Battery state of health detection method and system

Abstract

The application relates to the technical field of battery detection, and discloses a battery health state detection method and system, which comprises the following steps: collecting voltage and current signals during battery charging and discharging cycles, obtaining low-frequency and high-frequency components through wavelet transformation, and calculating energy distribution to determine abnormal characteristics; if the number of abnormal characteristics is less than a preset threshold, reconstructing the signal based on low-frequency component sparse representation, otherwise, constructing a state classification tree based on high-frequency component amplitude variation and a preset threshold, adjusting the traversal strategy and modifying the sampling frequency according to the number of abnormal characteristics; and finally generating a health state index. The system comprises a feature extraction module, a state analysis module and a health evaluation module. Through wavelet transformation, sparse representation and dynamic adjustment strategy, the precision and efficiency of battery health state detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery detection, in particular to a battery health state detection method and system. BACKGROUND

[0002] With the rapid development of new energy technology, batteries as energy storage devices are widely used in electric vehicles, energy storage systems and other fields. The health state (State of Health, SOH) of the battery is a key indicator for evaluating the performance and remaining life of the battery. Accurate detection of the health state of the battery is of great significance for ensuring the safety of the battery, optimizing battery management and improving system reliability.

[0003] Traditional battery health state detection methods are mainly based on electrochemical models and empirical formulas, such as measuring the voltage, current, temperature and other parameters of the battery, and combining the ampere-hour integral method, equivalent circuit model, etc. However, these methods have many limitations. First, the internal electrochemical reaction and physical characteristics of the battery during charging and discharging are complex and changeable, and traditional models are difficult to accurately depict the dynamic characteristics of the battery, resulting in low detection accuracy. Second, abnormal features generated during battery aging process (such as internal short circuit, plate corrosion, active material shedding, etc.) often have nonlinearity and nonstationarity, and traditional methods are difficult to effectively capture and identify these subtle changes, which may cause missed detection or false detection.

[0004] Existing detection methods usually use traditional means such as Fourier transform in signal processing, which has limited analysis capability for non-stationary signals and cannot meet the needs of time-frequency characteristic analysis of battery signals. At the same time, when facing complex abnormal feature recognition and state classification problems, traditional methods lack adaptive adjustment strategies and are difficult to optimize the detection process according to the real-time changes of the battery state, resulting in insufficient detection efficiency and reliability.

[0005] With the development of intelligent algorithms and signal processing technologies, wavelet transform, sparse representation, machine learning and other technologies have been gradually applied in the field of battery health state detection. Wavelet transform can perform multi-resolution analysis on signals, effectively extracting the low-frequency trend and high-frequency details of signals, and is suitable for non-stationary signal processing; sparse representation theory can reconstruct signals through sparse dictionary, retaining the main features of signals; machine learning models can automatically identify abnormal features and classify states through data training. However, the current application of these technologies in battery health state detection still has some problems, such as how to reasonably combine the time-frequency analysis capability of wavelet transform and the signal reconstruction advantage of sparse representation, how to construct an efficient abnormal feature recognition model and state classification tree, and how to dynamically adjust the detection strategy according to the real-time data in the detection process, etc. These problems need to be solved.

[0006] Therefore, there is an urgent need for a battery health status detection method and system that can effectively handle non-stationary battery signals, accurately identify abnormal features, and adaptively adjust detection strategies to improve the accuracy, reliability, and efficiency of battery health status detection and meet the urgent needs of the new energy field for battery management. Summary of the Invention

[0007] The purpose of this invention is to provide a battery health status detection method and system to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a battery health status detection method, the method comprising:

[0009] When the battery is in a charge-discharge cycle, the battery's voltage and current signals are collected, and wavelet transform processing is performed on the collected signals to obtain the low-frequency and high-frequency components of the signals; the energy distribution of the low-frequency components and the energy distribution of the high-frequency components are calculated, and the abnormal characteristics of the battery are determined based on the energy distribution.

[0010] If the number of abnormal features is less than a preset threshold, the original signal of the battery is reconstructed based on the low-frequency component using sparse representation; otherwise, a state classification tree is constructed based on the amplitude change of the high-frequency component and the preset threshold. If the number of levels in the classification tree is greater than 2, the traversal strategy of the state classification tree is adjusted according to the number of identified abnormal features and the total number of abnormal features. When the adjustment conditions are met, the sampling frequency of signal acquisition is modified.

[0011] The reconstructed battery signal is acquired, and the corresponding health status index is generated in the health status assessment system.

[0012] Preferably, determining the abnormal characteristics of the battery based on the energy distribution specifically involves:

[0013] The energy distribution of the low-frequency component is normalized using the energy distribution of the high-frequency component.

[0014] The normalized energy distribution is input into a pre-established anomaly feature recognition model, and the number of anomaly features is output.

[0015] Preferably, the construction of the state classification tree based on the amplitude changes of high-frequency components and a preset threshold specifically involves:

[0016] Calculate the ratio of the maximum amplitude of the high-frequency component to a preset threshold. If the number of abnormal features is greater than or equal to 50% of the maximum amplitude, the number of branches in the state classification tree is the ratio; otherwise, the number of branches is the integer part of the ratio.

[0017] Preferably, the strategy for adjusting the traversal of the state classification tree based on the number of identified abnormal features and the total number of abnormal features is as follows:

[0018] If, after traversing a certain level of the state classification tree, the total number of abnormal features minus the number of identified abnormal features is less than a preset threshold, then the adjustment condition is triggered.

[0019] Alternatively, after traversing the first 50% of the state classification tree, if the total number of abnormal features minus the number of identified abnormal features is less than a preset threshold, then the adjustment condition is triggered.

[0020] Preferably, the modified sampling frequency of the signal acquisition is specifically as follows:

[0021] Adjust the sampling frequency to 1.5 times the original frequency;

[0022] Alternatively, merge branches at adjacent levels in the state classification tree and adjust the segmentation strategy of sampling frequency based on the merging result.

[0023] Preferably, the present invention further includes a battery health status detection system, the system comprising:

[0024] The feature extraction module is used to collect the battery's voltage and current signals when the battery is in a charge-discharge cycle, perform wavelet transform processing on the collected signals to obtain the low-frequency and high-frequency components of the signals, calculate the energy distribution of the low-frequency components and the energy distribution of the high-frequency components, and determine the abnormal characteristics of the battery based on the energy distribution.

[0025] The state analysis module is used to reconstruct the original signal of the battery based on the low-frequency component using sparse representation if the number of abnormal features is less than a preset threshold; otherwise, it constructs a state classification tree based on the amplitude change of the high-frequency component and the preset threshold. If the number of levels in the classification tree is greater than 2, it adjusts the traversal strategy of the state classification tree according to the number of identified abnormal features and the total number of abnormal features. When the adjustment conditions are met, it modifies the sampling frequency of signal acquisition.

