A method and system for optimizing the state of a battery pack for electric scooters

Abstract

This invention provides a battery pack state optimization testing method and system for electric scooters, relating to the field of battery technology. The method includes: classifying cells based on their physical location coordinates within the battery pack; sampling cells using different sampling methods according to their type to obtain high-precision initial data; pre-sampling all cells within the battery pack; calculating the voltage jump rate of each cell using the pre-sampled data and the high-precision initial data; classifying cells based on their voltage jump rate; inputting the cell's physical location coordinates, electromagnetic interference intensity, voltage jump rate, and cell type into a priority calculation formula to calculate priorities; planning sampling paths based on these priorities; and collecting cell state data in real-time based on the sampling paths; inputting the cell state data into a three-dimensional threshold pool for anomaly detection; inputting the anomaly detection results into a fault diagnosis model to obtain fault indication; and improving accuracy and efficiency through a multi-dimensional integrated battery pack testing system.

Description

Technical Field

[0001] This invention relates to the field of battery technology, specifically to a method and system for optimizing the state of a battery pack for electric scooters. Background Technology

[0002] With the increasing popularity of electric scooters, the accuracy and reliability of battery pack status monitoring in complex operating environments has become a technical challenge. Market research indicates that current testing methods for electric scooter battery packs often rely on single-sensor data acquisition or preset fixed thresholds. These methods struggle to integrate multi-source information such as electromagnetic interference, temperature gradients, and cell group characteristics in real time. This leads to risks such as excessive sampling data errors, misjudgments of status, and delayed fault location in scenarios involving strong motor interference, high and low temperature environments, and cell aging.

[0003] For example, some solutions do not consider the dynamic correlation between electromagnetic interference intensity and the physical location of the battery cell, and only process the data through simple mean filtering. Under high-frequency interference from the motor, the voltage sampling error often exceeds 1mV, which cannot reflect the true state of the battery cell. Other solutions use a fixed threshold to determine the abnormality of the battery cell, ignoring the performance difference between new and old batteries and the influence of temperature on voltage characteristics. The misjudgment rate is as high as 20% in high or low temperature environments.

[0004] Furthermore, existing testing models largely rely on manual experience to set sampling paths and fault diagnosis rules. They lack self-optimization mechanisms when facing high-risk conditions (such as cell short circuits and deterioration of group consistency), making it difficult to guarantee long-term testing accuracy and efficiency. Therefore, how to construct an intelligent testing system that integrates multi-source data to achieve high-precision sampling of battery pack status, dynamic threshold determination, rapid fault diagnosis, and data traceability has become a key technical bottleneck in improving the safety and reliability of electric scooter battery packs. Summary of the Invention

[0005] To address the aforementioned technical challenges, this solution employs a multi-dimensional integrated battery pack testing system to achieve high-precision sampling and dynamic judgment, thereby improving accuracy and efficiency and ensuring safety.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a battery pack state optimization testing method for electric scooters, the method comprising:

[0007] Based on the physical location coordinates of the cells within the battery pack, the cells are divided into Class I cells and Class II cells. The sampling method for Class I cells is determined based on the electromagnetic interference intensity. Both Class I and Class II cells are sampled to obtain high-precision initial data.

[0008] All cells in the battery pack are pre-sampled, and the voltage jump rate of each cell is calculated by combining the pre-sampled data with high-precision initial data. The cells are then classified according to their voltage jump rate.

[0009] Input the physical location coordinates of the battery cell, electromagnetic interference intensity, voltage sag rate and battery cell type into the priority calculation formula to calculate the priority of each battery cell. Based on the priority of each battery cell, a sampling path is planned, and battery cell status data is collected in real time according to the sampling path.

[0010] The cell status data is input into a three-dimensional threshold pool containing individual thresholds, group thresholds, and environmental thresholds, and anomaly determination is made by using a veto and weighted voting method.

[0011] Input the anomaly determination results into the fault diagnosis model to obtain the fault location.

[0012] Furthermore, the sampling methods for the first type of battery cell and the second type of battery cell are as follows:

[0013] Battery cells with a distance of less than or equal to 5cm between their geometric center and the center of the motor output shaft are classified as Class I battery cells, and the rest are classified as Class II battery cells.

[0014] When the electromagnetic interference intensity is greater than the interference intensity threshold, multi-dimensional adaptive sampling is used to perform seven consecutive samplings on the first type of battery cell. When the electromagnetic interference intensity is less than or equal to the interference intensity threshold, the sampling method is the same as that for the second type of battery cell.

[0015] Three consecutive samplings were used to sample the two types of battery cells and the dynamic weighted mean was calculated.

[0016] The acquired signals are processed using a variational mode decomposition algorithm.

[0017] Furthermore, the specific process of the variational mode decomposition algorithm for processing signals is as follows:

[0018] The collected cell voltage signal is filtered by a 50Hz low-pass filter to remove power grid interference and retain the effective frequency band of 0-10kHz. The preprocessed signal is input into the algorithm in the form of a discrete time series and the sampling frequency is fixed at 10kHz.

[0019] Based on the signal characteristics of the battery pack, three modal components are preset, corresponding to the high-frequency interference mode, the mid-frequency coupling mode, and the low-frequency effective mode, respectively.

[0020] The signal is converted to the frequency domain by Fourier transform, and an objective function is constructed with the constraint that the superposition of all modal components equals the original signal.

[0021] The center frequency and amplitude of each modal component are updated iteratively. After each iteration, the spectral overlap of each component is calculated. The iteration is stopped when the overlap is less than 5%.

[0022] Through energy proportion analysis, low-frequency effective modes with an energy proportion greater than 80% are retained, while high-frequency interference modes and mid-frequency coupling modes are eliminated.

[0023] The selected low-frequency effective modes are subjected to inverse Fourier transform to convert them back to time domain signals. Then, missing data points are filled in by linear interpolation to output a smooth voltage curve.

[0024] Furthermore, the specific process of classifying these categories is as follows:

[0025] The absolute value of subtracting the high-precision initial data from the pre-sampling result is the voltage jump rate. If the voltage jump rate is greater than 5mV / 20ms, the cell is a transient fluctuation cell; if the voltage jump rate is between 1mV / 20ms and 5mV / 20ms, the cell is a metastable cell; if the voltage jump rate is less than 1mV / 20ms, the cell is a voltage-stable cell.

[0026] Furthermore, the specific process of sampling path planning is as follows:

[0027] The parameters of the physical location coordinates of the battery cell, electromagnetic interference intensity, and voltage sag rate are quantified into a numerical matrix. The numerical matrix and the battery cell type are then input into the priority calculation formula to obtain the priority of each battery cell. Based on the priority, the sampling order of the battery cells is determined from largest to smallest, and the sampling path is determined.

[0028] Every 5 seconds, the sampling path is replanned based on the newly collected data. During the 5-second interval between two path refreshes, the sampling system performs high-frequency monitoring of high-priority cells according to the current path, and performs one verification sampling of low-priority stable cells to obtain voltage data. For the second-class cells, the result after three dynamic weight averages is retained. The newly collected data is input into the priority calculation formula to re-sort the priorities and obtain a new sampling path.

[0029] Furthermore, the construction process of the three-dimensional threshold pool is as follows:

[0030] Extract the voltage data from the first three complete tests of each cell, calculate the voltage sequence of each test, calculate the voltage mean and standard deviation of the first three tests, and determine the individual threshold.

[0031] Collect real-time voltage data of all cells in the battery pack, calculate the group voltage distribution entropy, and determine the group threshold;

[0032] The ambient temperature is divided into 5 intervals, each with an independent temperature-voltage coupling coefficient. When the electromagnetic interference intensity is greater than the interference intensity threshold, the voltage threshold is corrected to determine the environmental threshold.

[0033] By statistically analyzing failure cases, the weights of three dimensions are determined, and individual thresholds, group thresholds, and environmental thresholds are mapped to a three-dimensional space. The comprehensive deviation is obtained by weighted summation of the thresholds for each dimension.

[0034] Furthermore, the specific processes of the veto and the weighted voting are as follows:

[0035] At the individual level, if the real-time voltage exceeds twice the individual threshold range, it is considered abnormal. At the group level, if the group distribution entropy is greater than 1.2 and the duration is greater than 2 seconds, it is considered abnormal. At the environmental level, if the electromagnetic interference intensity is greater than 150μT or the temperature is greater than 50℃, it is considered abnormal. When no veto is triggered, each dimension is classified into levels according to the degree of deviation, and a comprehensive score is calculated. When the comprehensive score is greater than or equal to 1.5, it is considered abnormal. A comprehensive score greater than 1.0 and less than 1.5 is considered a warning. A comprehensive score less than or equal to 1.0 is considered normal.

[0036] Furthermore, the fault location acquisition process is as follows:

[0037] Taking battery pack malfunction as the top event, the system is broken down into a three-level structure of system, subsystem, and component to form a fault tree containing several basic events. The fault tree contains first-level sub-events and second-level sub-events. Each second-level sub-event is further broken down into quantifiable fault modes, and each basic event is associated with specific characteristic parameters.

[0038] Real-time feature maps are constructed based on real-time monitoring data from a three-dimensional threshold pool.

[0039] The real-time feature map is matched with the basic event feature library of the fault tree, and a similarity score is calculated. Weights are assigned according to the importance of the features. A similarity score of 85% or higher is considered a confirmed case, while a similarity score between 60% and 85% is considered a suspected case, requiring further sampling verification.

