frequency point set {ω1, ω2, ..., ω n}, extract the corresponding complex impedance measurement value: Z_measured(ω i )=Z_real_measured(ω i )+jZ_imag_measured(ω i ). Construction of linearized equations: Convert the nonlinear complex impedance equation into a linear parameter identification form: Aθ=B; where the parameter vector is: θ=[R_s, R_dl, C_dl, R_ct, Y_CPE, n, σ_w] T ; The coefficient matrix A is constructed in the following way (each frequency point corresponds to two rows, corresponding to the real part and the imaginary part respectively): A_real_i=[1, g_dl_real(ω i ), h_dl_real(ω i ), g_ct_real(ω i ), h_ct_real(ω i ), k_ct_real(ω i ), 1 / sqrt(2ω i )];A_imag_i=[0,g_dl_imag(ω i ), h_dl_imag(ω i ), g_ct_imag(ω i ), h_ct_imag(ω i ), k_ct_imag(ω i ), -1 / sqrt(2ω i)]; the observation vector B is the measured real and imaginary values ​​of the impedance, resulting in a linearized parametric equation system. g_dl_real(ω) and h_dl_real(ω) are the real coefficients of the double layer branch; g_dl_imag(ω) and h_dl_imag(ω) are the imaginary coefficients of the double layer branch; g_ct_real(ω) and h_ct_real(ω) are the real coefficients of the charge transfer branch; g_ct_imag(ω) and h_ct_imag(ω) are the imaginary coefficients of the charge transfer branch; k_ct_real(ω) is the additional real correction coefficient for the charge transfer branch, and k_ct_imag(ω) is the additional imaginary correction coefficient for the charge transfer branch.

[0092] Perform constrained recursive least squares solution. Use the recursive least squares method with physical constraints to solve the linearized parametric equations in real time: Standard RLS update: K(k)=P(k-1)φ(k) / [λ+φ T (k)P(k-1)φ(k)];θ_temp(k)=θ(k-1)+K(k)[y(k)-φ T (k)θ(k-1)];P(k)=[P(k-1)-K(k)φ T (k)P(k-1)] / λ; where the forgetting factor λ is adaptively adjusted according to the dynamic characteristics of the echelon battery: λ(k)=λ_base+Δλexp(-|y(k)–y*(k)| / σ_noise); y*(k) is the predicted output value at the current moment, λ_base=0.95, Δλ=0.04; and the temporary parameter estimate is obtained. θ_temp(k) is the temporary parameter estimate, and σ_noise is the noise standard deviation.

[0093] Perform physical constraint projection correction. Perform physical constraint verification and projection correction on temporary parameter estimates: Constraint verification: Check whether the parameters meet the boundary conditions and physical relationship constraints in the physical constraint parameter matrix; Projection algorithm: For parameters that violate the constraints, use the quadratic programming projection algorithm: θ_projected = argmin||θ-θ_temp|| 2; subject to: P_constraint_lower ≤ θ ≤ P_constraint_upper; h(θ) ≤ 0% physical relationship inequality constraint; g(θ) = 0% physical relationship equality constraint. Constraint function definition: h(θ) = [R_dl - R_ct + ε_tolerance]%, allowing slight violations of the R_ct > R_dl constraint; g(θ) = []%, no equality constraint; obtain constraint correction parameters. Where P_constraint_lower and P_constraint_upper are the upper and lower limits of the physical constraint parameters, h(θ) is the physical relationship inequality constraint, ε_tolerance is the allowable constraint error, and g(θ) is the physical relationship equality constraint.

[0094] Alternatively, the parameter to be corrected, θ_temp, is first checked for boundary constraints. For out-of-bounds parameters, a truncated projection is applied: θ_bounded = max(min(θ_temp, θ_upper), θ_lower). Physical constraints are then checked, primarily including the impedance magnitude relationship R_s ≤ R_dl ≤ R_ct and the time constant relationship τ_dl < τ_ct. If a constraint is violated, a weighted adjustment strategy is employed: parameters with higher confidence levels remain unchanged, while other parameters are adjusted proportionally to meet the constraint conditions. The iteration terminates when all constraints are met or the maximum number of iterations, five, is reached to ensure real-time performance. θ_bounded is the parameter value after boundary constraint verification and truncation, θ_upper is the upper bound of each parameter to be estimated, θ_lower is the lower bound of each parameter to be estimated, and R_s is the series resistance between the battery's internal and external circuits.

[0095] Perform parameter continuity filtering. Calculate the change rate and continuity filtering of the constraint correction parameters at adjacent moments: Change rate calculation: Δθ_rel(k)=||θ(k)-θ(k-1)||2 / ||θ(k-1)||2; Δθ_individual(k)=|θ i (k)-θ i (k-1)| / |θ i(k-1)|, all i; Continuity alarm and filtering: When Δθ_rel(k)>0.15 or max(Δθ_individual(k))>0.2, the parameter continuity alarm is triggered, and adaptive sliding average filtering is used: α_adaptive=min(0.8, 0.3+0.5exp(-Δθ_rel(k) / 0.1)); θ_smooth(k)=α_adaptiveθ(k)+(1-α_adaptive)θ_smooth(k-1). Parameter physical meaning verification: Verify the electrochemical physical rationality of the filtered parameters to ensure that: R_ct can reflect the changes in charge transfer kinetics; σ_w conforms to the diffusion-limited characteristics; each time constant conforms to the electrochemical process hierarchy; the final output contains the electrochemical impedance parameter set containing the complete parameter set of each battery {R_s, R_dl, C_dl, R_ct, Y_CPE, n, σ_w}. Where Δθ_rel(k) is the relative parameter change rate, and Δθ_individual(k) is the change rate of a single parameter.

[0096] In another embodiment of the present application, the process of outputting the internal resistance mutation detection result is specifically: reading the R_ct parameter time series in the electrochemical impedance parameter set, and at the same time collecting the related information of the adjacent batteries: local detection features: Local_features={R_ct_change_rate:ΔR_ct(k)=|R_ct(k)-R_ct(k-1)| / R_ct(k-1); Voltage_response_slope:dV / d, change when the current pulse is injected; Temperature_gradient: temperature difference with the adjacent battery; Current_distribution: current distribution ratio in the parallel branch; Historical_anomaly_pattern: historical mutation pattern matching degree}. Local anomaly score: Local_anomaly_score = w1·R_ct_score + w2·Voltage_score + w3·Thermal_score + w4·Current_score + w5·Pattern_score; the local detection signal is obtained: Signal_local = (Local_anomaly_score, Confidence_local). Where R_ct_change_rate is the rate of change of the charge transfer resistance R_ct, Voltage_response_slope is the voltage response slope, Temperature_gradient is the temperature gradient, R_ct_score, Voltage_score, Thermal_score, Pattern_score, and Current_score are the anomaly scores of the battery's electrochemical, electrical, thermodynamic, historical pattern matching, and current distribution, respectively, and Confidence_local is the confidence level of the local detection.