[0026] The health assessment module is used to acquire the reconstructed battery signals and generate corresponding health status indicators in the health status assessment system.

[0027] Preferably, determining the abnormal characteristics of the battery based on the energy distribution specifically involves:

[0028] The energy distribution of the low-frequency component is normalized using the energy distribution of the high-frequency component.

[0029] The normalized energy distribution is input into a pre-established anomaly feature recognition model, and the number of anomaly features is output.

[0030] Preferably, the construction of the state classification tree based on the amplitude changes of high-frequency components and a preset threshold specifically involves:

[0031] Calculate the ratio of the maximum amplitude of the high-frequency component to a preset threshold. If the number of abnormal features is greater than or equal to 50% of the maximum amplitude, the number of branches in the state classification tree is the ratio; otherwise, the number of branches is the integer part of the ratio.

[0032] Preferably, the strategy for adjusting the traversal of the state classification tree based on the number of identified abnormal features and the total number of abnormal features is as follows:

[0033] If, after traversing a certain level of the state classification tree, the total number of abnormal features minus the number of identified abnormal features is less than a preset threshold, then the adjustment condition is triggered.

[0034] Alternatively, after traversing the first 50% of the state classification tree, if the total number of abnormal features minus the number of identified abnormal features is less than a preset threshold, then the adjustment condition is triggered.

[0035] Preferably, the modified sampling frequency of the signal acquisition is specifically as follows:

[0036] Adjust the sampling frequency to 1.5 times the original frequency;

[0037] Alternatively, merge branches at adjacent levels in the state classification tree and adjust the segmentation strategy of sampling frequency based on the merging result.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] In signal processing, when the battery is in a charge-discharge cycle, the acquired voltage and current signals are processed using wavelet transform to decompose the signals into low-frequency and high-frequency components. The low-frequency components reflect the overall trend and main energy distribution of the signal, while the high-frequency components contain the signal's detailed features and abrupt changes. By calculating the energy distribution of the low-frequency and high-frequency components separately, the time-frequency characteristics of the battery signal can be captured more comprehensively, providing a rich information foundation for subsequent anomaly identification and signal reconstruction. Compared to traditional methods such as Fourier transform, wavelet transform has a stronger ability to analyze non-stationary signals and can more accurately reveal the dynamic changes of the battery during the charge-discharge process.

[0040] In anomaly feature identification, the energy distribution of low-frequency components is normalized using the energy distribution of high-frequency components. The normalized energy distribution is then input into a pre-established anomaly feature identification model, which outputs the number of anomaly features. This normalization process eliminates the influence of different signal scales, making the energy distribution features more comparable and robust. Through learning and training on a large amount of historical data, the anomaly feature identification model can automatically extract key features related to battery health status, achieving accurate identification of battery anomalies. This improves the sensitivity and accuracy of anomaly feature detection, reducing missed and false detections.

[0041] In signal reconstruction and state analysis, different processing strategies are adopted based on the number of abnormal features. If the number of abnormal features is less than a preset threshold, it indicates that the battery state is relatively stable. In this case, the original battery signal is reconstructed using sparse representation based on low-frequency components. Sparse representation can compress and reconstruct the signal through a sparse dictionary, reducing the amount of data while retaining the main features of the signal, thus improving the efficiency and accuracy of signal processing. If the number of abnormal features is greater than or equal to the preset threshold, it indicates that the battery may have more serious anomalies. In this case, a state classification tree is constructed based on the amplitude changes of high-frequency components and the preset threshold. The number of branches in the state classification tree is determined by calculating the ratio of the maximum amplitude of the high-frequency components to the preset threshold, allowing the structure of the classification tree to adapt to the characteristic changes of the battery signal, thus improving the accuracy and flexibility of state classification.

[0042] Regarding the adjustment of the detection strategy, when the number of levels in the classification tree is greater than 2, the traversal strategy of the state classification tree is adjusted based on the number of identified abnormal features and the total number of abnormal features. If, after traversing a certain level or the first 50% of levels, the number of unidentified abnormal features is less than a preset threshold, an adjustment condition is triggered, modifying the sampling frequency of signal acquisition. This dynamic adjustment strategy can optimize the detection process based on real-time data during the detection process. When the detection of abnormal features is close to completion, the sampling frequency is adjusted (e.g., adjusting the sampling frequency to 1.5 times the original frequency, or merging adjacent level branches and adjusting the sampling frequency segmentation strategy) to improve detection efficiency while ensuring detection accuracy and reduce unnecessary consumption of computing resources.

[0043] In terms of health status assessment, the reconstructed battery signals are acquired, and corresponding health status indicators are generated in the health status assessment system. These indicators can comprehensively and accurately reflect the health status of the battery, providing a reliable basis for decision-making in the battery management system. This helps to achieve real-time monitoring, lifespan prediction, and optimized management of the battery, thereby improving the safety and economy of battery use. Attached Figure Description

[0044] Figure 1 This is a schematic diagram illustrating the working principle of the battery health status detection method described in this invention.

[0045] Figure 2 A flowchart for determining anomaly characteristics based on energy distribution;

[0046] Figure 3 Flowchart for modifying the signal acquisition sampling frequency;

[0047] Figure 4 Flowchart for determining abnormal features in the feature extraction module. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Please see Figures 1-4 The battery health status detection method involved in this invention has the following specific implementation steps:

[0050] The battery's voltage and current signals are collected, and wavelet transform is performed on the collected signals to decompose them into low-frequency and high-frequency components. Wavelet transform can effectively extract different frequency components from the signal. The low-frequency components mainly reflect the overall trend and energy distribution of the signal, while the high-frequency components correspond to the detailed changes and abnormal features in the signal.

[0051] The energy distributions of the low-frequency and high-frequency components are calculated, and the abnormal characteristics of the battery are determined based on these energy distributions. The energy distribution can be calculated by integrating the squares of the signal amplitudes of the corresponding frequency components. The process of determining abnormal characteristics involves analyzing the energy distribution to identify feature points that differ from the normal state.

[0052] Determine the relationship between the number of abnormal features and the preset threshold:

[0053] If the number of abnormal features is less than a preset threshold, it indicates that the battery state is relatively stable. In this case, the original battery signal is reconstructed based on the low-frequency components using a sparse representation method. Sparse representation achieves approximate reconstruction of the original signal by finding the sparse representation coefficients of the low-frequency components under a specific dictionary, thereby removing the interference of high-frequency noise and retaining the main energy features.