[0040] For the top 3 basic events with the highest similarity scores, backtrack the logical chain of the fault tree to verify whether the characteristics of their parent events are satisfied.

[0041] Furthermore, after obtaining the fault location, the source is traced through atomic data nodes, as shown in the following process:

[0042] Each sampled data is broken down into atomic units consisting of timestamps, physical location codes, and feature values;

[0043] A unique step fingerprint is generated for each test step, consisting of the step number, device status code, and environmental parameter hash value.

[0044] Atomized units and step fingerprints are linked through blockchain-style hashing;

[0045] When an anomaly occurs, the atomic unit corresponding to the anomaly parameter is retrieved, and the step fingerprint is reversed to locate the specific sampling time, physical location, and environmental state.

[0046] The present invention also provides a battery pack state optimization testing system for electric scooters, the system being used to execute the aforementioned battery pack state optimization testing method for electric scooters.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] 1. This solution uses environmentally coupled dynamic anti-interference sampling. For Class I cells, it employs 7 samplings combined with variational mode decomposition and wavelet thresholding for noise reduction. For Class II cells, it employs 3 dynamic weighted averaging processes to control voltage error within 1mV and temperature deviation within 0.5℃. This effectively counteracts the effects of electromagnetic interference and temperature gradients, providing high-precision initial data for subsequent testing.

[0049] 2. An evolutionary threshold pool is constructed based on three dimensions: individual, group, and environment. Combined with the "one-vote veto and weighted voting" mechanism, the threshold is adjusted in real time according to the cell status and environment. The accuracy of anomaly identification is increased to 98.2%, and the false judgment rate is reduced to 1.8%. This solves the problem that traditional fixed thresholds cannot adapt to complex working conditions and provides a reliable basis for fault diagnosis.

[0050] 3. By rapidly locating faults through bidirectional mapping of fault trees and instantaneous feature maps, and combining the "data atom and step fingerprint" binding mechanism, instantaneous source tracing is achieved within 1 second, reducing the anomaly investigation time from 15 seconds to 2 seconds, improving the anomaly source tracing efficiency by 90%, and forming a complete closed loop of "sampling-judgment-diagnosis-source tracing", which greatly improves production and maintenance efficiency. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0052] Figure 1 A schematic diagram illustrating the steps of a battery pack state optimization test method for electric scooters;

[0053] Figure 2 This is a schematic diagram of the battery pack state sampling and threshold determination process of a battery pack state optimization test device for electric scooters.

[0054] Figure 3 This is a schematic diagram of the battery pack fault diagnosis and tracing process of a battery pack state optimization testing device for electric scooters. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0056] like Figure 1 and Figure 2 As shown, a battery pack state optimization testing method for electric scooters addresses the low sampling accuracy of core data caused by electromagnetic interference and temperature gradients through environmentally coupled dynamic anti-disturbance sampling, reducing the error of initial sampling data and improving its accuracy. It also addresses low sampling efficiency and accuracy through instantaneous state feature extraction and sampling path planning. Furthermore, it overcomes the problem of traditional fixed thresholds being unable to adapt to different battery pack states and environmental changes through a self-evolving threshold pool and multi-parameter coupled decision-making, allowing the threshold to evolve in real-time with individual states and environmental changes, thereby reducing the false positive rate and improving test accuracy. Finally, it addresses the inefficiency of traditional anomaly troubleshooting relying on manual experience through fault tree and instantaneous feature map fusion diagnosis, quickly locating faults and improving test efficiency. Finally, it addresses the issue of easily tampered data leading to low traceability accuracy, which affects production and maintenance efficiency through atomic data node instantaneous tracing.

[0057] Furthermore, environment-coupled dynamic disturbance rejection sampling is the process of collecting static basic data during the initial startup of the test system. The specific process is as follows:

[0058] Cells located near the motor within the battery pack are subject to stronger electromagnetic interference due to the motor and require close monitoring. During the battery pack design phase, the relative positions of the motor and cells are clearly defined. Cells are categorized based on their distance from the motor by marking cell coordinates or pre-setting position parameters in the battery management system. Cells within 5cm of the motor are grouped into one category (these cells are most affected by electromagnetic interference). For these cells near the motor, a built-in miniature magnetic field sensor monitors the surrounding electromagnetic interference intensity (in μT) in real time. When the detected interference intensity exceeds a threshold, multi-dimensional adaptive sampling is automatically initiated. After sampling, the acquired signal is processed using a variational mode decomposition algorithm. This algorithm decomposes complex voltage signals into multiple modal components with different frequency characteristics, effectively separating interference signals from the true voltage signal. Next, wavelet thresholding is applied to the separated true voltage signal. By setting a reasonable threshold, noise components are removed from the signal, preserving the true voltage information. Three outliers are removed, and the average of the remaining four valid data points is taken. The data is then converted back to the time domain using an inverse Fourier transform, and linear interpolation is used to complete the data points, outputting a smooth voltage curve (e.g., 3800±1mV) to ensure an error ≤1mV. When the detected interference intensity is less than or equal to the interference intensity threshold, the sampling method is the same as that for Class II cells.

[0059] The remaining cells were classified into two categories. For these secondary cells, data processing was performed using a three-step method combined with dynamic weighted averaging. The dynamic weight allocation was based on the real-time temperature gradient. When the temperature difference between adjacent cells exceeded 2℃, the weight of the high-temperature cell was increased by 20% to more accurately reflect its state. Simultaneously, the sampling timing was synchronized with the device's electromagnetic pulse cycle (derived from the motor's operating status), deliberately avoiding the strong interference period of 0-200ms during motor startup. The total sampling time for each group was strictly controlled within 2 seconds to ensure data acquisition was completed within the limited testing time. A weighted average was used to calculate the final value: Final value = (High-temperature cell voltage × 1.2 + Low-temperature cell voltage × 0.8) / 2 (proportional expansion for multiple cells), offsetting deviations caused by temperature gradients (error ≤ 0.5℃).

[0060] Environmental data is recorded synchronously during sampling: electromagnetic interference intensity (μT), cell temperature (°C), motor speed (rpm), and sampling timestamp (accurate to 10μs). A label is added to the sampling results of each cell, in the format: [Cell ID]-[Distance from Motor (cm)]-[Interference Intensity (μT)]-[Temperature (°C)]-[Voltage (mV)], for example, "C1-3.2-65-28-3802". All valid data from all cells are converted into a standardized array containing: physical location code, sampling timestamp, voltage value (mV), temperature value (°C), interference intensity (μT), and data confidence level (based on error verification results, categorized into "high / medium / low" levels). This data is sent as a real-time data stream to the "Instantaneous State Feature Extraction Sampling Path Planning" module as the basis for path optimization, while the original sampling records (including outliers) are cached locally for traceability.

[0061] The design characteristics of electric scooters dictate a compact battery pack space, with close proximity between the battery cells and the motor. Electromagnetic interference generated during motor operation and temperature gradients generated during cell operation significantly impact cell data. Variational Mode Decomposition (VMD) algorithms can effectively separate signals of different frequencies based on their frequency characteristics, making them ideal for handling signals in such complex electromagnetic interference environments. Wavelet thresholding denoising can precisely filter noise based on the different characteristics of signals and noise. The combination of these two methods can specifically address the complex interference problems encountered in battery pack testing, meeting the industry's demand for high-precision cell data acquisition. By directly incorporating environmental interference parameters into the sampling decision process, and through multi-round sampling and signal processing algorithms, the signal-to-noise ratio of core data is increased to 95%, voltage error is strictly controlled within 1mV, and temperature deviation is controlled within 0.5℃. This provides high-precision initial data for subsequent instantaneous state feature extraction and sampling path planning, ensuring that subsequent stages can be carried out based on reliable data.

[0062] Specifically, the distance between the motor and the battery cell is determined by the center of the motor output shaft and the geometric center of the battery cell. The motor output shaft is the core component for power transmission in electric scooters, and its position has uniqueness and stability in terms of mechanical structure. The output shaft is the area with the strongest electromagnetic radiation from the motor (magnetic field strength decreases with the square of the distance), and using it as a reference can accurately reflect the actual intensity of electromagnetic interference to the battery cell. Furthermore, the relative position of the output shaft and the battery pack is fixed in the overall vehicle assembly (positioned through the frame screw holes), with an error of ≤1mm, avoiding measurement deviations caused by irregularities in the motor casing. The battery cell uses cylindrical or square packaging, and the voltage sampling point (tab) of the battery cell is usually located near the central axis. The electromagnetic interference at the center position is closest to the sampling point, and the geometric center can be directly calculated from the battery cell casing dimensions (e.g., for a cylindrical battery cell with a diameter of 18mm, the center distance from each end is 9mm), without the need to disassemble the battery cell, making operation convenient. The compact structure of electric scooters (the battery pack and motor are usually integrated into the same cavity of the frame) dictates that distance measurement must consider both "mechanical accuracy" and "electromagnetic characteristics." Using the output shaft and the center of the battery cell as reference points not only conforms to the standardized process of mechanical design, but also accurately correlates the physical characteristics of electromagnetic interference, avoiding the errors caused by "using the edge of the casing as a reference".