[0097] Collect information about neighboring batteries. Build a battery pack topology diagram: G = (V, E, W), where V = {Battery_1, Battery_2, ..., Battery_n}: battery node set; E = {(i, j) | physical proximity OR electrical coupling OR thermal coupling}: associated edge set; W = {w_ij}: association weight matrix, association weight calculation: w_ij = α·spatial_weight + β·electrical_weight + γ·thermal_weight; where spatial_weight = exp(-physical_distance_ij / σ_spatial); electrical_weight = |V_i - V_j| -1 ·|I_i-I_j|-1 ; thermal_weight = exp(-|T_i - T_j| / σ_thermal); Collect neighboring battery states: Neighbor_states = {(Battery_j, R_ct_j, Anomaly_score_j, Confidence_j) | j∈N(i)}; Obtain the neighboring state information set. Where Battery_n is the nth battery in the battery pack, spatial_weigh, electrical_weigh, and thermal_weight are the weights of spatial, electrical, and thermal coupling, respectively, physical_distance_ij is the physical distance between batteries i and j, and σ_spatial and σ_thermal are the characteristic scale parameters of spatial and thermal coupling, respectively.

[0098] Collaborative anomaly detection based on distributed consensus theory: Each battery i maintains a state vector: x_i = [R_ct_i, ΔR_ct_i, Local_anomaly_score_i, Confidence_i, Timestamp_i]. Consensus updates are iterated as follows: x_i(t+1) = x_i(t) + ε·∑_{j∈N(i)}w_ij·(x_j(t)-x_i(t)). ε is the convergence step size, which is adaptively adjusted as follows: ε(t) = ε_base·exp(-||x_i(t)-x_i(t-1)||2 / σ_convergence). Convergence is achieved when max_i||x_i(t+1)-x_i(t)||2 < Δ_threshold. The resulting distributed consensus state is: consensus_state = {x_1*, x_2*, ..., x_n*}. Where Timestamp_i is the timestamp of battery i, σ_convergence is the convergence parameter of the consistency algorithm, and Δ_threshold is the convergence threshold of the consistency algorithm.

[0099] Group anomaly detection based on the consensus state: Global anomaly indicator calculation: Global_anomaly_indicator = ∑_iw_i · |x_i* - μ_consensus|; where μ_consensus = (1 / n) · ∑_ix_i* is the mean of the consensus state. Outlier battery location: Outlier_battery = argmax_i |x_i* - μ_consensus|. Anomaly type classification: if Global_anomaly_indicator > θ_global: Anomaly type = "systemic anomaly" (caused by environmental factors); if max_i |x_i* - μ_consensus| > θ_individual: Anomaly type = "individual anomaly" (single battery mutation); else: Anomaly type = "normal state"; Confidence assessment: Group_confidence = f(consistency convergence speed, neighboring battery state similarity, historical detection accuracy); The group detection result: Group_detection = (anomaly type, outlier_battery, group_confidence).

[0100] Multi-scale spatio-temporal consistency test: Temporal consistency test: Short-term consistency (1 - 10 seconds): Test the temporal synchronization of mutation events T_consistency_short = Corr(Anomaly_time_series_i, Anomaly_time_series_j); Medium-term consistency (1 - 10 minutes): Test the consistency of gradual trends T_consistency_medium = DTW_distance(Trend_i, Trend_j), dynamic time warping distance; Long-term consistency (hourly level): Test the matching degree of periodic patterns T_consistency_long = Cross_correlation(Pattern_i, Pattern_j); Spatial consistency test: Local consistency: State correlation of adjacent cells (physical distance < d_local) S_consistency_local = ∑_{j∈Local(i)}w_ij·Similarity(State_i, State_j); Regional consistency: Consistency of the state distribution of cells within the same module S_consistency_regional = KL_divergence(Distribution_module_i, Distribution_expected); Global consistency: State distribution characteristics of the entire battery pack S_consistency_global = Chi_square_test(Global_distribution, Historical_distribution). Comprehensive consistency score: Consistency_score = α·T_consistency + β·S_consistency + γ·Historical_consistency; Obtain the spatio-temporal consistency verification result: Consistency_result = (Consistency_score, Validation_flag).Where Anomaly_time_series is the anomaly time series, DTW_distance is the dynamic time warping distance, Pattern_i is the historical anomaly pattern of battery i, KL_divergence is the Kullback-Leibler divergence, Distribution_module_i and Distribution_expected are the state distribution and expected distribution of the battery module, Chi_square_test is the chi-square test, Global_distribution and Historical_distribution are the state distribution and historical distribution of the current battery pack, α, β, and γ are the weights of temporal, spatial, and historical consistency, T_consistency, S_consistency, and Historical_consistency are the temporal, spatial, and historical consistency scores respectively, and Validation_flag is the consistency verification flag.

[0101] Fusion of three layers of detection information: Local detection layer: Signal_local = (Local_anomaly_score, Confidence_local); Collaborative detection layer: Group_detection = (Anomaly type, Outlier_battery, Group_confidence); Consistency verification layer: Consistency_result = (Consistency_score, Validation_flag). Bayesian information fusion: P(Anomaly|All_Evidence) = P(Anomaly)·∏_kP(Evidence_k|Anomaly) / P(All_Evidence). Where Evidence_1 = Local_anomaly_score > θ_local; Evidence_2 = Group_detection. Anomaly type = "Individual anomaly"; Evidence_3 = Consistency_score < θ_consistency. The fusion weights are adaptively adjusted: w_local = f (local sensor quality, historical detection accuracy); w_group = f (number of neighboring cells, network connectivity); w_consistency = f (data integrity, time series length); and the final anomaly probability is calculated as: P_final = w_local · P_local + w_group · P_group + w_consistency · P_consistency, resulting in the fused detection probability. P_local, P_group, and P_consistency are the anomaly probabilities for local detection, group detection, and consistency verification, respectively.

[0102] Build a triple confirmation mechanism to avoid false alarms: First confirmation: continuity test within the time window ifP_final>θ_primaryforcontinuousT_window:trigger_secondary_confirmation(); Second confirmation: cross-validation of multi-dimensional features Cross_validation_features=[electrochemical characteristics: consistency of R_ct, C_dl, σ_w changes; Electrical characteristics: consistency of voltage and current response anomalies; Thermodynamic characteristics: consistency of temperature distribution and heat dissipation characteristics changes; Mechanical characteristics: internal pressure and volume changes (if measurable)]; Feature_consistency=∑_fw_f·Consistency_check(f); Third confirmation: historical pattern matching Historical_pattern_match=DTW_distance(Current_pattern, Historical_anomaly_patterns). Comprehensive confirmation result: if (P_final > θ_primary) AND (Feature_consistency > θ_feature) AND (Historical_pattern_match < θ_pattern): Final_decision = "Confirm the anomaly"; Confidence_final = min(P_final, Feature_consistency, 1 - Historical_pattern_match); else: Final_decision = "Reject the anomaly"; obtain the final confirmation result. Here, trigger_secondary_confirmation triggers the secondary confirmation mechanism, consistency_check is the consistency check function, and historical_anomaly_patterns is the historical anomaly pattern library.