[0054] If the number of abnormal features is greater than or equal to a preset threshold, it indicates that the battery may have a significant abnormality. In this case, a state classification tree is constructed based on the amplitude changes of high-frequency components and the preset threshold. If the number of levels in the constructed state classification tree is greater than 2, the traversal strategy of the state classification tree is further adjusted according to the number of identified abnormal features and the total number of abnormal features. When the adjustment conditions are met, the sampling frequency of signal acquisition is modified.

[0055] The reconstructed battery signal is acquired and input into the health status assessment system, which generates a corresponding health status indicator that is used to intuitively reflect the battery's health status.

[0056] Example 1: In the battery health status detection process, the specific implementation method for determining abnormal battery characteristics based on energy distribution is as follows: When the battery is in a charge-discharge cycle, the battery voltage and current signals are first acquired through signal acquisition devices such as sensors. These signals contain rich information about the battery's operation, including both low-frequency components reflecting the basic working state of the battery and high-frequency components that may characterize battery abnormalities. After acquiring the signals, wavelet transform processing is required. As a time-frequency analysis tool, wavelet transform can decompose the original signal into components of different frequency bands, thereby achieving multi-resolution analysis of the signal. Through wavelet transform, the original voltage and current signals are decomposed into low-frequency components and high-frequency components. The low-frequency components mainly reflect the overall trend and energy distribution of the signal, while the high-frequency components focus on reflecting detailed changes and abrupt changes in the signal. These abrupt changes are often associated with abnormal battery characteristics.

[0057] After completing the wavelet transform decomposition of the signal, the next step is to calculate the energy distribution of the low-frequency and high-frequency components. The energy distribution calculation is based on the signal amplitude. For both low-frequency and high-frequency components, their energy can be obtained by integrating or summing the squares of the signal amplitude within their respective frequency bands. The specific calculation method can be chosen according to the signal type and processing requirements. For example, for discrete signal sequences, the energy can be approximated by accumulating the squares of the amplitude at each sampling point. The calculated energy distributions of the low-frequency and high-frequency components can reflect the energy consumption and distribution of the battery in different frequency bands, providing a data foundation for subsequent anomaly feature analysis.

[0058] After obtaining the energy distributions of the low-frequency and high-frequency components, the energy distribution of the low-frequency component needs to be normalized using the energy distribution of the high-frequency component. The purpose of normalization is to eliminate the influence of energy scale differences between the components, making them comparable. Specific normalization methods can include dividing the energy distribution of the low-frequency component by the sum of the energy distributions of the high-frequency components, or using other suitable normalization factors to ensure that the normalized energy distribution values ​​are within a reasonable range. Through normalization, the energy distributions of the low-frequency and high-frequency components can be converted into relative energy distributions, highlighting the proportional relationship between them, thus facilitating the analysis of the correlation between changes in energy distribution and abnormal battery characteristics.

[0059] After normalization, the normalized energy distribution needs to be input into a pre-established anomaly feature recognition model to output the number of anomaly features. The anomaly feature recognition model is trained on a large amount of historical data. Its construction process requires collecting battery signal data under different health states and preprocessing this data, including signal acquisition, wavelet transform decomposition, energy distribution calculation, and normalization, to obtain the corresponding normalized energy distribution data as input samples. Simultaneously, the number of anomaly features corresponding to each sample needs to be manually labeled or determined through other reliable methods, serving as the model's output label.

[0060] During the model training phase, machine learning algorithms, such as neural networks and support vector machines, can be used to construct anomaly recognition models. Taking a neural network as an example, it can contain an input layer, hidden layers, and an output layer. The number of nodes in the input layer corresponds to the dimension of the normalized energy distribution data, and the number of nodes in the output layer represents the predicted number of anomaly features. By inputting a large amount of normalized energy distribution data into the neural network and comparing it with the corresponding anomaly feature quantity labels, the backpropagation algorithm is used to adjust the weights and bias parameters of the neural network, minimizing the error between the model's predicted output and the actual labels, thereby achieving model training and optimization.

[0061] In practical applications, when new normalized energy distribution data is input, the model extracts features and recognizes patterns based on the parameters and structure obtained during training. Ultimately, it outputs the number of anomalous features corresponding to that set of energy distribution data. It is important to note that the training data for the anomalous feature recognition model needs to be sufficiently diverse and representative, covering battery data of different types, service lives, and operating conditions to ensure the model's generalization ability and accuracy in real-world applications.

[0062] Throughout the processing, the accuracy and stability of signal acquisition are crucial. The selected sensor must meet the requirements of battery detection, accurately capture minute changes in voltage and current, and maintain reliable operation during battery charge-discharge cycles. Simultaneously, anti-interference measures must be implemented during signal transmission to prevent external electromagnetic interference and other factors from affecting the quality of the acquired signal, ensuring that subsequent wavelet transform processing and energy distribution calculations are based on real and valid signal data.

[0063] In the wavelet transform processing stage, it is necessary to rationally select the wavelet basis functions and the number of decomposition levels. Different wavelet basis functions have different time-frequency characteristics and are suitable for different types of signal analysis. Therefore, it is necessary to select an appropriate wavelet basis function based on the characteristics of the battery signal to ensure effective decomposition of low-frequency and high-frequency components. The choice of the number of decomposition levels also affects the signal decomposition effect. Too few decomposition levels may lead to incomplete separation of low-frequency and high-frequency components, making it impossible to accurately extract the energy distribution; too many decomposition levels may introduce excessive detail noise, increasing computational complexity. Therefore, it is necessary to determine the optimal wavelet basis functions and the number of decomposition levels through experiments and analysis to achieve the best decomposition effect on the battery signal.

[0064] When calculating energy distribution, the signal duration and sampling frequency must be considered. Energy distribution is calculated based on the signal within a certain time window, the size of which needs to be appropriately set according to the battery charge-discharge cycle period and the timescale of abnormal characteristics. If the time window is too small, it may fail to capture sufficient energy change information; if the time window is too large, it may smooth out some transient anomalies. Simultaneously, the sampling frequency also affects the accuracy of energy distribution calculation. Higher sampling frequencies provide denser signal sample points but increase the workload of data processing; lower sampling frequencies may lose some crucial details. Therefore, a trade-off must be struck between computational accuracy and efficiency to select an appropriate time window and sampling frequency.

[0065] During the normalization process, it is important to prevent the denominator from being zero. When the sum of the energy distributions of the high-frequency components is zero, the normalization process cannot be performed. Therefore, it is necessary to check the energy distribution of the high-frequency components before processing. If the sum is zero, measures such as re-acquiring the signal, adjusting the wavelet transform parameters, or using other normalization methods can be taken to ensure the smooth progress of the normalization process.