[0063] Specifically, the interference intensity threshold is preferably 50μT. The normal operating voltage range of the electric scooter battery cell is 3.2-4.2V (3200-4200mV), and the test requires a voltage acquisition error ≤1mV. Experiments show that when the electromagnetic interference intensity is ≤50μT, the induced electromotive force generated by the interference signal in the battery cell voltage sampling circuit is ≤0.8mV, which is within the allowable error range and can be eliminated by conventional 3 samplings and mean filtering. When the interference intensity is >50μT, the induced electromotive force increases sharply to 1.5-3mV, exceeding the basic filtering capability, and the interference must be offset by increasing the number of samplings (7 times) and using a composite filtering algorithm. This threshold directly corresponds to the anti-interference critical value of the battery cell signal, ensuring that the sampling accuracy does not exceed the design error limit. The electromagnetic radiation intensity of the drive motor of an electric scooter (usually a brushless DC motor) exhibits a "segmented attenuation" characteristic with distance. The interference intensity within 3cm of the center of the motor output shaft can reach 100-300μT (mainly high-frequency pulse interference). The intensity in the 3-5cm range drops to 50-100μT (fluctuating with the motor speed, with the peak value possibly exceeding 50μT). Beyond 5cm, it stabilizes below 50μT (mainly low-frequency magnetic field, with smooth interference characteristics). 50μT is precisely the dividing point between the "strong interference zone" and the "weak interference zone" of the motor radiation, which corresponds to the physical threshold of the battery cell being 5cm away from the motor (the median interference intensity at a distance of 5cm is 48μT, with a 90% probability of being ≤50μT).

[0064] Referring to GB / T18487.1-2015 "Conductive Charging Systems for Electric Vehicles - Part 1: General Requirements" regarding electromagnetic immunity requirements for on-board battery systems, the battery management system (BMS) has an immunity limit of 3V / m (corresponding to a magnetic field strength of approximately 50μT) in the 30-1000MHz frequency band. Exceeding this limit requires additional anti-interference measures. The 50μT threshold in this solution directly corresponds to the immunity level in the national standard, ensuring that the testing process complies with electromagnetic compatibility specifications. 50μT is a critical value constrained by cell signal accuracy, motor radiation characteristics, measured data, and industry standards; it represents both the upper limit of tolerable interference and the optimal balance point for initiating enhanced sampling.

[0065] Actual tests were conducted on battery packs of 10 brands of electric scooters (each group included 100 samples under different operating conditions). When the interference intensity was ≤50μT, the pass rate of voltage data from 3 samples (error ≤1mV) was 98.3%. When the interference intensity was >50μT, the pass rate from 3 samples plummeted to 72.6%, while the pass rate from 7 samples + composite filtering could recover to 97.8%. 50μT is the intersection of the two pass rate curves. At this threshold, enhanced sampling can be initiated to achieve a pass rate improvement of more than 25% with the lowest sampling cost (adding 4 samples).

[0066] Specifically, multi-dimensional adaptive sampling employs seven consecutive samples to sample a type of battery cell. Electromagnetic interference generated by electric scooter motors exhibits significant randomness (e.g., high-frequency pulse interference) and periodicity (fluctuations of 100-300Hz related to motor speed). Extensive experimental verification shows that seven samples can cover at least two complete interference cycles (based on a maximum frequency of 300Hz, seven samples cover approximately 23ms, containing 6-7 interference pulses), thus offsetting the impact of random interference through data averaging. Fewer than five samples fail to cover a complete interference cycle, making the data susceptible to transient strong interference; more than nine samples result in data redundancy due to repeated interference patterns, hindering further improvement of the signal-to-noise ratio. Statistically, seven samples yield the fastest convergence speed in terms of standard deviation. When the sampling frequency is 3 times, the data stability is insufficient (95% confidence interval error > 5mV); although 5 sampling times can reduce the error to 3mV, it is still insufficient for strong interference scenarios near the motor (interference amplitude can reach 8-12mV); 7 sampling times, through a combination strategy of 3 outlier removals and 4 mean calculations, can stably control the error within 1mV, and the computational load is only 78% of that of 9 sampling times. Within the total test cycle, 7 sampling times can meet the accuracy requirements without affecting the timing of subsequent steps due to excessive computation time. Furthermore, the variational mode decomposition algorithm used in the scheme requires at least 5 data points to complete the effective separation of modal components, while the cell signal near the motor contains three components: "fundamental voltage + 3rd harmonic interference + random noise". 7 sampling times can provide sufficient degrees of freedom for the algorithm: the first 2 sampling times are used to initially determine the interference intensity, the middle 3 sampling times are used for mode decomposition, and the last 2 sampling times are used to verify the consistency of the decomposition results. If the number of samplings is less than 5, the algorithm cannot complete multimodal separation; if it is more than 7, it will lead to redundancy in the decomposition dimensions, increasing the computation time by more than 15%.

[0067] Comparative testing of 1000 electric scooter battery packs revealed that, under strong electromagnetic interference, the effective data retention rate (the percentage of usable data after removing outliers) reached 92% after 7 samplings, compared to 78% after 5 samplings and 93% after 9 samplings (only 1% higher than 7 samplings). However, 9 samplings increased the sampling time per cell from 1.2 seconds to 1.8 seconds. In a strong interference area containing 8 cells, the total time increased by 4.8 seconds, exceeding the 40-second time allocation threshold. Therefore, 7 samplings represent the optimal balance between accuracy, efficiency, and scenario adaptability.

[0068] Specifically, the variational mode decomposition algorithm processes the acquired signal as follows:

[0069] The acquired cell voltage signal (such as a fluctuation signal of around 3800mV) is first subjected to a 50Hz low-pass filter to remove power grid frequency interference and retain the effective frequency band of 0-10kHz (covering the main components of cell voltage changes and high-frequency interference from the motor). The preprocessed signal is input into the algorithm in discrete time series form, with a sampling frequency fixed at 10kHz (one data point is collected every 0.1ms).

[0070] Based on the signal characteristics of the electric scooter battery pack, three modal components (K=3) are preset: component 1 is the high-frequency interference mode (3-10kHz), corresponding to electromagnetic pulse interference from the motor; component 2 is the mid-frequency coupling mode (1-3kHz), corresponding to voltage fluctuations caused by temperature changes; and component 3 is the low-frequency effective mode (0-1kHz), corresponding to the actual voltage changes of the battery cell. A penalty factor is also set. =2000 (Balanced decomposition accuracy vs. computational efficiency), convergence tolerance (Ensure the decomposition results are stable).

[0071] The signal is transformed to the frequency domain using Fourier transform, and the objective function constrained by "the center frequencies of each modal component do not overlap" is constructed as follows:

[0072] ;

[0073] The constraints are:

[0074] ;

[0075] By minimizing the total energy of each modal component, the decomposed... Each modal component ( It exhibits sparsity (energy is concentrated in a specific frequency range). Among them, Indicates time Find the partial derivative. For convolution operations, It is the kernel function of the Hilbert transform, used to obtain the analytic form of the signal. To convert the signal to the center frequency The frequency domain as the reference, for The square of the norm represents the signal energy. The superposition of all modal components must equal the original signal f(t) to ensure that no useful information is lost during the decomposition process.

[0076] The center frequency and amplitude of each modal component are updated iteratively. After each iteration, the spectral overlap of each component is calculated. The iteration stops when the overlap is <5% (usually 20-30 iterations are required, taking about 0.5 seconds).

[0077] By analyzing the energy percentage, the low-frequency effective mode (component 3) with an energy percentage greater than 80% is retained, while the high-frequency interference mode (component 1) and the mid-frequency coupling mode (component 2) are removed to obtain the purified cell voltage signal.

[0078] The selected low-frequency effective modes are subjected to inverse Fourier transform to convert them back to time-domain signals. Then, missing data points are filled in by linear interpolation, and finally a smooth voltage curve (such as a stable fluctuation curve of 3800±1mV) is output. The reconstructed signal can be directly used for subsequent threshold determination. The total processing time is controlled within 1 second, shortening the test cycle.

[0079] Electromagnetic interference from electric scooters is characterized by its wide bandwidth, suddenness, and strong correlation with motor speed. The interference covers a frequency range of 1-10kHz. Variational mode decomposition (VMD) can effectively handle this multi-band separation. During rapid motor acceleration, the interference pulse lasts 0.5-2ms, and the algorithm's rapid iteration capability (30 iterations / 0.5 seconds) allows for real-time response. The interference frequency changes linearly with the motor speed (1000-3000rpm), and the adaptive adjustment characteristic of the VMD center frequency can be dynamically matched. Wavelet thresholding denoising truncates high-frequency coefficients using a preset threshold, but when the high-frequency pulse interference (3-10kHz) from the motor overlaps with the voltage fluctuations in the battery cell (1-3kHz), "mode aliasing" easily occurs, misjudging some effective voltage fluctuations as interference noise, leading to signal distortion (with an error of up to 2-3mV). Empirical Mode Decomposition (EMD) relies on selecting intrinsic mode functions (EMFs) from local extrema of the signal. When processing signals with significant impulse interference, it is prone to "endpoint effects" (edge ​​data distortion), and the number of decomposition iterations is uncertain (typically requiring 5-8 iterations), with each iteration taking approximately 2 seconds. Kalman filtering requires knowledge of the statistical characteristics of the interference noise (such as mean and variance). However, the interference intensity of electric scooter motors dynamically changes with rotational speed (e.g., interference increases 3 times during startup), and a fixed noise model leads to deterioration in filtering performance (error > 1.5mV). Compared to traditional algorithms such as wavelet thresholding, comparative EMD, and Kalman filtering, this approach is better suited to the strong electromagnetic interference scenario of electric scooter battery packs, providing high-quality and effective signals for subsequent threshold determination.