[0103] Integrate all detection information to generate a comprehensive report: Detection_report = {Battery ID: Battery_ID, Detection result: Final_decision, Confidence: Confidence_final, Anomaly type: Individual anomaly / Systematic anomaly / Normal, Detection level result: {Local detection: (Local_anomaly_score, Confidence_local), Collaborative detection: (Group_detection, Group_confidence), Consistency verification: (Consistency_score, Validation_flag)}, Spatiotemporal features: {Anomaly time: Anomaly_timestamp, Duration: Duration, Affected range: Affected_batteries, Propagation pattern: Propagation_pattern}, Recommended action: Recommended_action}; Output the internal resistance mutation detection result.

[0104] The mutation level classification adds a time duration correction based on the basic change rate judgment. When the duration of the mutation exceeds the preset threshold, the system automatically increases the response level: if the duration is 10-30 seconds, the original level is maintained; if the duration is 30-60 seconds, the response speed is increased by 50%; if the duration exceeds 60 seconds, the level is automatically increased by one level. It is suitable for gradual anomalies of cascade batteries and can capture slowly developing failure modes in a timely manner. At the same time, a comprehensive judgment is made based on the detection confidence: when Confidence_final>0.9, a high-level response is allowed to be triggered in advance; when Confidence_final<0.6, the response is delayed and a verification step is added.

[0105] In a specific embodiment of the present application, an energy storage module is based on eight series-connected ternary lithium batteries used in cascade utilization. The batteries are numbered B1-B8, with a rated capacity of 50Ah and a nominal voltage of 3.7V. The system operates under energy storage conditions, with a charge and discharge rate of 0.5C and an ambient temperature of 28°C. The process of the BMS balancing strategy intelligent switching method for cascade utilization battery packs is as follows:

[0106] Step 1: Data collection and parameter identification.

[0107] 1.1. Raw data collection.

[0108] At t=1000s, the collected raw sensor data are: battery voltage: V1=3.82V, V2=3.85V, V3=3.79V, V4=3.84V, V5=3.81V, V6=3.83V, V7=3.80V, V8=3.86V; excitation test current: i(t)=0.1sin(2π×1000t)+0.05sin(2π×100t)+0.02sin(2π×10t)A; battery temperature: T1= 28.2°C, T2=29.1°C, T3=31.5°C, T4=28.8°C, T5=29.3°C, T6=28.7°C, T7=29.8°C, T8=28.5°C; SOC valuation: SOC1=87%, SOC2=89%, SOC3=85%, SOC4=88%, SOC5=86%, SOC6=89%, SOC7=87%, SOC8=90%; After data synchronization processing, the synchronized original data matrix is ​​obtained.

[0109] 1.2. Data cleaning processing.

[0110] Abnormally high temperature was detected for battery B3, and outlier processing was performed: anomaly detection threshold: T_threshold = 30°C; B3 temperature 31.5°C > 30°C, marked as a potential anomaly; verification using the 3σ rule: T_mean = 29.1°C, T_std = 0.9°C; anomaly determination: |31.5-29.1| = 2.4°C < 3×0.9 = 2.7°C, determined as normal fluctuation; all data were retained after cleaning to form a clean data set.

[0111] 1.3. Construction of three-branch equivalent circuit model.

[0112] Taking battery B3 as an example, based on its voltage response v3(t) and excitation current i(t): the complex impedance model of battery B3 is: Z3_total(jω)=R_s3+R_dl3 / (1+jωR_dl3C_dl3)+R_ct3 / (1+R_ct3Y_CPE3(jω) n3 )+σ_w3 / sqrt(jω); parameter physical constraint range: R_s3: [0.001Ω, 0.1Ω]; R_dl3: [0.001Ω, 0.05Ω]; C_dl3: [0.1F, 10F]; R_ct3: [0.01Ω, 1.0Ω]; σ_w3: [0.001, 0.1]Ω·s -1 / 2 ; Get the equivalent circuit model.

[0113] 1.4. Recursive least squares parameter identification.

[0114] Real-time parameter identification of battery B3: Initial parameter estimation: θ3(999)=[0.008, 0.012, 1.5, 0.045, 0.25, 0.85, 0.035] T Parameter update at t=1000s: Forgetting factor: λ=0.95; Measurement vector: y(1000)=3.79V; Regression vector: φ T (1000)=[1, 0.85, 0.67, 0.92, 0.78, 0.88, 0.71]; Gain vector calculation: K(1000)=P(999)φ(1000) / [λ+φ T (1000)P(999)φ(1000)]=P(999)φ(1000) / [0.95+2.34]=P(999)φ(1000) / 3.29; parameter update: θ3(1000)=θ3(999)+K(1000)[y(1000)-φ T (1000)θ3(999)]=[0.008, 0.012, 1.5, 0.045, 0.25, 0.85, 0.035] T +K(1000)×0.021. Updated parameters (after constraint correction): R_s3=0.008Ω; R_dl3=0.012Ω; C_dl3=1.48F; R_ct3=0.047Ω; Y_CPE3=0.26; n3=0.84; σ_w3=0.036Ω·s -1 / 2 ; Repeat the above process for all 8 batteries to obtain the electrochemical impedance parameter set.

[0115] Step 2: Construction of three-dimensional state space and fuzzy partitioning.

[0116] 2.1. Calculation of activity parameters.

[0117] Taking battery B3 as an example, the key state parameters are calculated: SOH3 calculation: SOH3=(R_ct_initial-R_ct_current) / R_ct_initial=(0.040-0.047) / 0.040=-0.175, corrected to SOH3=1-|0.175|=0.825=82.5%; temperature change rate calculation: ΔT3=(T3(1000s)-T3(990s)) / 10s=(31.5-30.8) / 10=0.07°C / s; current rate calculation: I_rate3=|I_current| / I_nominal=|25A| / |50A|=0.5C; activity parameter matrix B3=[SOH3=0.825, ΔT3=0.07°C / s, I_rate3=0.5C].

[0118] 2.2. Three-dimensional state space partitioning.

[0119] Based on the B3 state parameters, the following zones are determined: high activity zone boundary: SOH > 0.8, |ΔT| > 0.05°C / s, and |I_rate| > 0.3°C; medium activity zone boundary: 0.6 ≤ SOH ≤ 0.8 or 0.02 ≤ |ΔT| ≤ 0.05°C / s or 0.1 ≤ |I_rate| ≤ 0.3°C; low activity zone boundary: SOH < 0.6, |ΔT| < 0.02°C / s, and |I_rate| < 0.1°C. B3 determination: SOH3 = 0.825 > 0.8, ΔT3 = 0.07 > 0.05, I_rate3 = 0.5 > 0.3, meeting the high activity zone conditions. Fixed thresholds for each zone: high activity zone: Th_high = 0.08 (8%); medium activity zone: Th_mid = 0.12 (12%); low activity zone: Th_low = 0.15 (15%).