[0066] Example 2: In the state analysis stage of battery health status detection, the specific process of constructing a state classification tree based on the amplitude changes of high-frequency components and a preset threshold is as follows: When the number of abnormal features determined through preprocessing is greater than or equal to the preset threshold, it indicates that the battery may have a significant abnormal condition, and a state classification tree needs to be constructed further based on the amplitude characteristics of high-frequency components for more detailed analysis. First, the amplitude of the high-frequency components needs to be quantitatively analyzed. The core is to calculate the ratio of the maximum amplitude of the high-frequency components to the preset threshold. This preset threshold is a reference value pre-set based on multiple dimensions such as battery type, specifications, historical operating data, and industry standards, and is used to measure whether the amplitude of the high-frequency components exceeds the normal fluctuation range. For example, for a certain type of lithium-ion battery, historical data statistics show that the amplitude of the high-frequency components during normal operation usually does not exceed 8mV, so the preset threshold can be set to 8mV. If the maximum amplitude of the high-frequency components in actual detection is 12mV, then the ratio is 12 / 8 = 1.5.

[0067] After calculating the ratio, the number of abnormal features needs to be compared with 50% of the maximum amplitude to determine the correlation between the abnormal features and the amplitude changes of the high-frequency components. This 50% of the maximum amplitude is not a simple numerical calculation, but an empirical threshold derived from battery fault mechanism analysis, used to reflect the potential impact weight of amplitude changes on abnormal features. For example, if the maximum amplitude is 10mV, 50% is 5mV. If the number of abnormal features is greater than or equal to 5 (i.e., a proportion corresponding to 5mV), then the occurrence of abnormal features is considered highly correlated with the amplitude changes of the high-frequency components. In this case, the number of branches in the state classification tree is directly set to the calculated ratio. Assuming the ratio is 2.3, considering that the number of branches in the classification tree must be an integer, the ratio can be rounded according to the actual situation (e.g., rounding to the nearest integer or rounding up). If the original value of the ratio is directly used as a reference for the number of branches, it may need to be adjusted to an integer number of branches based on engineering practice, for example, adjusting 2.3 to 2 or 3 branches, which can be determined according to the classification tree construction rules.

[0068] If the number of outliers is less than 50% of the maximum amplitude, it indicates that the contribution of high-frequency component amplitude changes to outliers is relatively low. In this case, the number of branches in the state classification tree is the integer part of the ratio. For example, if the ratio is 2.6 and the number of outliers is less than 1.3 (i.e., the threshold for the number of outliers corresponding to 50% of the maximum amplitude), then the number of branches is the integer part 2. The logic behind this method of determining the number of branches is that when the correlation between amplitude changes and outliers is weak, the classification tree structure is simplified to avoid overfitting and improve classification efficiency.

[0069] When constructing a state classification tree, it is necessary to clearly define the tree's hierarchical structure and branching logic. Each level of the state classification tree corresponds to a judgment of the battery state, and the branches represent different state paths. For example, the first level can be set to determine whether the amplitude of the high-frequency component exceeds a preset threshold. If it does, it enters one branch; otherwise, it enters another branch. The next level can further refine the judgment based on ratios or other derived parameters (such as amplitude change rate). The dynamic adjustment mechanism of the number of branches allows the classification tree to adaptively adjust its structure according to the relationship between the number of abnormal features detected in real time and amplitude changes, thereby more accurately mapping the actual state differences of the battery.

[0070] In constructing the classification tree, the accuracy of the data preprocessing stage must also be considered. The extraction of high-frequency component amplitudes requires a precise signal acquisition and processing workflow, including anti-aliasing filtering and analog-to-digital conversion accuracy control. If noise interference exists during signal acquisition, it may cause deviations in the calculation of high-frequency component amplitudes, thus affecting the accuracy of the number of branches in the classification tree. Therefore, during the signal acquisition stage, high signal-to-noise ratio sensors and anti-interference circuit designs are necessary to ensure that the acquired voltage and current signals accurately reflect the actual operating state of the battery.

[0071] A dynamic update mechanism for the preset threshold is also crucial. As battery usage time increases and environmental conditions change, the amplitude range of its high-frequency components during normal operation may drift. Therefore, the preset threshold needs to be calibrated periodically based on the latest historical data. For example, it can be set that after every 100 charge-discharge cycles, the system automatically recalculates the preset threshold based on the normal signal data during that period to adapt to the effects of battery performance degradation or environmental changes.

[0072] In the hierarchical design of the classification tree, if the number of levels in the initially constructed classification tree is greater than 2 (such as 3 or more), the strategy needs to be adjusted in subsequent traversal processes based on the identification of abnormal features. However, this part falls under the scope of other embodiments, and this section focuses on the construction logic of the number of branches. The algorithm for constructing the classification tree can adopt classic decision tree algorithms (such as ID3, C4.5, etc.), and select the optimal splitting attribute through indicators such as information gain and gain ratio to ensure that the branches at each level can separate battery data of different states to the greatest extent.

[0073] It's important to note that building a state classification tree is not a one-time process, but rather a dynamic optimization process. After the initial construction, the tree structure can be pruned or expanded based on subsequent detection data to improve the model's generalization ability. For example, if the sample data volume under a certain branch is found to be too small, that branch can be removed through pruning to avoid classification bias caused by data sparsity; if new abnormal patterns are discovered, branches can be added at the corresponding levels to refine the classification.

[0074] In practical applications, the amplitude changes of high-frequency components may exhibit nonlinear characteristics. Therefore, when calculating the ratio, nonlinear transformation methods such as logarithmic transformation and exponential transformation can be considered to better fit the relationship between amplitude changes and the number of anomalous features. For example, when the amplitude change exceeds a certain range, its influence on anomalous features may increase exponentially. In this case, nonlinear transformation can make the ratio calculation more consistent with actual physical laws.

[0075] Example 3: In battery health status detection, the specific method for adjusting the state classification tree traversal strategy based on the number of identified abnormal features and the total number of abnormal features is as follows: When the state classification tree is completed and the number of levels is greater than 2, the dynamic relationship between the number of identified abnormal features and the total number of abnormal features needs to be monitored in real time during the traversal process to trigger corresponding adjustment conditions and optimize the traversal path. The entire process involves progress judgment of traversal levels, difference calculation, condition triggering logic, and strategy adjustment mechanism, which requires refined design in conjunction with the real-time processing requirements of battery signals.