[0080] Specifically, the electromagnetic interference intensity at the location of the secondary battery cell is ≤50μT (far lower than in the strong interference zone), and the interference signal has little impact on voltage sampling (single sampling error ≤1mV). Experimental data shows that a single sampling is affected by random noise, and the error may reach 2-3mV (pass rate only 82%); three samplings can cancel out most of the random noise through averaging, and the error is stabilized within 0.5mV (pass rate increases to 98%); although the error of five or more samplings can be reduced to 0.3mV, it is only 0.2mV higher than three samplings, but it increases the sampling time of a single battery cell from 0.5 seconds to 1 second. In a battery pack containing multiple cells, the total time will increase significantly. Therefore, using three samplings can meet the accuracy requirement of "error ≤1mV" for the secondary battery cell, and also control the sampling time of a single group within a reasonable range, avoiding occupying the testing resources of cells in the strong interference zone. According to the normal distribution theory, when the sample size is ≥3, the standard error of the mean can be reduced to 58% of the single sampling error, which is sufficient to cover the noise characteristics of the secondary battery cell (mainly low-frequency temperature noise, with an approximately normal distribution). If the sample size is <3 (e.g., 2 samplings), the stability of the mean decreases significantly (the standard error is 71% of the single sampling error), and it cannot reliably offset random fluctuations; if the sample size is >3, although the error can be further reduced, the marginal benefit diminishes (the error reduction from 3 to 5 samplings is only 0.2mV), which does not conform to the principle of maximizing testing efficiency.

[0081] Specifically, although secondary battery cells have weak electromagnetic interference, they may exhibit temperature gradients due to differences in heat dissipation (temperature differences between adjacent cells can reach 2-5℃). For every 1℃ increase in temperature, the cell voltage will shift by approximately 0.5mV. If a simple arithmetic average is used, the voltage shift of high-temperature cells may be "diluted" by low-temperature cells, causing the overall mean to deviate from the true state (the error can reach 1-2mV). Dynamic weighted averaging increases the weight of high-temperature cells (the weight increases by 20% when the temperature difference is >2℃), making the final result closer to the actual voltage of high-temperature cells (experiments have verified that the temperature-induced error can be controlled within 0.3mV), solving the problem of insufficient adaptation of static averaging to temperature-sensitive cells. For secondary battery cells, there is no need to use 7 samplings and composite filtering in the strong interference area (the algorithm takes 1 second). Dynamic weighted averaging can achieve improved accuracy through simple weighted calculation (takes <0.1 seconds), and the weight adjustment rules (based on the temperature difference threshold) are easy to program and will not increase the processor load. In contrast, while using complex algorithms such as adaptive filtering can further optimize the results, it will increase the program complexity by 30%, which does not meet the "simple and reliable" requirements of industrial testing.

[0082] Furthermore, the specific details of the instantaneous state feature extraction sampling path planning are as follows:

[0083] After the test is initiated, all cells in the battery pack are pre-sampled once. Following pre-sampling, based on the high-precision initial data output from the environmentally coupled dynamic disturbance rejection sampling and the pre-sampling results, the "voltage jump rate" (i.e., the absolute value of the difference between the high-precision initial data and the pre-sampling results), representing the voltage change over a continuous 20ms, is calculated for each cell. Based on the calculation results, cells exhibiting instantaneous fluctuations (voltage jump rate greater than 5mV / 20ms), "voltage-stable cells" (voltage jump rate less than 1mV / 20ms), and "metastable cells" (voltage jump rate between 1mV / 20ms and 5mV / 20ms, inclusive) are categorized. Sampling paths are ordered in a two-dimensional order based on fluctuation intensity and physical location. Cells with higher fluctuation intensity are more unstable and require more frequent monitoring; therefore, for every 1mV / 20ms increase in fluctuation intensity, the sampling priority of that cell is increased by 30%. Furthermore, since "metastable cells" may fluctuate during operation and potentially transform into "instantaneously fluctuating cells" or "voltage-stable cells," the sampling path is automatically refreshed every 5 seconds based on newly acquired data to adapt to real-time changes in cell status. For voltage-stable cells, due to their smaller state changes, a "one-time accurate sampling and feature value locking" approach is adopted, eliminating the need for repeated sampling to save testing time.

[0084] Electric scooters are used in various scenarios, such as urban roads, hill climbing, and sudden braking. The performance of the battery pack's cells changes in real time under these different scenarios. Path planning based on historical data cannot adapt to these dynamic changes in a timely manner, easily leading to misjudgments of the cell's current state. However, planning the sampling path based on real-time voltage jump rate can accurately capture the instantaneous state of the cells, aligning with the industry's characteristics of rapid and dynamic battery pack testing, ensuring that the test results reflect the battery pack's current true state. By rationally allocating sampling resources, more testing time is devoted to monitoring unstable cells, improving sampling efficiency by 30% and ensuring that all cells can be collected with fine detail within the testing cycle. Simultaneously, it provides comprehensive and real-time cell state data for the self-evolving threshold pool and multi-parameter coupled decision-making, ensuring that subsequent decision-making processes are based on complete and accurate information.

[0085] Specifically, the sampling path planning process is as follows:

[0086] The three parameters obtained from pre-sampling—"voltage jump rate" (e.g., 8mV / 20ms), "electromagnetic interference intensity" (e.g., 60μT), and "physical location coordinates" (e.g., 3cm distance from the motor)—are quantized into a numerical matrix. Priority weights are then calculated. The priority calculation formula is: Priority ( = Baseline value (100) + Jump rate correction term ( ) + Position Correction Item ( ) × Interference intensity correction factor ( ) + State mutation correction term ( ). , , Voltage jump rate, As the reference jump rate, The priority increase is the percentage increase for every 1mV / 20ms (used to prioritize). (It becomes dimensionless), and 10 is the amplification factor. , This represents the distance between the battery cell and the motor. Median filtering is applied to 10 consecutive electromagnetic interference intensity monitoring values ​​to remove transient pulse interference, resulting in a stable interference intensity. , (used to) (revised to be dimensionless), when When ≤50μT, =1; when 50μT < When ≤100μT, Follow Linear growth; when When >100μT, =2, to avoid excessive amplification. Compare the voltage jump rate this time with the previous time to determine if the cell type has changed (e.g., from metastable state to transient dynamic). (New and old category benchmark priority: stable state = 100, metastable state = 300, instantaneous dynamics = 800) Dimensionless), the first time priority calculation is performed .

[0087] Specifically, the process of automatically refreshing the sampling path every 5 seconds based on newly collected data is as follows:

[0088] During the 5-second interval between two path refreshes, the sampling system performs high-frequency monitoring of high-priority cells according to the current path (e.g., cells with transient fluctuations are sampled once every 100ms, and cells with metastable states are sampled once every 2 seconds). Simultaneously, it performs a verification sampling of low-priority stable cells to obtain voltage data (for Class I cells, 5 valid values ​​are retained from 7 samples (excluding 2 outliers); for Class II cells, the result after averaging 3 samples with dynamic weights is retained, ensuring that each cell has at least 3 new voltage data points within 5 seconds (interval of 20ms)). Electromagnetic interference intensity is also measured (a miniature magnetic field sensor continuously collects data at a frequency of 1kHz, generating one median-filtered valid value every 50ms, accumulating 100 data points within 5 seconds, and finally taking the average as the final value). Temperature gradient (temperature data of adjacent cells, used to correct the dynamic weighted average). During refresh, the newly collected data is input into the priority calculation formula, and the priority is re-sorted to obtain a new sampling path.

[0089] Furthermore, the specific details of the self-evolving threshold pooling and multi-parameter coupled decision-making are as follows:

[0090] It receives comprehensive cell status data output during the instantaneous state feature extraction and sampling process, and constructs a three-dimensional threshold pool based on the three dimensions of individual, group, and environment.

[0091] The individual threshold is generated based on the voltage standard deviation of the first three tests of each cell (if the standard deviation is σ). The threshold is set as the range consisting of the mean plus three times the standard deviation and the mean minus three times the standard deviation, which can cover the voltage fluctuation range of the cell under normal conditions.

[0092] The group threshold is determined by real-time calculation of the voltage distribution entropy of all cells. The voltage distribution entropy can reflect the degree of voltage balance among cells. When the entropy value is greater than 0.8, it indicates that the voltage difference between cells is large, which is judged as group imbalance.

[0093] The environmental threshold divides the test environment temperature (range -10~45℃) into 5 intervals, each interval corresponding to an independent temperature-voltage coupling coefficient. For example, when the temperature is in the 30-35℃ range, since the increase in temperature will affect the chemical reaction rate of the battery cell, the voltage threshold will be reduced by 0.5mV for every 1℃ increase in temperature.

[0094] The judgment process employs a mechanism combining veto power and weighted voting. When a single parameter exceeds its individual threshold, a potential problem may exist, triggering a warning; when the group threshold or environmental threshold is exceeded, it indicates an abnormal overall state of the battery pack, and it is directly judged as a fault. The weights are dynamically allocated according to the following proportions: voltage 60% (voltage is a core parameter reflecting the cell state), temperature 30% (temperature has a significant impact on cell performance), and equipment status 10% (equipment operating status indirectly affects battery performance).

[0095] The performance of electric scooter battery packs varies significantly at different stages of use (e.g., new batteries vs. aged batteries). The voltage, capacity, and other parameters of new batteries differ markedly from those of aged batteries. Furthermore, the normal parameter ranges of battery packs also vary considerably under different environmental conditions, such as high and low temperatures. A three-dimensional threshold pool can dynamically adjust thresholds based on the individual state, group state, and environmental conditions of the battery pack. Multi-parameter coupled decision-making comprehensively considers the influence of various factors, avoiding the limitations of single-parameter judgments, making it suitable for the complex and variable testing scenarios of battery packs in this industry. The thresholds evolve in real-time with individual states and the environment, reducing the false positive rate to below 2%. This provides accurate anomaly judgment results for fault tree and instantaneous feature map fusion diagnosis, ensuring that subsequent diagnostic processes can proceed based on correct anomaly judgments.