[0120] 2.3. Fuzzy boundary processing.

[0121] Calculate the Euclidean distance from B3 to the center of each partition: Center of high activity zone: [0.9, 0.1, 0.8]; Center of medium activity zone: [0.7, 0.035, 0.2]; Center of low activity zone: [0.5, 0.01, 0.05]; d_high = sqrt[(0.825-0.9)] 2 +(0.07-0.1) 2 +(0.5-0.8) 2 ]=sqrt[0.005625+0.0009+0.09]=sqrt(0.096525)=0.311; d_mid=sqrt[(0.825-0.7) 2 +(0.07-0.035) 2 +(0.5-0.2) 2 ]=sqrt[0.015625+0.001225+0.09]=sqrt(0.10685)=0.327; d_low=sqrt[(0.825-0.5) 2 +(0.07-0.01) 2 +(0.5-0.05) 2 ]=sqrt[0.105625+0.0036+0.2025]=sqrt(0.311725)=0.558. Gaussian membership calculation (σ=0.1): μ_high=exp(-0.311 2 / 2×0.1 2 )=exp(-48.36)≈0;μ_mid=exp(-0.327 2 / 2×0.12 )=exp(-53.47)≈0;μ_low=exp(-0.558 2 / 2×0.1 2 )=exp(-155.65)≈0. Since the distance is too large, adjust σ=0.3: μ_high=exp(-0.311 2 / 2×0.3 2 )=exp(-5.37)=0.0047;μ_mid=exp(-0.327 2 / 2×0.3 2 )=exp(-5.94)=0.0026;μ_low=exp(-0.558 2 / 2×0.3 2 )=exp(-17.29)≈0; interpolation threshold calculation: Th_B3=(μ_high×Th_high+μ_mid×Th_mid+μ_low×Th_low) / (μ_high+μ_mid+μ_low)=(0.0047×0.08+0.0026×0.12+0×0.15) / (0.0047+0.0026+0)=0.00069 / 0.0073=0.0945; since B3 obviously belongs to the high-activity area, the final threshold is determined to be Th_B3=0.08; the fuzzy partition threshold table is obtained.

[0122] Step 3: Collaborative internal resistance mutation detection based on swarm intelligence.

[0123] 3.1. Local feature extraction and neighbor association establishment.

[0124] 3.1.1. Multi-dimensional local feature calculation. Taking battery B3 as the monitoring target, a parameter change is detected at t = 1005s: R_ct3(1005) = 0.054Ω (compared to 0.047Ω at t = 1000s). Local feature calculation: electrochemical dimension: ΔR_ct3(1005)=|0.054-0.047| / 0.047=0.149=14.9%; impedance phase angle change: Δφ3(1005)=|arg(Z3(1005))-arg(Z3(1000))|=|42.3°-41.8°|=0.5°; electrical dimension: dV / dt3=(3.76-3.79) / (1005-1000)=-0.006V / s; current distribution deviation: I_deviation3=|I3_actual-I3_expected| / I3_expected=|24.2-25.0| / 25.0=0.032; thermodynamic dimension: dT / dt3=(32.1-31.5) / (10 05-1000)=0.12°C / s; Local anomaly score calculation: Local_score3=w1×f1(ΔR_ct3)+w2×f2(Δφ3)+w3×f3(dV / dt3)+w4×f4(thermal gradient); where w1=0.4 is the electrochemical weight; w2=0.2 is the phase weight; w3=0.3 is the electrical weight; w4=0.1 is the thermodynamic weight Weights; f1(14.9%)=0.89; f2(0.5°)=0.12; f3(0.006)=0.15; f4(0.12)=0.45; Local_score3=0.4×0.89+0.2×0.12+0.3×0.15+0.1×0.45=0.356+0.024+0.045+0.045=0.47.

[0125] 3.1.2. Constructing the topological association graph of the battery pack. Constructing the topological association graph G = (V, E, W) of 8 batteries: Node set: V = {B1, B2, B3, B4, B5, B6, B7, B8}; Association weight calculation (centered on B3): Spatial weight: w_spatial(B3, B2) = exp(-0.1 2 / 2×0.5 2 )=exp(-0.02)=0.98; Electrical weight: w_electrical(B3, B2)=|3.79-3.85| -1 ×|25-25| -1=16.67×∞, set to 0.1; thermal coupling weight: w_thermal(B3, B2) = exp(-|31.5-29.1| / 5) = exp(-0.48) = 0.62; comprehensive weight: w32 = α × w_spatial + β × w_electrical + γ × w_thermal = 0.5 × 0.98 + 0.3 × 0.1 + 0.2 × 0.62 = 0.49 + 0.03 + 0.124 = 0.644. Similar calculations yield the neighboring weight matrix for B3: w31 = 0.45, w32 = 0.644, w34 = 0.67, w35 = 0.52, w36 = 0.58, w37 = 0.49, w38 = 0.41.

[0126] 3.1.3. Neighboring battery status information collection. Based on the threshold value of w_min=0.5, the neighboring battery set of B3 is: N(3)={B2, B4, B5, B6}; the neighboring battery status is collected: B2:State2=[R_ct2=0.043, ΔR_ct2=5.2%, SOC2=89%, T2=29.1°C, V2=3.85V, Score2=0.15]; B4:State4=[R_ct4=0.041, ΔR_ct4=3.8%, SOC4=88%, T4=28.8°C, V4=3.84V, Score4=0.12]; B5:State5=[R_ct5=0.044, ΔR_ct5=6.1%, S OC5=86%, T5=29.3°C, V5=3.81V, Score5=0.18]; B6: State6=[R_ct6=0.040, ΔR_ct6=2.9%, SOC6=89%, T6=28.7°C, V6=3.83V, Score6=0.08]; State correlation calculation: Correlation_32=Pearson_corr(History_B3, History_B2)=0.78; Dynamic similarity: Similarity_32(1005)=exp(-||State3(1005)-State2(1005)||2 2 / 2×0.1 2 )=exp(-0.85 2 / 0.02)=exp(-36.13)≈0; adjust σ_sim=0.5: Similarity_32(1005)=exp(-0.85 2 / 2×0.5 2 )=exp(-1.445)=0.236.

[0127] 3.2. Distributed collaborative detection and consistency analysis.