[0076] The process of traversing the state classification tree is a layer-by-layer decision-making process, with each level's judgment node corresponding to different feature attributes or threshold conditions. For example, the first level might make branch judgments based on the ratio of the maximum amplitude of a high-frequency component to a preset threshold, the second level might further subdivide based on the energy distribution differences in a certain frequency band, and subsequent levels would progressively delve deeper to uncover more specific abnormal features. During the traversal, the system selects the appropriate branch based on the judgment result of the current node and accumulates the number of identified abnormal features. The total number of abnormal features refers to the total number of possible abnormal features of the battery initially determined through preliminary energy distribution analysis and other steps. This value is output by the abnormal feature recognition model at the start of detection or preset manually (if the scenario is known).

[0077] The first scenario triggering the adjustment condition is as follows: After traversing a certain level of the state classification tree, calculate the difference between the total number of anomalous features and the number of identified anomalous features, and compare this difference with a preset threshold. Here, "a certain level" refers to any complete level in the classification tree, such as level 3 or level 5. The specific level to be traversed for judgment can be determined based on the tree structure design or real-time data processing requirements. The preset threshold is a small positive integer (such as 1, 2, or 3) used to characterize the acceptable range of the remaining unidentified anomalous features. For example, assuming the total number of anomalous features is 8 and the preset threshold is 2, after traversing level 2, 6 anomalous features have been identified, and the difference is 2 (equal to the preset threshold). At this point, it is considered that the number of remaining unidentified anomalous features is at a low level, and the current traversal path may be redundant or inefficient, requiring the adjustment condition to be triggered.

[0078] The second trigger condition for adjustment is as follows: After traversing the first 50% of the state classification tree, the difference between the total number of abnormal features and the number of identified abnormal features is calculated and compared with a preset threshold. The calculation of "first 50% of the levels" is based on the total number of levels in the classification tree. For example, if the classification tree has 6 levels, then the first 50% of the levels is the first 3 levels; if the total number of levels is odd (e.g., 7 levels), then the first 50% of the levels can be rounded up or down (e.g., the first 3 or the first 4 levels, the specific rules need to be clearly defined in the system design). For example, if the total number of levels is 8, after traversing the first 4 levels, if the total number of abnormal features is 10, and 8 abnormal features have been identified, the difference is 2 (less than the preset threshold of 3), then the adjustment condition is triggered. The design logic of this condition is that if the identification of abnormal features is nearly completed when traversing half of the levels, it means that the judgment of subsequent levels may not need to be executed completely along the original path. The path can be optimized in advance by adjusting the traversal strategy, reducing the consumption of computing resources.

[0079] Once the adjustment condition is triggered, the system needs to adjust the traversal strategy according to preset rules. Specific adjustment methods include, but are not limited to, the following: First, changing the traversal order, such as switching from depth-first traversal to breadth-first traversal, or skipping certain branches that have been proven irrelevant to the remaining anomaly features; second, dynamically merging adjacent branches, merging them into a single path to simplify subsequent judgments when the anomaly features corresponding to several branches have been sufficiently identified; and third, adjusting the threshold parameters of judgment nodes, such as increasing or decreasing the judgment threshold in subsequent levels to more sensitively or leniently identify the remaining anomaly features. The selection of these adjustment methods must be based on the structural characteristics of the classification tree and the real-time data characteristics. For example, if the remaining anomaly features are mainly concentrated in a specific branch, the traversal order or parameters of that branch should be adjusted first.

[0080] During implementation, the following key technical points should be noted: First, the real-time statistics and synchronization of the number of abnormal features. The number of identified abnormal features needs to be updated promptly after each node traversal, ensuring consistency with the calculation benchmark of the total number of abnormal features. This requires the system to have efficient data storage and real-time computing capabilities, such as storing the current traversal progress and feature count data through a caching mechanism to avoid judgment deviations due to data delays. Second, the reasonable setting of the preset threshold. The preset threshold needs to comprehensively consider the accuracy requirements and computational efficiency of battery detection. If the threshold is too small, it may cause frequent triggering of adjustment conditions, increasing the system burden; if the threshold is too large, it may not be able to optimize the traversal path in time, affecting the detection speed. It is recommended to determine the initial threshold through historical data simulation or engineering experience, and provide a user-configurable interface for subsequent optimization.

[0081] Furthermore, the hierarchical division of the classification tree must match the complexity of the anomaly features. If the battery anomaly features are relatively simple, the number of levels in the classification tree may be small, and the judgments of the first 50% of levels may not fully reflect the feature distribution. If the anomaly features are complex and diverse, and the number of levels is large, then it is necessary to ensure that the first 50% of levels can capture the main feature differences. Therefore, when constructing a classification tree, the hierarchical structure should be reasonably allocated through feature importance analysis (such as Gini coefficient, information gain ratio, etc.), so that higher levels prioritize the processing of anomaly features with high impact, while lower levels process detailed differences.

[0082] During the adjustment of the traversal strategy, it is also necessary to avoid missing out on abnormal features due to the adjustment. For example, when skipping a branch, it is necessary to ensure through logical judgment that the abnormal features corresponding to that branch have been fully identified or are indeed unrelated to the remaining features. This can be achieved by adding a "safety check" mechanism to the classification tree node. That is, when deciding to skip a branch, first verify whether there is a probability of unidentified abnormal features in that branch. If the probability exceeds a certain threshold (such as 5%), then the traversal path of that branch is retained.

[0083] It's important to note that adjusting the traversal strategy is not a one-time operation, but a dynamic, iterative process. Each time the adjustment condition is triggered and strategy optimization is completed, the system continues traversing subsequent levels, monitoring new difference changes in real time. If the adjustment condition is met again, the adjustment logic can be repeated. This dynamic optimization mechanism allows the classification tree traversal process to constantly adapt to real-time changes in battery state, improving the efficiency and accuracy of anomaly feature identification.

[0084] In engineering implementation, the traversal and adjustment of the state classification tree can be achieved through software algorithms, such as using a recursive function to implement depth-first traversal. After each recursive call, the number of identified abnormal features is updated, and the recursive parameters are modified when conditions are met (such as skipping certain child nodes). For embedded systems with limited computing resources, the time and space complexity of the algorithm must also be considered to avoid system response delays caused by adjustments to the traversal strategy.

[0085] Example 4: During battery health status detection, the specific implementation methods for modifying the sampling frequency of signal acquisition when adjustment conditions are met include the following two cases:

[0086] 1. Adjust the sampling frequency to a fixed multiple of the original frequency.

[0087] When adjustment conditions are triggered (such as meeting the difference requirement for the number of abnormal features during state classification tree traversal), the sampling frequency can be adjusted to 1.5 times the original frequency. Original frequency This refers to the default sampling frequency set in the initial stage of detection. Its value needs to be determined based on the battery type, charge / discharge rate, and signal characteristic frequency, and usually needs to satisfy the Nyquist sampling theorem, i.e. (in This refers to the highest effective frequency component in the signal. For example, if the original sampling frequency is 100Hz (corresponding to the signal component that can be effectively acquired up to 50Hz), after adjustment it becomes... At this point, signal components up to 75Hz can be collected, thus enabling the capture of detailed changes at higher frequencies, making it suitable for scenarios requiring precise analysis of high-frequency anomalies.