[0096] Specifically, the construction process of the three-dimensional threshold pool is as follows:

[0097] In the initial stage of battery pack operation, high-precision voltage sensors are used to continuously sample each cell within the battery pack at extremely short time intervals (e.g., 2ms) for 100ms, thus acquiring 50 voltage data points. These data are arranged and stored sequentially according to cell number to construct an initial voltage dataset. Miniature magnetic field sensors are deployed at key locations inside the battery pack, such as near strong interference sources like motors and power converters. These sensors collect electromagnetic interference intensity data at a frequency of 1kHz, collecting 1000 data values ​​per second. Median filtering is applied to these data to remove outliers caused by transient pulses, and the mean value is then calculated as the representative value of the electromagnetic interference intensity at that location within that one second. High-precision temperature sensors are attached to key components such as the positive and negative terminals of the cells and the casing, collecting temperature data every 5 seconds. To reflect temperature differences between cells, the temperature difference between adjacent cells is calculated, constructing a temperature gradient dataset.

[0098] Individual thresholds are determined based on the historical stability characteristics of a single battery cell. Voltage data from the first three complete tests of each cell are extracted (each test includes 500 sampling points, covering static, dynamic, and load fluctuation scenarios), and the voltage sequence for each test is calculated. The average voltage of the first three tests is then calculated. and standard deviation Individual threshold upper limit = For example, if the average voltage of a certain battery cell in the first three tests is 3800mV, the standard deviation is... =2mV, then the individual threshold range is 3794mV-3806mV. The individual threshold reflects the "historical reference fluctuation range" of a single cell. When the real-time voltage exceeds this range, it indicates that there is a significant deviation between the current state and the historical stable state (which may be due to individual problems such as cell aging or abnormal internal resistance).

[0099] The group threshold is based on the uniformity characteristics of the cell group. Real-time voltage data of all cells within the battery pack are collected, and the group voltage distribution characteristics are calculated. The distribution entropy H is used to quantify the dispersion of the group voltage, and its calculation formula is as follows: ,in, This refers to the number of voltage intervals (divided into 0.5mV intervals). For a certain cell voltage to fall into the first The probability of an interval. The larger the distribution entropy, the more significant the voltage difference within the group. The group threshold is H=0.8 (determined through testing 1000 normal battery packs: the distribution entropy of a normal group ≤0.6, the distribution entropy of an unbalanced group ≥0.8, and 0.6-0.8 is the warning interval). When H>0.8, it is judged as a group imbalance. The group threshold reflects the "cooperative working state" of multiple battery cells. An excessive entropy value means that some cells may experience a deterioration in consistency (such as a sudden drop in the voltage of a certain cell leading to a dispersion in the group distribution), requiring a focus on investigating group coupling problems (such as series and parallel circuit faults).

[0100] The environmental threshold is based on the coupling characteristics between environmental interference and cell state. The ambient temperature is divided into five intervals, each corresponding to an independent temperature-voltage coupling coefficient k (unit: mV / ℃, representing the correction amount for the voltage threshold for every 1℃ change in temperature). Specifically, the intervals are: -10℃ to 1℃, k = +0.3; 1℃ to 12℃, k = +0.1; 12℃ to 23℃, k = 0; 23℃ to 34℃, k = -0.3; 34℃ to 45℃, k = -0.5. The electromagnetic interference intensity threshold is 50μT (consistent with the previous description). When the interference intensity > 50μT, the voltage threshold is corrected (e.g., for every 10μT exceeding 50μT, the voltage threshold range is widened by 0.5mV to offset measurement errors caused by interference). The environmental threshold reflects the "boundary of the influence of environmental interference on cell state." Individual / group thresholds are dynamically corrected through the coupling coefficient to avoid misjudgments caused by environmental fluctuations (such as high temperature or strong electromagnetic fields) (e.g., voltage naturally decreases at high temperatures, requiring a lower limit for individual thresholds to adapt to physical characteristics).

[0101] Through statistical analysis of 1000 battery pack anomaly cases, the weights of three dimensions were determined. Individual anomalies, accounting for the highest percentage (58%), are directly related to cell safety and thus have the highest weight. The individual dimension threshold is [not specified in the original text]. Anomalies caused by environmental disturbances account for 32% (such as voltage shifts due to high temperatures), requiring dynamic threshold adjustment. The weighting of these anomalies is secondary; environmental dimension thresholds are the next most important factor. Group imbalance is mostly caused by individual abnormalities or environmental disturbances (accounting for 10%), which is a derived feature with the lowest weight and is the threshold for the group dimension. .

[0102] Individual, group, and environmental thresholds are mapped to a three-dimensional space (X-axis: individual voltage deviation, Y-axis: group distribution entropy, Z-axis: environmental disturbance intensity). The thresholds for each dimension are weighted to form a comprehensive criterion, with the comprehensive deviation S = ×(Individual voltage deviation / Individual threshold range)+ ×(Environmental disturbance exceeding value / Environmental threshold)+ ×(population entropy / population threshold).

[0103] A single cell is the smallest functional unit of a battery pack. Its individual parameters, such as voltage and capacity, directly determine its performance (e.g., aging of a single cell can cause a sudden voltage drop). Voltage is the easiest core indicator to collect in real time (sampling frequency can reach 1kHz) and it has a strong correlation with capacity and state of health (SOH) (correlation coefficient > 0.9), making it more suitable as a benchmark for individual monitoring. A battery pack is a series / parallel system of multiple cells. An abnormality in a single cell can affect the overall state through circuit coupling (e.g., a sudden increase in the internal resistance of a single cell can cause other cells to be overloaded). The group dimension quantifies this "coordination" through distribution entropy, solving the blind spot of "individual cells being normal but the group being unbalanced". Cell performance (such as voltage temperature coefficient) is highly sensitive to the environment: for every 1°C increase in temperature, the voltage may deviate by 0.5mV; when electromagnetic interference is > 50μT, the sampling error will exceed the allowable range. The environmental dimension dynamically corrects the threshold through coupling coefficients to avoid "misjudging environmental influences as cell failures". The three-dimensional threshold pool formed by these three elements can effectively cover all aspects of the cell status, and redundant parameters are eliminated through cross-validation, making it a perfect fit for the monitoring needs of electric scooter battery packs. Therefore, it has become the optimal choice for constructing a threshold pool.

[0104] A comparison of 1000 sets of fault simulation experiments showed that the missed detection rate due to the missing individual dimension increased from 1.8% to 23% (mainly due to missed detection of latent faults in single cells); the misjudgment rate due to the missing group dimension increased from 1.8% to 17% (misjudging group imbalance as environmental interference); and the misjudgment rate due to the missing environmental dimension increased from 1.8% to 21% (misjudging voltage drop caused by high temperature as cell fault). However, the anomaly identification accuracy of the three-dimensional threshold pool (98.2%) was significantly higher than that of the individual threshold (82.5%), group threshold (76.3%), or two-dimensional threshold (90.1%), and the misjudgment rate was reduced to 1.8% (mainly for edge cases under extreme environments), meeting the high-precision monitoring requirements of battery packs.

[0105] Specifically, the mechanism combining "veto power with weighted voting" is as follows:

[0106] To address extreme situations that could lead to safety accidents or core performance failures, three "one-vote veto" conditions are set (triggering any one of them constitutes an anomaly). At the individual level, if the real-time voltage exceeds twice the individual threshold range (i.e., >μ+6σ or <μ-6σ), for example, if the individual threshold for a cell is 3794mV-3806mV, and the real-time voltage suddenly drops to 3780mV (exceeding the lower limit by 14mV, i.e., 7σ), it is considered a risk of internal short circuit within the cell, directly triggering an alarm. At the group level, if the group distribution entropy H > 1.2 (far exceeding the imbalance threshold of 0.8) and the duration is > 2 seconds, the cell group has experienced severe inconsistency deterioration (e.g., voltage deviation of 5 cells > 10mV), potentially leading to circuit overload, and is directly vetoed without weighted calculation. At the environmental level, if the electromagnetic interference intensity > 150μT (3 times the interference threshold of 50μT) or the temperature > 50℃ (exceeding the upper limit of the normal operating range by 5℃), the anomaly is detected. Strong interference may cause the sampling system to fail, and high temperature may cause thermal runaway. In both cases, the test must be interrupted immediately and the protection activated.

[0107] When no veto is triggered, a comprehensive score is calculated through a three-dimensional "weighted voting" to determine the cell status. Each dimension is divided into three levels according to the degree of deviation: "Normal (0 points)," "Warning (1 point)," and "Abnormal (2 points)," and the total score is calculated based on the dimension weights. In the individual dimension, voltage within (μ±3σ) is 0 points, (μ±3σ-μ±4σ) is 1 point, and (μ±4σ-μ±6σ) is 2 points; in the environmental dimension, interference intensity ≤50μT and temperature within the normal range is 0 points, interference 50-100μT or temperature 34-45℃ is 1 point, and interference 100-150μT or temperature 45-50℃ is 2 points; in the environmental dimension, distribution entropy H≤0.6 is 0 points, 0.6-0.8 is 1 point, and 0.8-1.2 is 2 points. The comprehensive score S = (individual score × 0.5) + (environmental score × 0.3) + (group score × 0.2). When S ≥ 1.5, it is judged as abnormal; 1.0-1.5 is a warning; S ≤ 1.0 is normal. For example, if a battery cell scores 1 point (voltage slightly exceeds 3σ), 1 point for the environment (temperature 36℃), and 0 points for the group, the comprehensive score = 0.5 + 0.3 + 0 = 0.8 points, which is judged as normal; if the individual score is 2 points and the environment score is 1 point, the comprehensive score = 1.0 + 0.3 = 1.3 points, which is judged as a warning.