[0128] 3.2.1 Distributed Consensus Algorithm. Establish the battery pack state vector, taking four related batteries as an example: x3(1005)=[0.054, 14.9%, 0.47, 0.85]; x2(1005)=[0.043, 5.2%, 0.15, 0.92]; x4(1005)=[0.041, 3.8%, 0.12, 0.88]; x5(1005)=[0.044, 6.1%, 0.18, 0.79]. Consistency update iteration (first iteration): x3(1006)=x3(1005)+ε(1005)×[w32×(x2(1005)-x3(1005))+w34×(x4(1005)-x3(1005))+w35×(x5(1005)-x3(1005))]; Convergence step: ε(1005)=ε_base×exp(-||X(1005)-X(1004)||_F / σ_ convergence)=0.1×exp(-2.3 / 1.0)=0.1×0.1=0.01; x3 update calculation: Δx32=w32×(x2-x3)=0.644×[-0.011,-9.7%,-0.32,0.07]=[-0.007,-6.25%,-0.206,0.045] Δx34=w34×(x4-x3)=0.67×[-0.013,-11.1%, -0.35, 0.03]=[-0.0087, -7.44%, -0.235, 0.02]; Δx35=w35×(x5-x3)=0.52×[-0.010, -8.8%, -0.29 , -0.06]=[-0.0052, -4.58%, -0.151, -0.031]; x3(1006)=[0.054, 14.9%, 0.47, 0.85]+0.01×[(-0.0 The results show that the proposed method has the following three parameters: (1) the first parameter is [0.07-0.0087-0.0052), (-6.25-7.44-4.58), (-0.206-0.235-0.151), (0.045+0.02-0.031)] = [0.054, 14.9%, 0.47, 0.85] + 0.01×[-0.0209, -18.27, -0.592, 0.034] = [0.0538, 14.72%, 0.464, 0.8503]. After 5 iterations, the consensus center is: μ_consensus = [0.0455, 7.5%, 0.217, 0.86]; the state dispersion is: σ_consensus = 0.089.

[0129] 3.2.2 Multi-scale Anomaly Pattern Recognition. Calculation of individual anomaly deviation: Deviation_3 = ||x3*-μ_consensus||2 / σ_consensus = ||[0.0538, 14.72%, 0.464, 0.8503]-[0.0455, 7.5%, 0.217, 0.86]||2 / 0.089 = ||[0.0083, 7.22%, 0.247, -0.0097]||2 / 0.089 = 0.254 / 0.089 = 2.85. Individual anomaly determination: Deviation_3 = 2.85 > θ_individual = 2.5, identifying the individual anomaly candidate. Global anomaly indicator: Global_anomaly = σ_consensus / σ_historical = 0.089 / 0.075 = 1.19; Global anomaly determination: Global_anomaly = 1.19 < θ_global = 1.8, determined as a non-systematic anomaly. Anomaly type classification: Since Global_anomaly ≤ 1.8 and Deviation_3 > 2.5, the classification result is "individual mutation anomaly", and the anomaly battery ID is B3. Group detection confidence: Group_confidence = min(1.0, Convergence_speed × Neighbor_consistency × Historical_accuracy) = min(1.0, 0.92 × 0.85 × 0.88) = min(1.0, 0.688) = 0.688.

[0130] 3.2.3. Temporal and Spatial Correlation Verification. Temporal correlation verification: Analyze anomaly persistence within the time window T_window = 30 seconds: anomaly duration: Duration_anomaly = 25 seconds; persistence index: Persistence = 25 / 30 = 0.833 > 0.6, confirming a persistent anomaly; anomaly intensity trend: Trend = linear_regression_slope([0.47, 0.52, 0.58, 0.61, 0.59]) = 0.025; trend determination: |Trend| = 0.025 < 0.1, confirming a stable anomaly. Spatial Correlation Verification: Analysis of the state changes of the neighboring cells of B3 within the time series [1000s, 1010s] yields the following results: The B2 anomaly propagation correlation coefficient is: Cross_correlation(Anomaly_B3(t), Anomaly_B2(t+τ)) = max_τ∈[-5, 5]Corr = 0.42; the B4 anomaly propagation correlation coefficient is: Cross_correlation(Anomaly_B3(t), Anomaly_B4(t+τ)) = 0.38; and the maximum propagation correlation coefficient is: Propagation_corr = max(0.42, 0.38) = 0.42. Spatial propagation determination: Propagation_corr = 0.42 < 0.7, confirming that the anomaly has no obvious spatial propagation characteristics. Spatiotemporal consistency score: Spatiotemporal_consistency=w_t×Temporal_consistency+w_s×Spatial_consistency=0.6×0.833+0.4×(1-0.42)=0.6×0.833+0.4×0.58=0.50+0.232=0.732.

[0131] 3.3. Multi-layer information fusion and final confirmation decision.

[0132] 3.3.1. Bayesian Multi-Source Information Fusion. Evidence Set Definition: E1: Local_score3 = 0.47 > θ_local = 0.4, True; E2: Anomaly Type Classification Result = "Individual Mutation Anomaly", True; E3: Spatiotemporal_consistency = 0.732 > θ_consistency = 0.7, True. Likelihood Probability Calculation: P(E1|Anomaly) = sigmoid(0.47 / 0.4-1) = sigmoid(0.175) = 0.544; P(E2|Anomaly) = Group_confidence = 0.688; P(E3|Anomaly) = Spatiotemporal_consistency = 0.732. Prior Probability: P(Anomaly) = Base_rate × Age_factor × Load_factor = 0.01 × 1.2 × 1.1 = 0.0132. Dynamic weight fusion: Historical accuracy: Accuracy_local = 0.85, Accuracy_group = 0.78, Accuracy_spatial = 0.82; Weight normalization: w_local = 0.85 / 2.45 = 0.347, w_group = 0.78 / 2.45 = 0.318, w_spatial = 0.82 / 2.45 = 0.335; Final anomaly probability: Final_prob = w_local × P(E1|Anomaly) + w_group × P(E2|Anomaly) + w_spatial × P(E3|Anomaly) = 0.347 × 0.544 + 0.318 × 0.688 + 0.335 × 0.732 = 0.189 + 0.219 + 0.245 = 0.653.

[0133] 3.3.2. Multiple Timing Confirmation Mechanism. First-level confirmation (continuity test): Within the confirmation window T_confirm = 15s, Final_prob > θ_primary = 0.6 must be maintained continuously: t = 1005s: Final_prob = 0.653 > 0.6; t = 1008s: Final_prob = 0.678 > 0.6; t = 1011s: Final_prob = 0.671 > 0.6; t = 1014s: Final_prob = 0.665 > 0.6; t = 1017s: Final_prob = 0.659 > 0.6. Continuity indicator: Continuity = 15 / 15 = 1.0 > 0.8, passing the first-level confirmation. Second-level confirmation (feature consistency test): Three characteristic anomaly indicators were tested: electrochemical anomaly: ΔR_ct3 = 14.9% > 8% (threshold); electrical anomaly: |dV / dt3| = 0.006 > 0.005 (threshold); thermodynamic anomaly: |dT / dt3| = 0.12 > 0.08 (threshold); and feature consistency: Feature_consistency = 3 / 3 = 1.0 > 0.67. Passing the second-level confirmation. Third-level confirmation (historical pattern matching): Current anomaly feature vector: Current_pattern = [anomaly intensity = 0.653, propagation path = localized, duration = 25s]. DTW distance calculations were performed against the historical anomaly pattern library: Pattern_1 (internal resistance gradient pattern): DTW_distance = 0.45; Pattern_2 (temperature change pattern): DTW_distance = 0.62; Pattern_3 (charge transfer anomaly pattern): DTW_distance = 0.28; Pattern_4 (diffusion anomaly pattern): DTW_distance = 0.38. Minimum matching distance: Min_DTW = 0.28 < θ_pattern = 0.3. The pattern passed the third level of confirmation and was matched as a "charge transfer anomaly pattern." Comprehensive confirmation decision: All three levels of confirmation passed, Final_decision = "Confirmed anomaly." Final confidence: Confidence_final = min(Final_prob, Continuity, Feature_consistency) = min(0.653, 1.0, 1.0) = 0.653.