[0088] This adjustment method relies on the hardware configurability of the signal acquisition module, such as using an analog-to-digital converter (ADC) with a programmable sampling rate. On the software level, the system triggers a sampling rate switching instruction based on detected adjustment conditions. This instruction modifies the ADC's control register via the hardware driver, reconfiguring the sampling clock frequency. When adjusting the sampling frequency, it is crucial to ensure the continuity of signal acquisition to avoid data loss or acquisition misalignment due to frequency switching. To this end, the current sampling period can be processed before the switch, and a short buffer period (such as acquiring several cycles of signal) can be implemented after the switch to stabilize the sampling timing at the new frequency.

[0089] Furthermore, it is necessary to consider the potential increase in data volume that may result from increasing the sampling frequency. For example, if 100 data points are collected per second at the original frequency, and 150 data points are collected per second after adjustment, the data volume increases by 50%, which places higher demands on the data storage and subsequent processing modules' computing power. Therefore, the hardware's processing bandwidth needs to be evaluated during system design, or data compression algorithms (such as differential coding, wavelet compression, etc.) should be adopted to reduce data transmission and storage pressure.

[0090] II. Segmented Adjustment of Sampling Frequency Based on Branch Merging of State Classification Tree

[0091] The second method for modifying the sampling frequency is to merge branches at adjacent levels in the state classification tree and adjust the segmentation strategy of the sampling frequency based on the merging result. The specific steps are as follows:

[0092] Branch merging logic: Merging branches at adjacent levels refers to merging branches at adjacent levels in the state classification tree (such as the first level). Layer and First Branches with similar characteristics (e.g., those in a layer) can be merged. Similarity can be determined based on the amplitude range of the high-frequency components corresponding to the branches, the energy distribution pattern, or the type of abnormal features. For example, if the amplitude changes of two adjacent branches are both less than 1.2 times the preset threshold, and the abnormal feature type is "slight voltage fluctuation", they can be merged into a composite branch.

[0093] Segmentation strategy design: After merging branches, the sampling frequency is divided into several segmented intervals based on the number and characteristic distribution of the remaining branches. Assume that after merging... There are 10 effective branches, which can divide the sampling frequency into 1000 branches. Each segment corresponds to a different sampling frequency range. For example, if three branches remain after merging, corresponding to "normal state," "minor abnormality," and "serious abnormality" respectively, then three sampling frequency segments can be set:

[0094] Segment 1 (Normal State): Using the original frequency This meets basic detection accuracy requirements and reduces data processing volume;

[0095] Segment 2 (Minor Anomaly): Using a medium frequency It is used to capture feature changes corresponding to minor anomalies;

[0096] Segment 3 (Severe Anomaly): High Frequency Utilization High-resolution acquisition is performed to capture rapidly changing features of severe anomalies.

[0097] Dynamic switching mechanism: The system automatically selects the corresponding sampling frequency segment based on the current branch state. For example, when the classification tree traversal result enters the "severe anomaly" branch, a high-frequency sampling segment is triggered, and the signal is collected at 1.8 times the original frequency; if it returns to the "normal state" branch, it switches back to the original frequency. This segmentation strategy can ensure detection accuracy while avoiding resource waste caused by using high-frequency sampling throughout the process.

[0098] When implementing branch merging, explicit merging rules must be established. For example, defining merging thresholds. When the difference between the characteristic parameters (such as the mean amplitude, variance of energy distribution, etc.) of two adjacent branches is less than At that time, a merging operation is performed. The selection of characteristic parameters needs to be related to the abnormal mechanism of the battery. For example, for lithium-ion batteries, the amplitude standard deviation of high-frequency components and energy entropy can be selected as similarity metrics.

[0099] The segmented adjustment of the sampling frequency needs to be synchronized in real time with the traversal process of the state classification tree. To this end, sampling frequency configuration information can be embedded in each node of the classification tree. When a node is reached, the system reads the segmentation parameters corresponding to that node and sends a frequency switching command to the signal acquisition module. Simultaneously, a frequency switching anti-jitter mechanism needs to be designed to prevent frequent sampling frequency switching due to brief branch fluctuations, which could affect signal stability. For example, a minimum hold time can be set. Once a branch is entered, the sampling frequency must be maintained at at least The duration is then determined, and a decision on whether to switch is made based on the results of subsequent iterations.

[0100] Key technical points and implementation details

[0101] ① Hardware compatibility: Whether adjusting by a fixed multiple or in segments, the signal acquisition hardware must support a programmable sampling rate. Common implementation methods include:

[0102] The sampling clock can be dynamically modified via software programming using an acquisition module equipped with a digital signal processor (DSP) or a field-programmable gate array (FPGA).

[0103] A multi-rate sampling circuit is used to switch frequencies by switching different clock sources (such as crystal oscillators) using an analog switch.

[0104] ② Anti-aliasing processing: When the sampling frequency is increased, the cutoff frequency of the anti-aliasing filter needs to be adjusted accordingly to prevent high-frequency noise from entering the acquisition channel. For example, when the original frequency is 100Hz, the cutoff frequency of the anti-aliasing filter is set to 50Hz; after adjusting to 150Hz, the cutoff frequency needs to be increased to 75Hz to meet the sampling theorem requirements.

[0105] ③ Data Synchronization and Calibration: After modifying the sampling frequency, the acquired data needs to be timestamped to ensure that signal data at different frequencies are aligned on the time axis. This can be achieved through a global clock synchronization mechanism or by embedding frequency identification information in the data frame.

[0106] ④ Computing resource allocation: Data processing at high sampling frequencies requires higher computing power. Pipeline processing, parallel computing (such as GPU acceleration), or edge computing architecture can be adopted to complete some signal preprocessing (such as wavelet transform) at the acquisition end, reducing the amount of data transmitted to the main processor.

[0107] Applicable Scenarios and Selection Logic for the Two Adjustment Methods

[0108] Fixed-multiplier adjustment is suitable for scenarios where the distribution of abnormal features is relatively uniform and a global improvement in detection accuracy is required. For example, when the number of abnormal features detected for the first time is close to the threshold, a comprehensive investigation can be carried out by uniformly increasing the sampling frequency.

[0109] The segmented adjustment after branch merging is suitable for scenarios where the abnormal features have a clear hierarchical structure, such as when the state classification tree has clearly distinguished different types of abnormalities with different degrees of severity. In this case, the sampling frequency can be dynamically adjusted according to the branch status to achieve "on-demand sampling" and optimize resource utilization efficiency.