[0108] First, a veto decision is made. If it is triggered, an abnormal result is output directly. If it is not triggered, a weighted vote is entered, the status is determined according to the score range, and the sampling strategy is adjusted according to the warning level (e.g., the sampling frequency is increased to 100ms / time in the warning state).

[0109] like Figure 3 As shown, the specific details of the fault tree and instantaneous feature map fusion diagnosis are as follows:

[0110] Based on the anomaly determination results of the self-evolving threshold pool and multi-parameter coupled decision output, diagnosis is performed using a bidirectional mapping method of "structured fault tree and real-time feature map".

[0111] The fault tree retains a two-layer structure: top event and direct features. The top event represents common battery pack fault types, such as "cell short circuit" and "cell overcharge". Each top event is associated with a specific direct feature. For example, the direct feature associated with the top event "cell short circuit" is "voltage less than 3000mV and temperature rise rate greater than 5℃ / s".

[0112] For each top event, an "instantaneous feature map" is constructed. Dynamic features such as voltage jump curves, temperature change slopes, and current fluctuation frequencies at the time of fault occurrence are transformed into quantifiable feature values ​​using a specific algorithm. For example, a voltage jump curvature greater than 0.5V / ms² can be used as a feature value for a cell short circuit.

[0113] During the diagnosis process, the suspected top event is first located by using the fault tree to determine the anomaly. Then, the dynamic features collected in the current test are compared with the spectral feature values ​​corresponding to the suspected top event. When the matching degree is greater than 85%, the fault is confirmed.

[0114] The types of faults in electric scooter battery packs are relatively concentrated, mainly including cell short circuits, overcharging, over-discharging, and overheating, and each fault has obvious characteristic manifestations. Fault trees can quickly narrow down the investigation scope based on anomalies, locking in a small range of suspected fault types. Real-time feature maps can accurately match dynamic features at the time of fault occurrence without complex probability calculations, providing diagnostic results in a short time, meeting the efficiency requirements of short-cycle testing in the industry. The diagnostic response time is less than 500ms, compressing the anomaly location time to 2 seconds, providing clear fault indications for instantaneous tracing of atomic data nodes, enabling subsequent tracing work to be targeted and effective.

[0115] Specifically, the diagnostic method using "structured fault tree and real-time feature map" is as follows:

[0116] Using "battery pack malfunction" as the top event, a fault tree (based on over 1000 historical fault cases) is constructed using a three-tiered structure of "system-subsystem-component". The first-level sub-events are categorized into individual malfunctions (58%), group imbalances (10%), and environmental disturbances (32%), corresponding to the three-dimensional threshold pool. Second-level sub-events are further subdivided into individual malfunctions such as "voltage drop", "capacity decay", and "sudden increase in internal resistance"; group imbalances are subdivided into "consistency deterioration" and "circuit coupling failure"; and environmental disturbances are subdivided into "high-temperature runaway" and "strong electromagnetic interference". Each second-level sub-event is further broken down into quantifiable fault modes; for example, "voltage drop" corresponds to "internal short circuit", "lack connection", and "electrolyte drying". Each basic event is associated with specific characteristic parameters (e.g., an internal short circuit will cause "voltage drop > 50mV within 50ms + temperature rise > 5℃").

[0117] Based on real-time monitoring data from a three-dimensional threshold pool (updated every 50ms), a multi-dimensional feature map is constructed. The individual feature layer includes single-cell voltage jump rate (mV / 20ms), temperature rise rate (℃ / s), and voltage-temperature coupling coefficient (mV / ℃); the group feature layer includes group voltage distribution entropy, maximum voltage difference (mV), and equalization current response speed (A / s); and the environmental feature layer includes electromagnetic interference intensity (μT), temperature gradient (℃ / cm), and interference-voltage error coefficient (mV / μT). The feature map is stored in matrix form; for example, the feature vector at a certain moment is: [8mV / 20ms (individual jump rate), 0.9 (group entropy), 120μT (interference intensity),...].

[0118] Forward mapping (feature to fault): Match the real-time feature map with the basic event feature library of the fault tree and calculate the similarity score. For example, if the real-time feature is "voltage drops by 60mV in 50ms + temperature rises by 3℃ / s", and the feature matching degree with "internal short circuit" reaches 92%, then the basic event can be preliminarily located.

[0119] Reverse mapping (fault to feature): For the top 3 basic events with the highest matching degree, trace back the logical chain of the fault tree to verify whether the features of its parent events are satisfied. For example, after locating "pole ear virtual connection", it is necessary to verify whether "group entropy < 0.8" (because a single event does not affect group equilibrium). If it is not satisfied, the fault is excluded and rematching is performed.

[0120] Weights are assigned based on the importance of the features (e.g., voltage jump rate weight 0.4, temperature rise rate weight 0.3). A matching score ≥85% is considered a confirmed case, while 60%-85% is considered a suspected case, requiring further sampling verification.

[0121] Battery pack failures often manifest as "one cause, multiple effects" or "multiple causes, one effect": for example, both "loose connection of the electrode tabs" and "internal short circuit" can lead to a voltage drop, but the former results in a gradual temperature change, while the latter causes a sudden temperature rise. Structured fault trees, through explicit causal logic decomposition and multi-dimensional matching of real-time feature maps, can accurately distinguish similar faults (experimental data shows that the misjudgment rate of similar faults has decreased from 35% with single feature matching to below 5%). The bidirectional mapping not only diagnoses faults but also outputs the "scope of influence" of the fault tree (e.g., a basic event leading to "group imbalance turning into circuit overload") and the "repair guidance" of the feature map (e.g., "loose connection of the electrode tabs" requires adjusting "contact pressure + applying conductive paste"), directly connecting to subsequent repair strategies. The bidirectional mapping of "structured fault trees and real-time feature maps," through deep integration of logical decomposition and real-time features, solves core problems in battery pack fault diagnosis such as "ambiguity, difficulty in tracing the source, and poor adaptability." Its accuracy, real-time performance, and closed-loop capability have irreplaceable advantages in the electric scooter battery pack scenario.

[0122] Furthermore, the specific process of instantaneous tracing of atomized data nodes is as follows:

[0123] Based on the fault indication output from the fusion diagnosis of the fault tree and the instantaneous feature map, the "data atom and step fingerprint" binding mechanism is invoked.

[0124] Each sampled data is broken down into an atomic unit consisting of a timestamp (accurate to 10μs), a physical location code, and a feature value. The timestamp ensures the time accuracy of the data, the physical location code clarifies the cell location corresponding to the data, and the feature value is the specific data of that sample.

[0125] A unique "step fingerprint" is generated for each of the 20 test steps. It consists of a step number, a device status code, and an environmental parameter hash value. The step number identifies the specific test step, the device status code reflects the operating status of the device during the test, and the environmental parameter hash value is an encrypted processing of the environmental parameters to ensure the uniqueness and integrity of the environmental parameters.

[0126] Atomized data and step fingerprints are linked via blockchain-style hashing. This association method is tamper-proof, ensuring the authenticity and reliability of the data. When an anomaly occurs, the atomic unit corresponding to the anomaly parameter is retrieved, and the step fingerprint is matched in reverse through the atomic unit to pinpoint the specific sampling time, physical location, and environmental state within one second.

[0127] In the production and repair of electric scooters, it is crucial to quickly pinpoint the source of malfunctions to improve production efficiency and repair quality. Atomized data units can accurately identify the smallest data unit related to the fault, avoiding interference from a large amount of irrelevant data. Blockchain-style hash association ensures the immutability of data, preventing tampering during storage or transmission, guaranteeing the accuracy of traceability results, and meeting industry requirements for accuracy and efficiency in fault tracing. This improves anomaly tracing efficiency by 90%, eliminating the need to trace historical data, and directly providing precise evidence for subsequent repair work. It clarifies the specific time, location, and environment of the fault occurrence, helping repair personnel quickly find the problem and completing the closed loop of the entire testing process.

[0128] Specifically, the binding mechanism for "data atoms and step fingerprints" is as follows:

[0129] A data atom refers to the smallest indivisible data unit, which must meet three characteristics: "single source, clear semantics, and unique timestamp." This includes individual data atoms (e.g., the voltage value of a battery cell at 10:00:00.000 is 3800mV (including sensor ID and sampling accuracy ±0.5mV)), feature data atoms (e.g., the jump rate of 2mV / 20ms calculated based on 3 voltage samples (including calculation algorithm version and outlier removal markers)), and diagnostic data atoms (e.g., the matching score of "internal short circuit" is 92% (including feature map ID and fault tree node number)). Each data atom carries a 128-bit unique identifier (UUID) containing metadata such as generation time (accurate to μs), device number, and data type. A step fingerprint is a unique code for an operation step. It uses a hash algorithm to convert key parameters of the step (such as sampling frequency, algorithm threshold, operator) into a 64-bit hash value (fingerprint), covering 12 core steps in the entire process. For example, the acquisition step (such as "7 samplings of a type of battery cell + variational mode decomposition", the fingerprint includes the number of samplings, algorithm parameters (penalty factor 2000), sensor model), the processing step (such as "wavelet threshold denoising (threshold 0.8mV)", the fingerprint includes threshold type (hard threshold / soft threshold) and number of decomposition layers), the diagnostic step (such as "bidirectional mapping matching (weight 0.5 / 0.3 / 0.2)", the fingerprint includes matching algorithm version and confidence threshold (85%), etc.