[0134] 3.3.3. Test Result Integration and Propagation Impact Assessment. Internal resistance mutation test result generation: {Abnormal battery identifier: B3, Abnormal type: Individual mutation abnormality, Detection confidence: 0.653, Abnormal time: t = 1005s, Duration: 25s, Internal resistance change: 0.047Ω → 0.054Ω (+14.9%), Propagation impact assessment: {Potentially affected batteries: [B2, B4, B5], Number of affected batteries: 3, Maximum propagation probability: 0.42, Propagation risk level: Medium risk}, Detection details for each layer: {Local detection layer: {Abnormality score: 0.47, Confidence: 0.85}, Collaborative detection layer: {Group confidence: 0.688, Abnormal type: Individual abnormality}, Consistency verification layer: {Consistency score: 0.732, Verification passed: true}}, Recommended handling measures: [Enhance monitoring, adjust balancing strategy, and control temperature]}.

[0135] Step 4: Mutation level classification and strategy switching.

[0136] 4.1. Mutation level classification.

[0137] Based on the internal resistance mutation test results, we extracted mutation data for B3: R_ct_before = 0.047Ω; R_ct_after = 0.054Ω; and change rate: Change_rate = |0.054 - 0.047| / 0.047 × 100% = 14.9%. Mutation level assessment: Mild mutation: 8%-15%; Moderate mutation: 15%-25%; Severe mutation: >25%. Since Change_rate = 14.9% < 15%, it was determined to be a mild mutation.

[0138] 4.2. Strategy mapping and target strategy determination.

[0139] Current balancing state: Passive balancing mode, current 0.5A; Strategy mapping rule: Mild mutation → Active balancing strategy; Target balancing strategy: Strategy type: Active balancing; Target current: 1.0A; Balancing time: 240s; Topology adjustment: No reconstruction required.

[0140] 4.3. Execution parameter generation.

[0141] Specific execution parameters: Safety current calculation: I_safe = min(1.0, 2.0, 1.5) = 1.0A; dynamic time adjustment: T_dynamic = 240×(1+14.9 / 100)×0.653 = 240×1.149×0.653 = 180.1s; gradual switching time: 15s linear climb to target current; strategy switching instruction: {target battery: B3, switching type: passive balancing → active balancing, current parameters: {start: 0.5A, target: 1.0A, climb time: 15s}, time parameters: {balancing duration: 180s, monitoring interval: 5s}, safety protection: {current upper limit: 1.5A, temperature protection: 45°C}, execution timing: {start time: t0+1s, expected completion: t0+196s}}.

[0142] In another specific embodiment, taking an energy storage module composed of 8 second-use ternary lithium batteries as an example, the batteries are numbered B1-B8, with a rated capacity of 50Ah, a rated voltage of 3.7V, a current SOC range of 85%-92%, and an operating temperature of 25°C.

[0143] Step S41: Quantitative calculation of mutation level classification.

[0144] When a sudden change in internal resistance was detected for battery B3, key data was extracted from the internal resistance mutation test results: Battery B3's R_ct value before the mutation: R_ct_before = 0.045Ω; Battery B3's R_ct value after the mutation: R_ct_after = 0.052Ω; mutation detection confidence: Confidence = 0.89. The internal resistance change rate was calculated as Change_rate = |R_ct_after - R_ct_before| / R_ct_before × 100% = |0.052 - 0.045| / 0.045 × 100% = 15.56%. Change_rate is the relative rate of change in internal resistance; R_ct_before is the charge transfer impedance value before the mutation; and R_ct_after is the charge transfer impedance value after the mutation. Mutation classification is performed based on the preset mutation classification criteria: Light mutation threshold: Threshold_light = 8%-15%; Moderate mutation threshold: Threshold_medium = 15%-25%; Severe mutation threshold: Threshold_severe=>25%. Since Change_rate = 15.56%, which falls within the moderate mutation range, the mutation classification result is: Mutation Classification = {Battery ID: B3, Class: Moderate Mutation, Change Rate: 15.56%, Confidence: 0.89}.

[0145] Step S42: query the policy mapping table and determine the target policy.

[0146] Read the current balancing status of the battery pack: Current balancing mode: Passive balancing mode; Current balancing current: I_balance_current = 0.5A; Battery pack voltage inconsistency: ΔV_max = 0.12V; Balancing topology: Standard series topology. Query the preset strategy mapping table. The mapping rule for moderate mutations is: strategy mapping function (mutation level, current state) → target strategy. When mutation level = "moderate mutation" and current mode = "passive balancing", the target strategy is fast balancing mode. Fast balancing mode includes two sub-strategies: balancing current enhancement and balancing time reduction. The balancing current enhancement factor is 2.5, and the balancing time reduction ratio is 0.6. Determine the target balancing strategy based on the mapping results: Target balancing strategy = {Strategy type: Fast balancing mode, Current setting: I_target = I_balance_current × Enhancement_factor = 0.5 × 2.5 = 1.25 A, Time adjustment: T_target = T_standard × Time_reduction = 300 s × 0.6 = 180 s, Topology maintained: Series topology remains unchanged}.

[0147] The execution parameters in the strategy mapping table contain three levels of control elements. The current control layer includes the starting current I_start, target current, ramp-up time, and safety upper limit. The ramp-up time is dynamically adjusted based on the battery temperature and SOC: when the temperature is below 25°C, the ramp-up time is extended by 20% to protect the battery; when the SOC is below 30%, the ramp-up time is extended by 30% to avoid excessive stress. The time control layer includes the balancing duration, monitoring sampling interval, and timeout protection. The monitoring interval is adaptively adjusted based on the mutation level: 10 seconds for mild mutations, 5 seconds for moderate mutations, and 2 seconds for severe mutations. The topology control layer includes the switch action sequence, battery isolation procedure, and reconnection logic to ensure the safety and continuity of the topology change process.

[0148] Step S43: Execute parameter generation and gradual switching.