[0110] In practical applications, the two methods can be used in combination. For example, the overall sampling frequency can be increased by a fixed factor first, and then the sampling segments can be further subdivided based on the branch merging results of the classification tree to form a multi-level optimized sampling strategy. Regardless of the method used, the modification of the sampling frequency must not affect the normal charging and discharging process of the battery to ensure the reliability and safety of the detection system.

[0111] Example 5: During battery health status detection, the specific implementation methods for modifying the sampling frequency of signal acquisition when adjustment conditions are met include the following two cases, the technical principles, implementation process, and key points of which will be described in detail below:

[0112] 1. Adjust the sampling frequency to 1.5 times the original frequency.

[0113] In the battery health status detection system, the original sampling frequency is a default frequency preset based on the battery's basic characteristics, signal bandwidth, and detection accuracy requirements. When an adjustment condition is triggered during the traversal of the status classification tree (such as when the number of remaining unidentified abnormal features is less than a preset threshold after traversing a certain level or the first 50% of levels), the system will activate the sampling frequency boosting mechanism to adjust the current sampling frequency to 1.5 times the original frequency.

[0114] The technical logic behind this adjustment is as follows: when a potential battery anomaly is detected or more finely detailed signal information needs to be captured, increasing the sampling frequency increases the number of signal sampling points per unit time, thereby more accurately restoring the true form of the signal, especially anomalies in the high-frequency band. For example, if the original sampling frequency is 200Hz, and it is adjusted to 300Hz, the system can now collect 300 voltage and current signal samples per second. Compared to the original 200 samples, this allows for a more dense capture of transient signal changes, helping to identify minute abnormal fluctuations.

[0115] At the hardware implementation level, this adjustment relies on the programmability of the signal acquisition module. Specifically, the analog-to-digital converter (ADC), as the core acquisition component, needs to support dynamic modification of the sampling clock frequency via software instructions. After detecting the adjustment condition, the system software layer sends a frequency switching command to the ADC driver. The driver adjusts the sampling clock frequency to 1.5 times the original frequency by configuring the ADC's control register or an external clock source. During this process, it is crucial to ensure that the switching of the sampling clock is seamless to avoid signal acquisition interruptions or data misalignment due to clock jumps. This is typically achieved by waiting for the current sampling period to complete before switching the sampling frequency, then enabling the new sampling frequency at the start of the new clock period, and calibrating the first few sampling points to eliminate timing deviations caused by the clock switch.

[0116] Increasing the sampling frequency leads to a significant increase in data volume. Taking the original frequency of 200Hz as an example, the adjusted data volume increases from 200 sample points per second to 300, a 50% increase. This places higher demands on data storage, transmission, and processing modules. To address this challenge, the system can employ data compression techniques, such as differential coding and run-length coding, to compress the acquired raw signal in real time, reducing data transmission bandwidth and storage capacity requirements. Simultaneously, the signal processing algorithm design needs to optimize the computational flow, for example, by adopting pipelined processing or parallel computing architectures, to ensure that wavelet transform, energy distribution calculation, and other processing steps can be completed promptly even at high data rates.

[0117] Increasing the sampling frequency also requires consideration of anti-aliasing filter matching. According to the Nyquist sampling theorem, the sampling frequency must be at least twice the highest frequency component of the signal to avoid aliasing distortion. Therefore, when the sampling frequency is adjusted to 1.5 times the original frequency, the cutoff frequency of the anti-aliasing filter also needs to be increased accordingly to ensure that noise signals higher than half of the new sampling frequency are effectively filtered out. For example, if the cutoff frequency of the anti-aliasing filter corresponding to the original frequency of 200Hz is 100Hz, the cutoff frequency corresponding to the adjusted sampling frequency of 300Hz should be increased to 150Hz to ensure the integrity and accuracy of the sampled signal.

[0118] II. Merging adjacent branches of the state classification tree and adjusting the sampling frequency segmentation strategy

[0119] When the number of levels in the state classification tree is greater than 2 and the adjustment condition is triggered, another method to modify the sampling frequency is to merge branches of adjacent levels and dynamically adjust the segmentation strategy of the sampling frequency based on the merging result. This process involves the structural optimization of the state classification tree and the collaborative design of the sampling strategy. The specific steps are as follows:

[0120] ① Merging adjacent level branches: Merging adjacent level branches in the state classification tree is based on the similarity of branch features. The system analyzes adjacent levels (such as the first level...) Layer and First If the differences in the high-frequency component amplitude range, energy distribution pattern, or anomalous feature type corresponding to the branches (layers) are less than a preset merging threshold, then these branches are merged into a composite branch. For example, suppose two adjacent branches correspond to high-frequency component amplitudes in the ranges of [5mV, 8mV] and [6mV, 9mV], respectively, and both have the anomalous feature of "periodic voltage fluctuations". Then, these two branches can be considered to have high similarity and are merged into one branch, corresponding to the amplitude range of [5mV, 9mV] and a unified anomalous feature type.

[0121] The core objective of branch merging is to simplify the classification tree structure, reduce unnecessary hierarchical judgments, and retain the main discriminative features of anomalies. The merging threshold should be set based on the specific battery type and historical data. For example, for lithium iron phosphate batteries, the amplitude difference threshold can be set to 2mV, and the energy distribution variance difference threshold can be set to 0.1J. When the difference in feature parameters between adjacent branches is less than these thresholds, the merging operation is triggered.

[0122] ② Segmented Sampling Frequency Based on Merging Results: After branch merging, the system divides the sampling frequency into several segmented intervals based on the number and characteristics of the remaining branches. Each segment corresponds to a specific battery state or anomaly level and is matched with a corresponding sampling frequency. For example, if there are three remaining branches after merging, corresponding to "normal state," "moderate anomaly," and "severe anomaly," the following three-segmented sampling frequency strategy can be designed:

[0123] Normal state segmentation: using the original sampling frequency This meets basic testing needs and balances data volume with testing accuracy.

[0124] Moderate anomaly segmentation: using a medium sampling frequency (e.g., 1.2 times the original frequency) is used to capture subtle changes in signal details corresponding to moderate anomalies;

[0125] Severe anomaly segmentation: using a high sampling frequency (e.g., 1.8 times the original frequency) to collect high-frequency data for rapidly changing characteristics of severe anomalies.

[0126] The number and specific values ​​of sampling frequency segments need to be dynamically adjusted based on the characteristics of the merged branches. For example, if two branches remain after merging ("normal" and "abnormal"), a two-segment strategy can be adopted: the normal state maintains the original frequency, while the abnormal state switches to 1.5 times the original frequency. The segmentation strategy design should follow the "on-demand sampling" principle, that is, while ensuring that abnormal features can be identified, the time proportion of high-frequency sampling should be minimized to reduce system resource consumption.