[0130] For each raw data atom (such as a voltage value) generated by the sensor, the system automatically extracts the fingerprint of the current sampling step (e.g., "Class I cell sampling_20231001_V1.2"), and binds it to the fingerprint via a blockchain-like structure using a UUID (written to a local encrypted database, immutable). When an algorithm (such as variational mode decomposition) generates a feature data atom (e.g., the energy percentage of mode component 3 is 82%), it binds the step fingerprint of that algorithm (including the number of iterations 25 and the convergence tolerance 1e-6), and simultaneously associates it with the input raw data atom UUID, forming a traceability chain from "raw data to feature data". The diagnostic result (e.g., "abs are loosely connected") serves as a diagnostic data atom, bound to the fingerprint of the bidirectional mapping step (including forward / backward matching parameters), and associated with the feature data atom UUID, completely recording the entire chain from "raw data to feature to diagnosis".

[0131] When the parameters of a step change (e.g., the sampling frequency is adjusted from 1kHz to 2kHz), the step fingerprint is recalculated (the hash value changes), and all data atoms generated based on that step are marked as "derived version". For example, if the threshold of the "wavelet denoising" step is adjusted from 0.8mV to 1.0mV, the newly generated feature data atom UUID will have "_V2" appended to it, and will be associated with both the old and new step fingerprints to clarify the source of the data difference. The system records all changes through a version tree and supports backtracking queries (e.g., comparing the diagnostic results between versions V1 and V2).

[0132] The binding relationship is verified for integrity every 5 minutes by reverse-engineering the step fingerprint using the data atom UUID. If the fingerprint hash value does not match the step parameters (e.g., data has been tampered with), an alarm is triggered immediately. Any data atom can be traced back to its original step level by level using its UUID. For example, a diagnostic result of "internal short circuit" can be used to trace back to the original step. Feature data atoms (jump rate 8mV / 20ms) Raw data atoms (3 voltage samples) Fingerprint of sampling steps (7 samplings + algorithm V2.1).

[0133] The "data atom and step fingerprint" binding mechanism constructs an immutable traceability chain by deeply binding the smallest data unit with the operational steps, solving the core pain points of "unreliable data, unreproducible processes, and insufficient compliance" in battery pack testing. Its lightweight, precise, and dynamic adaptability have irreplaceable advantages in the electric scooter battery pack scenario, and are a key support for the transformation from "experience-driven" to "data-driven" diagnostics.

[0134] Before testing, device connection is required. After connection, the system initializes and performs a self-test to ensure all hardware is functioning correctly. The interface displays 20 test steps, all initially yellow, indicating the test has not yet started. Simultaneously, the interface displays the current magnetic field strength and temperature gradient, providing initial environmental parameters for environmentally coupled dynamic disturbance rejection sampling, allowing testers to intuitively understand the initial state of the test environment. Sampling and verification then proceed. The system samples according to the sampling path extracted from instantaneous state features, with the interface updating the dynamic changes in voltage and temperature in real time, directly reflecting the changing trends of cell parameters. The self-evolving threshold pool, fault tree, and instantaneous feature graph run in the background, processing and diagnosing the collected data in real time. After the test cycle, the interface displays the final result as "qualified" or "faulty." For battery packs determined to be faulty, the fault item is directly associated with an atomized data node. Testers can click on the fault item to view the instantaneous feature graph, which clearly shows the parameter changes at the time of the fault. The interface does not have historical data association fields; it only clearly presents the final test results and the source tracing entry, avoiding interference from irrelevant information.

[0135] This solution, through seamless integration and innovative design across all stages, completely eliminates reliance on historical data and probabilistic models. It achieves core data accuracy (voltage error ≤1mV, temperature deviation ≤0.5℃) exceeding industry standards, providing a more accurate reflection of the battery pack's condition. Anomaly detection time is reduced from 15 seconds to 2 seconds, perfectly adapting to the fast-paced demands of production lines and significantly improving production efficiency.

[0136] Example 1:

[0137] High-temperature testing was conducted. A sealed, constant-temperature chamber was used to simulate the high-temperature environment of an electric scooter exposed to direct sunlight in summer or during prolonged uphill climbing. The temperature was set at four gradients: 40℃, 45℃, 50℃, and 55℃, with each gradient stabilized for 30 minutes. Simultaneously, an electromagnetic interference simulator was used to reproduce the high-frequency interference of 100-150μT from the motor within the chamber. The battery pack was fixed within a fixture simulating the scooter frame. The distance between the battery cells and the center of the motor output shaft was set according to the aforementioned standard (3cm, 5cm, etc.) to ensure that the correlation between physical location and interference intensity conformed to the "Class I / Class II battery cell" classification logic. The dynamic operating conditions of the electric scooter under "urban commuting + uphill climbing" were simulated, with a charging current of 1.5A and a discharging current of 3A. The duration of a single charge-discharge cycle was controlled within 20 minutes to avoid exceeding the test cycle limits.

[0138] For Class I battery cells within 5cm of the motor, seven consecutive samples are initiated. Variational mode decomposition (preset with three modal components, K=3) separates high-frequency interference from the motor in the 3-10kHz range. Wavelet thresholding (threshold 0.8mV) is then applied to remove three outliers, and the average of the four samples is taken to ensure a voltage error ≤1mV. For Class II battery cells, three dynamic weighted average samples are used. When the temperature difference between adjacent cells is >2℃ (e.g., some cells at 50℃ have a 4℃ temperature difference due to heat dissipation differences), the weight of the high-temperature cell is increased by 20% (calculation formula: final value = (high-temperature cell voltage × 1.2 + low-temperature cell voltage × 0.8) / 2) to offset temperature gradient deviation.

[0139] After pre-sampling, the voltage jump rate is calculated, and "instantaneous fluctuation cells" are marked (e.g., the jump rate of a cell reaches 8mV / 20ms at 55℃). The sampling path is then calculated according to the priority calculation formula.

[0140] Priority P = 100 + (8-1) × 30% × 10 + (5-3) × 20 × (150 / 50) = 100 + 21 + 40 × 3 = 241, prioritizing high-frequency sampling (once every 100ms). The path is refreshed every 5 seconds. Combining the newly acquired electromagnetic interference intensity (e.g., interference rises to 150μT at 55℃) and temperature data (cell temperature reaches 52℃), the sampling frequency of the metastable cell (jump rate 3mV / 20ms) is adjusted to 2 seconds / time.

[0141] Individual threshold: Based on the data from the first three tests, the average value of a certain cell is μ=3800mV, σ=2mV, and the individual threshold range is 3794-3806mV. Combining the coupling coefficients of k=-0.5mV / ℃ in the 34-45℃ range and k=-0.8mV / ℃ above 45℃ in the environmental threshold, the lower limit of the individual threshold at 55℃ is corrected to 3794-(55-25)×0.8=3794-24=3770mV.

[0142] Group threshold: Real-time calculation of voltage distribution entropy. When the distribution entropy H of a certain group of cells at 55℃ is 0.9 (>0.8), a "group imbalance" warning is triggered.

[0143] Decision-making mechanism: A certain battery cell has a voltage of 3765mV (lower than the corrected individual threshold) and a group entropy of 0.9. Although it did not trigger the "one-vote veto" (not exceeding 6σ), the weighted voting score S = (individual score 2 × 0.5) + (environmental score 2 × 0.3) + (group score 2 × 0.2) = 1 + 0.6 + 0.4 = 2.0 ≥ 1.5, and is judged as abnormal.

[0144] The abnormal data was matched with the second-level sub-event "voltage drop" in the fault tree, corresponding to the basic event "electrolyte drying out" (characteristic parameters: voltage drop > 50mV within 50ms + temperature rise rate > 5℃ / s). In the real-time feature graph, the cell's voltage jump curvature was 0.6V / ms² and the temperature rise rate was 6℃ / s, with a 91% (≥85%) match with the "electrolyte drying out" feature, confirming the fault.

[0145] Retrieve the data atoms of the faulty cell (timestamp 10:15:30.000000, voltage 3765mV, sensor ID=C1), associate the step fingerprint (sampling steps: 7 samplings of a type of cell + variational mode decomposition V2.1), confirm that the fault occurred at the 15th minute of the 55℃ discharge stage, and locate the C1 cell 3cm away from the motor.

[0146] Example 2:

[0147] Low-temperature testing was conducted. A low-temperature constant-temperature chamber simulated a winter outdoor environment, with temperatures set at -20℃, -30℃, and -40℃. Each temperature gradient was stabilized for 60 minutes (due to lower cell activity at low temperatures, a longer stabilization time is required), and the electromagnetic interference intensity was ≤50μT. The cell temperature sensor accuracy was ±0.5℃, and the sampling timing avoided strong interference during the 0-200ms of motor startup. Simulation of "low-temperature start-up + smooth driving" was performed, with a discharge current of 1A (low load), and charging using a "0.5A pre-charge + 1A standard charge" mode (to address the issue of increased electrolyte viscosity).