[0149] Generate specific execution parameter configuration based on the target balancing strategy: Balance current size setting calculation: Safe current calculation I_safe = min(I_target, I_max_hardware, I_thermal_limit) = min(1.25, 2.0, 1.5) = 1.25A; where I_target is the target balancing current; I_max_hardware is the maximum hardware current limit; I_thermal_limit is the thermal management current limit; finally determine the safe balancing current I_balance_final = 1.25A. Balancing duration calculation: Considering the sudden change of battery B3, the balancing time needs to be dynamically adjusted: Dynamic time calculation T_dynamic = T_target × (1 + Change_rate / 100) × Confidence = 180 × (1 + 15.56 / 100) × 0.89 = 180 × 1.1556 × 0.89 = 185.2s; where T_target is the baseline balancing time; Change_rate is the internal resistance change rate; Confidence is the detection confidence level; and the final balancing duration T_balance_final = 185s. Gradual switching parameter configuration: To avoid system shock caused by policy jumps, a gradual current ramp-up process is constructed: the current ramp-up curve is I(t)=I_current+(I_balance_final-I_current)×(1-exp(-t / τ)); where I_current is the current current of 0.5A; I_balance_final is the target current of 1.25A; τ is the time constant of 5s; and t is the switching time variable. At t=15s, 95% of the target current is reached, that is, I(15)=0.5+(1.25-0.5)×(1-exp(-15 / 5))=0.5+0.75×0.95=1.21A. Topology reconstruction instruction configuration: Since this is a moderate mutation, no topology reconstruction is required, and the current series connection is maintained: Topology reconstruction instruction = {Reconstruction type: No reconstruction required, Switch state: Maintain current state, Connection mode: Series topology, Isolation instruction: None}. Integrate all execution parameters to generate a complete strategy switching instruction: Strategy switching instruction = {target battery: B3, switching type: passive balancing → fast balancing mode, current parameters: {starting current: 0.5A, target current: 1.25A, ramp time: 15s, ramp curve: exponential}, time parameters: {balancing duration: 185s, monitoring interval: 5s, timeout protection: 300s}, topology parameters: {reconstruction instruction: none, switch action: none, connection state: maintain series connection}, safety protection: {current upper limit: 1.5A, temperature protection: 45°C, voltage protection: 4.2V}, execution timing: {instruction issuance time: t0, execution start time: t0+1s, expected completion time: t0+201s}}.

[0150] The study found that the parameters established by the offline method based on historical test data had a deviation of 18% after 3 months, while the parameter deviation of the real-time identification method was always controlled within 5%. When the battery undergoes progressive attenuation, the offline method needs to be re-tested and calibrated, which takes 2-3 days; the present invention can complete the parameter update within 1 hour and realize continuous tracking of the attenuation process. The present application is specifically optimized for the electrochemical characteristics of cascade lithium batteries. The uneven distribution of SEI film thickness of cascade batteries leads to the complication of double-layer characteristics. A thickness distribution factor is introduced in the double-layer branch to correct the calculation of the double-layer capacitance C_dl. The local failure of the active material of the cascade battery affects the charge transfer process. The present invention adds an active area correction coefficient to the charge transfer branch. This improves the fitting accuracy of the model in cascade battery applications, which is significantly better than the direct application effect of the original supercapacitor model.

[0151] In some embodiments, when the ambient temperature fluctuates significantly (the rate of change exceeds 1°C / min), the fuzzy partition threshold table needs to be corrected in real time. When the temperature rises, the internal resistance of the battery usually decreases, and the system will lower the detection threshold by 10-15% to maintain detection sensitivity; when the temperature drops, the internal resistance of the battery increases, and the system will increase the detection threshold by 15-20% to avoid false alarms caused by temperature changes. The correction process uses a sliding average method to achieve a smooth transition: Th_corrected(k) = 0.7×Th_original + 0.3×Th_temperature_adjusted to avoid the impact of threshold mutations on detection stability. Among them, Th_corrected is the fuzzy partition detection threshold after temperature correction at the current moment, Th_original is the initial threshold without temperature adjustment, and Th_temperature_adjusted is the temperature response threshold calculated based on the current temperature change trend.

[0152] When the load current experiences a step change (a change exceeding 0.5C), the system temporarily increases the weight of collaborative detection to enhance robustness. For example, the group detection weight w_group is temporarily increased from 0.318 to 0.45, the spatiotemporal correlation weight w_spatial is increased from 0.335 to 0.4, and the local detection weight w_local is correspondingly reduced to 0.15. This weight adjustment lasts for 30 seconds before returning to normal. During this period, the uncertainty of single-cell detection is reduced by correlating information about the status of neighboring batteries. The load change compensation mechanism is also activated to identify and filter normal internal resistance fluctuations caused by sudden load changes.

[0153] In response to the nonlinear distortion problem of frequency domain impedance analysis, the present invention differs from the traditional mathematical frequency domain decomposition method and adopts a physical modeling method based on the electrochemical mechanism. Specifically, a three-branch electrochemical equivalent circuit model is constructed, and the complex impedance response of the cascade battery is decomposed into three components with clear physical meanings: the double-layer capacitance branch, the charge transfer impedance branch, and the Warburg impedance branch. Each branch corresponds to a different electrochemical process and time constant. The equivalent circuit parameters are directly identified in real time in the time domain, and electrochemical constraints and parameter continuity verification mechanisms are introduced to ensure the physical rationality of the identification results; the nonlinear intermodulation interference problem in the frequency domain decomposition process is avoided. At the same time, the extracted charge transfer impedance parameters are highly sensitive to changes in the internal structure of the cascade battery, and can accurately capture instantaneous internal resistance mutations, shortening the detection delay. In response to the contradiction between the convergence and responsiveness of the adaptive threshold, a state space partition threshold strategy is proposed to transform the continuous adaptive problem into a discrete partition problem. Based on the electrochemical activity theory, three key physical quantities, SOH, temperature change rate and current rate, are selected to construct a three-dimensional state space, reducing the high-dimensional parameter space to a three-dimensional space with clear physical meaning. According to the electrochemical activity level, the state space is divided into three active zones: high, medium and low. An optimized fixed threshold is used in each partition, which completely avoids the convergence problem of the traditional adaptive algorithm. In order to solve the hard switching problem of the partition boundary, a fuzzy boundary processing mechanism based on the Gaussian membership function is constructed. By calculating the Euclidean distance from the battery state point to the center of each partition, a weighted average method is used to achieve a smooth transition of the threshold. This strategy of combining partition fixation with boundary fuzziness not only ensures the stability and predictability of the threshold setting, but also achieves a fast response to different battery states. The response time is controlled within 3 seconds, while avoiding the strategy oscillation phenomenon.