[0148] Primarily using Class II battery cells (distance from the motor > 5cm), the data was sampled using a three-time dynamic weighted average. At -40℃, the temperature difference between adjacent cells was 1℃ (< 2℃), with each cell having a weight of 1.0. The voltage data, after three averages, had an error ≤ 1mV. Environmental data recorded included: electromagnetic interference 30μT, cell temperature -40℃, and motor speed 800rpm. The data was labeled in the format "C5-6.0-30--40-3205".

[0149] The pre-sampling shows a voltage jump rate of 2mV / 20ms (metastable state), with a priority P = 100 + (2-1) × 30% × 10 + (5-6) × 20 × (30 / 50) = 100 + 3 - 4 × 0.6 = 100 + 3 - 2.4 = 100.6. The sampling frequency is set to 2 seconds / time. At a 5-second refresh rate, the jump rate increases to 4mV / 20ms at -40℃, and the priority is adjusted to 100 + (4-1) × 3 + (5-6) × 20 × 0.6 = 100 + 9 - 12 = 97, still in a metastable state.

[0150] Individual threshold: The average value of the first 3 tests for a certain battery cell is μ=3200mV, σ=3mV, and the threshold range is 3191-3209mV. Combined with the environmental threshold below -10℃, k=+0.3mV / ℃, the upper limit at -40℃ is corrected to 3209+(25+40)×0.3=3209+19.5=3228.5mV.

[0151] The population distribution entropy H = 0.5 (≤ 0.6) indicates that the population is in equilibrium.

[0152] Decision-making mechanism: The cell voltage is 3220mV (within the corrected threshold), and the weighted score S = (0×0.5) + (1×0.3) + (0×0.2) = 0.3 ≤ 1.0, which is considered normal.

[0153] When a certain battery cell is charged at -40℃, the voltage jump rate is 6mV / 20ms. Matching the fault tree sub-event "charging difficulty", the characteristic spectrum shows that the current response speed is 0.2A / s (< normal 0.5A / s), which matches the "electrolyte freezing" fault by 88%, and is diagnosed as a suspected fault.

[0154] Data atoms (timestamp 14:30:20.000000, voltage 3180mV) were associated with step fingerprints, confirming that the fault was caused by the surge in electrolyte viscosity at -40℃, which hindered ion migration.

[0155] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for optimizing the state of a battery pack for electric scooters, characterized in that, The method includes: Based on the physical location coordinates of the cells within the battery pack, the cells are divided into Class I cells and Class II cells. The sampling method for Class I cells is determined based on the electromagnetic interference intensity. Both Class I and Class II cells are sampled to obtain high-precision initial data. All cells in the battery pack are pre-sampled, and the voltage jump rate of each cell is calculated by combining the pre-sampled data with high-precision initial data. The cells are then classified according to their voltage jump rate. Input the physical location coordinates of the battery cell, electromagnetic interference intensity, voltage sag rate and battery cell type into the priority calculation formula to calculate the priority of each battery cell. Based on the priority of each battery cell, a sampling path is planned, and battery cell status data is collected in real time according to the sampling path. The cell status data is input into a three-dimensional threshold pool containing individual thresholds, group thresholds, and environmental thresholds, and anomaly determination is made by using a veto and weighted voting method. Input the anomaly determination results into the fault diagnosis model to obtain the fault location.

2. The method according to claim 1, characterized in that, The sampling methods for the first type of battery cell and the second type of battery cell are as follows: Battery cells with a distance of less than or equal to 5cm between their geometric center and the center of the motor output shaft are classified as Class I battery cells, and the rest are classified as Class II battery cells. When the electromagnetic interference intensity is greater than the interference intensity threshold, multi-dimensional adaptive sampling is used to perform seven consecutive samplings on the first type of battery cell. When the electromagnetic interference intensity is less than or equal to the interference intensity threshold, the sampling method is the same as that for the second type of battery cell. Three consecutive samplings were used to sample the two types of battery cells and the dynamic weighted mean was calculated. The acquired signals are processed using a variational mode decomposition algorithm.

3. The method according to claim 2, characterized in that, The specific process of signal processing by the variational mode decomposition algorithm is as follows: The collected cell voltage signal is filtered by a 50Hz low-pass filter to remove power grid interference and retain the effective frequency band of 0-10kHz. The preprocessed signal is input into the algorithm in the form of a discrete time series and the sampling frequency is fixed at 10kHz. Based on the signal characteristics of the battery pack, three modal components are preset, corresponding to the high-frequency interference mode, the mid-frequency coupling mode, and the low-frequency effective mode, respectively. The signal is converted to the frequency domain by Fourier transform, and an objective function is constructed with the constraint that the superposition of all modal components equals the original signal. The center frequency and amplitude of each modal component are updated iteratively. After each iteration, the spectral overlap of each component is calculated. The iteration is stopped when the overlap is less than 5%. Through energy proportion analysis, low-frequency effective modes with an energy proportion greater than 80% are retained, while high-frequency interference modes and mid-frequency coupling modes are eliminated. The selected low-frequency effective modes are subjected to inverse Fourier transform to convert them back to time domain signals. Then, missing data points are filled in by linear interpolation to output a smooth voltage curve.

4. The method according to claim 3, characterized in that, The specific process for classifying these categories is as follows: The absolute value of subtracting the high-precision initial data from the pre-sampling result is the voltage jump rate. If the voltage jump rate is greater than 5mV / 20ms, the cell is a transient fluctuation cell; if the voltage jump rate is between 1mV / 20ms and 5mV / 20ms, the cell is a metastable cell; if the voltage jump rate is less than 1mV / 20ms, the cell is a voltage-stable cell.

5. The method according to claim 4, characterized in that, The specific process of sampling path planning is as follows: The parameters of the physical location coordinates of the battery cell, electromagnetic interference intensity, and voltage sag rate are quantified into a numerical matrix. The numerical matrix and the battery cell type are then input into the priority calculation formula to obtain the priority of each battery cell. Based on the priority, the sampling order of the battery cells is determined from largest to smallest, and the sampling path is determined. Every 5 seconds, the sampling path is replanned based on the newly collected data. During the 5-second interval between two path refreshes, the sampling system performs high-frequency monitoring of high-priority cells according to the current path, and performs one verification sampling of low-priority stable cells to obtain voltage data. For the second-class cells, the result after three dynamic weight averages is retained. The newly collected data is input into the priority calculation formula to re-sort the priorities and obtain a new sampling path.

6. The method according to claim 5, characterized in that, The process of constructing the three-dimensional threshold pool is as follows: Extract the voltage data from the first three complete tests of each cell, calculate the voltage sequence of each test, calculate the voltage mean and standard deviation of the first three tests, and determine the individual threshold. Collect real-time voltage data of all cells in the battery pack, calculate the group voltage distribution entropy, and determine the group threshold; The ambient temperature is divided into 5 intervals, each with an independent temperature-voltage coupling coefficient. When the electromagnetic interference intensity is greater than the interference intensity threshold, the voltage threshold is corrected to determine the environmental threshold. By statistically analyzing failure cases, the weights of three dimensions are determined, and individual thresholds, group thresholds, and environmental thresholds are mapped to a three-dimensional space. The comprehensive deviation is obtained by weighted summation of the thresholds for each dimension.

7. The method according to claim 6, characterized in that, The specific processes for the veto and the weighted voting are as follows: At the individual level, if the real-time voltage exceeds twice the individual threshold range, it is considered abnormal. At the group level, if the group distribution entropy is greater than 1.2 and the duration is greater than 2 seconds, it is considered abnormal. At the environmental level, if the electromagnetic interference intensity is greater than 150μT or the temperature is greater than 50℃, it is considered abnormal. When no veto is triggered, each dimension is classified into levels according to the degree of deviation, and a comprehensive score is calculated. When the comprehensive score is greater than or equal to 1.5, it is considered abnormal. A comprehensive score greater than 1.0 and less than 1.5 is considered a warning. A comprehensive score less than or equal to 1.0 is considered normal.

8. The method according to claim 7, characterized in that, The fault location acquisition process is as follows: Taking battery pack malfunction as the top event, the system is broken down into a three-level structure of system, subsystem, and component to form a fault tree containing several basic events. The fault tree contains first-level sub-events and second-level sub-events. Each second-level sub-event is further broken down into quantifiable fault modes, and each basic event is associated with specific characteristic parameters. Real-time feature maps are constructed based on real-time monitoring data from a three-dimensional threshold pool. The real-time feature map is matched with the basic event feature library of the fault tree, and a similarity score is calculated. Weights are assigned according to the importance of the features. A similarity score of 85% or higher is considered a confirmed case, while a similarity score between 60% and 85% is considered a suspected case, requiring further sampling verification. For the top 3 basic events with the highest similarity scores, backtrack the logical chain of the fault tree to verify whether the characteristics of their parent events are satisfied.

9. The method according to claim 8, characterized in that, After obtaining the fault location, the source is traced through atomic data nodes, as shown in the following process: Each sampled data is broken down into atomic units consisting of timestamps, physical location codes, and feature values; A unique step fingerprint is generated for each test step, consisting of the step number, device status code, and environmental parameter hash value. Atomized units and step fingerprints are linked through blockchain-style hashing; When an anomaly occurs, the atomic unit corresponding to the anomaly parameter is retrieved, and the step fingerprint is reversed to locate the specific sampling time, physical location, and environmental state.

10. A battery pack state optimization testing system for electric scooters, characterized in that, The system is used to perform a battery pack state optimization test method for electric scooters as described in any one of claims 1-9.

Images