[0154] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A BMS balancing strategy intelligent switching method for a second-use battery pack, characterized in that: include: The current sensor data of the battery pack is collected and, after preprocessing and noise filtering, real-time parameter identification is performed based on the three-branch electrochemical equivalent circuit model to obtain the electrochemical impedance parameter set and clean data set; The electrochemical impedance parameter set is monitored using a mutation detection mechanism, and threshold matching judgment is performed in combination with a pre-configured fuzzy partition threshold table to output the internal resistance mutation detection result. Based on the internal resistance mutation detection results, mutation level classification and strategy mapping are performed to generate specific execution parameters and output strategy switching instructions; Get the electrochemical impedance spectroscopy parameter set and clean data set, including: Collect sensor data from battery packs in continuous use, including cell voltage, injected excitation current, battery temperature, and SOC estimation. Perform data synchronization, timestamp standardization, noise filtering, and outlier detection to obtain a clean data set. Based on the voltage and current data in the clean dataset, a three-branch electrochemical equivalent circuit model consisting of a double-layer capacitance branch, a charge transfer impedance branch, and a Warburg impedance branch was constructed. The equivalent circuit model was obtained and parameter identification was performed in real time to extract the electrochemical impedance parameter set, including double-layer impedance, charge transfer impedance, and Warburg impedance parameters. Get the equivalent circuit model, including: Construct a topological structure of a three-branch parallel equivalent circuit: the first branch is a double-layer capacitor in parallel with a double-layer resistor, the second branch is a charge transfer resistor in parallel with a constant phase element, and the third branch is a Warburg impedance; Based on the electrochemical characteristics of the cascade battery, the physical boundary conditions of the parameters are set, including the value ranges of the solution resistance, double layer resistance, and charge transfer resistance, as well as the impedance magnitude relationship constraints, to obtain the physical constraint parameter matrix; The topological structure and physical constraint parameter matrix are integrated to construct an equivalent circuit model suitable for the nonlinear electrochemical characteristics of cascade batteries.

2. The method according to claim 1, characterized in that Extract electrochemical impedance spectroscopy parameters, including: Based on the equivalent circuit model and the clean data set, a linear parameter identification equation set is established and solved to obtain the initial parameter estimates. Physical constraint checks are then performed on the equations. When the boundary conditions in the physical constraint parameter matrix are violated, the parameters are corrected back to the physically feasible region to obtain the constraint correction parameters. The change rate of the constraint correction parameters at adjacent moments is calculated, and when the change rate exceeds the continuity threshold, filtering is performed to obtain the electrochemical impedance parameter set.

3. The method according to claim 1, characterized in that The configuration of the fuzzy partition threshold table includes: Based on the historical sensor data of the used battery pack, the historical electrochemical impedance parameter set and the clean data set are obtained. Based on this, the SOH, temperature change rate and current rate are calculated to obtain the activity parameter matrix. Based on the activity parameter matrix, the electrochemical activity is used to set the boundaries of high, medium, and low activity zones. Different double layer impedance mutation detection thresholds are set for each zone, resulting in a three-dimensional zone boundary and fixed zone threshold table. For the boundary area of ​​the three-dimensional partition boundary, the Euclidean distance to the center of each partition is calculated, and the interpolation threshold is obtained by weighted average and smoothed. Based on this, the fixed partition threshold table is updated to the fuzzy partition threshold table.

4. The method according to claim 3, characterized in that Get the fixed partition threshold table, including: Calculate quantitative activity indicators based on the activity parameter matrix, including SOH, temperature change rate and current multiplier; A three-dimensional physical state space coordinate system is constructed based on the quantitative activity index. The high activity area is the area where the quantitative activity index is higher than the first threshold, the low activity area is the area where the quantitative activity index is lower than the second threshold, and the medium activity area is the area between the two, thus obtaining the three-dimensional partition boundaries. Based on the inverse relationship between electrochemical activity and internal resistance mutation sensitivity, corresponding mutation detection thresholds are set for high-activity areas, low-activity areas, and medium-activity areas within the three-dimensional partition boundaries, and a fixed partition threshold table is obtained.

5. The method according to claim 3, characterized in that Update the fixed partition threshold table to the fuzzy partition threshold table, including: Based on the fixed partition threshold table and the three-dimensional partition boundaries, the Euclidean distance from the battery status point to each partition center is calculated using the Gaussian membership function, and the weighted average is used to obtain the original interpolation threshold. Based on this, the maximum and minimum membership differences are calculated. When the difference is less than the preset threshold, it is determined to be a fuzzy area and the fuzzy area identifier is obtained. For the boundary area in the fuzzy area identification, the Sigmoid function is used to calculate the smoothing weight according to the distance from the state point to the center of the partition boundary to obtain the smoothing weight coefficient; The original interpolation threshold is modified based on the smoothing weight coefficient, and the fixed partition threshold table is updated to a fuzzy partition threshold table accordingly.

6. The method according to claim 1, characterized in that Output internal resistance mutation detection results, including: Monitor the rate of change of the charge transfer impedance in the electrochemical impedance parameter set, obtain the individualized detection threshold through three-dimensional interpolation in combination with the fuzzy partition threshold table, perform mutation judgment, and obtain the main channel detection signal; Calculate the abnormal change of the voltage response slope in the clean data set when the current pulse is injected, perform independent mutation detection, and obtain the auxiliary channel detection signal; A differential confirmation analysis is performed on the main channel detection signal and the auxiliary channel detection signal: when a mutation is detected by both channels at the same time, it is confirmed as a real mutation; when only a single channel is detected, it is determined to be measurement interference, and the internal resistance mutation detection result is obtained.

7. The method according to claim 1, characterized in that Output internal resistance mutation detection results, including: The multi-dimensional local features of the double-layer impedance in the electrochemical impedance parameter set are extracted to construct a battery pack topology association diagram that includes physical, electrical, and thermal coupling. Based on this, the state information of neighboring batteries is collected and the association weight is calculated to obtain a local anomaly score and a collection of neighboring state information. Based on the neighboring state information set, state propagation is performed to identify and distinguish individual mutation anomalies and systemic interference. Combined with spatiotemporal correlation verification to confirm the abnormal propagation characteristics, the group detection confidence and spatiotemporal correlation verification results are obtained; The local anomaly score, group detection confidence and spatiotemporal correlation verification results are integrated to perform persistence and feature consistency tests, as well as historical pattern matching. The final confirmation decision is made in combination with the fuzzy partition threshold table to obtain the internal resistance mutation detection result.

8. The method according to claim 7, characterized in that Obtain group detection confidence and spatiotemporal correlation verification results, including: Based on the set of neighboring state information, a battery pack state vector is established, and state information propagation is performed. When the global convergence condition is met, the consistency center and state dispersion are calculated to obtain the consistency convergence state. Based on this, the abnormal deviation of the individual battery and the global abnormality index are calculated. When the individual deviation exceeds the threshold and the global abnormality index is normal, it is classified as an individual mutation abnormality. When the global abnormality index exceeds the threshold, it is classified as a systematic interference. The abnormality type classification result is obtained and the group detection confidence is calculated. Based on the anomaly type classification results, the anomaly persistence and intensity change trends are analyzed, the propagation correlation coefficient is calculated, the anomaly propagation mode is identified, the spatiotemporal consistency score is calculated, and the spatiotemporal correlation verification results are obtained.

Images