[0127] ③ Dynamic sampling frequency switching mechanism: When traversing the state classification tree, the system switches the sampling frequency segment in real time according to the current branch node. For example, when the classification tree traversal result enters the "severe anomaly" branch, the system immediately triggers a high-frequency sampling segment, acquiring signals at 1.8 times the original frequency; when the traversal result returns to the "normal state" branch, it automatically switches back to the original frequency. To avoid frequent fluctuations in the sampling frequency due to rapid branch switching, a switching anti-jitter mechanism can be set, for example, stipulating that after entering a certain branch, the sampling frequency should be maintained for at least a certain period of time (such as 5 charge-discharge cycles), and then deciding whether to adjust it based on subsequent traversal results.

[0128] To implement dynamic switching, a mapping relationship between branch nodes and sampling frequency segments needs to be established. This mapping can be achieved by embedding metadata in the classification tree nodes, with each node storing its corresponding sampling frequency parameters (such as multipliers or absolute frequency values). When traversing to a certain node, the system reads the sampling frequency parameters of that node and updates the configuration of the signal acquisition module through the hardware control interface, thereby achieving real-time response to frequency switching.

[0129] III. Key Technology Implementation and System Coordination

[0130] a. Co-design of hardware and software: Whether adjusting by a fixed multiple or in segments, the signal acquisition hardware must have flexible frequency configuration capabilities. Common implementation methods include:

[0131] Using an ADC chip with an integrated programmable clock manager, the sampling clock division factor can be dynamically modified via SPI or I2C interface;

[0132] A field-programmable gate array (FPGA) is used as the acquisition controller, and sampling clocks of different frequencies are generated in real time by reconfiguring the FPGA logic;

[0133] By combining a multi-crystal clock source, different frequency clock signals are input to the ADC through analog switches.

[0134] At the software level, the system needs to achieve seamless integration between the sampling frequency adjustment logic and the state analysis module. After detecting the adjustment condition, the state analysis module sends a frequency adjustment command to the acquisition control module through an event-driven mechanism. The acquisition control module parses the command, executes the hardware configuration, and feeds back the adjustment result to the state analysis module, forming a closed-loop control.

[0135] b. Data Consistency and Time Synchronization: Modifications to the sampling frequency may lead to inconsistent data rates at different times; therefore, a data time synchronization mechanism needs to be established. Specific methods include:

[0136] A timestamp is embedded in each sampled data frame to record the absolute time or relative time offset of data acquisition;

[0137] Global clock synchronization technology is used to ensure that the clock references of the signal acquisition module, data processing module and health assessment module are consistent;

[0138] Inserting synchronization markers at frequency switching points facilitates the identification of frequency change boundaries during subsequent data splicing and analysis.

[0139] c. Anomaly Handling and Fault Tolerance Mechanisms: If a hardware failure (such as clock source failure or ADC configuration error) occurs during sampling frequency adjustment, the system must have the capability to detect and recover from the anomaly. For example, when it is detected that the sampling frequency cannot be adjusted as expected, the system should automatically revert to the original frequency, trigger a fault alarm, and record a fault log for subsequent troubleshooting. Furthermore, it is necessary to avoid asynchrony between battery charging / discharging control and signal acquisition due to frequency adjustment. For example, the charging / discharging control loop should be paused before frequency adjustment and resumed after the sampling frequency stabilizes, ensuring that the battery's operating status is not affected by the detection system's adjustment.

[0140] IV. Applicable Scenarios and Selection Logic of the Two Adjustment Methods

[0141] Fixed-multiplier adjustment is suitable for the following scenarios: when the number of abnormal features detected for the first time is close to the preset threshold, it is necessary to increase the global sampling accuracy for comprehensive investigation; when the battery operating environment changes significantly (such as a sudden temperature rise), which may cause the signal feature frequency range to expand, it is necessary to quickly adapt through fixed-multiplier adjustment; when the state classification tree structure is relatively simple with few levels, and no complex segmentation strategy is required.

[0142] Branch merging and segment adjustment are suitable for the following scenarios: the state classification tree has many levels (e.g., more than 5 levels) and the abnormal features have obvious hierarchical distribution (e.g., different severity of abnormalities correspond to different frequency ranges); the battery abnormal features exhibit a dynamic evolution process, and the sampling accuracy needs to be dynamically matched according to the current abnormality level; the system resources are limited, and the allocation of computing and storage resources needs to be optimized through the "on-demand sampling" strategy.

[0143] In practical applications, the two methods can be used in combination. For example, when the adjustment condition is triggered for the first time, the sampling frequency is first increased by a fixed multiple to obtain richer signal data; as the state classification tree traversal deepens, the sampling frequency segments are further refined based on the branch merging results, forming a multi-level adjustment strategy of "global improvement + local optimization", thereby achieving a dynamic balance between sampling efficiency and accuracy at different detection stages.

[0144] By employing the two sampling frequency modification methods described above, the battery health status detection system can dynamically adjust its signal acquisition strategy based on real-time analysis results. This ensures both high-resolution capture of abnormal features and avoids unnecessary resource consumption. The core of this mechanism lies in deeply coupling the structural features of the state classification tree with the signal acquisition parameters. Through intelligent adjustment logic, it achieves adaptive optimization of the detection process, providing crucial technical support for accurate assessment of battery health status.

[0145] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0146] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting battery health status, characterized in that, The method includes: When the battery is in a charge-discharge cycle, the battery's voltage and current signals are collected, and wavelet transform processing is performed on the collected signals to obtain the low-frequency and high-frequency components of the signals; the energy distribution of the low-frequency components and the energy distribution of the high-frequency components are calculated, and the abnormal characteristics of the battery are determined based on the energy distribution. If the number of abnormal features is less than a preset threshold, the original signal of the battery is reconstructed based on the low-frequency component using sparse representation; otherwise, a state classification tree is constructed based on the amplitude change of the high-frequency component and the preset threshold. If the number of levels in the classification tree is greater than 2, the traversal strategy of the state classification tree is adjusted according to the number of identified abnormal features and the total number of abnormal features. When the adjustment conditions are met, the sampling frequency of signal acquisition is modified. The reconstructed battery signal is acquired, and the corresponding health status index is generated in the health status assessment system. The strategy for adjusting the traversal of the state classification tree based on the number of identified abnormal features and the total number of abnormal features is as follows: If, after traversing a certain level of the state classification tree, the total number of abnormal features minus the number of identified abnormal features is less than a preset threshold, then the adjustment condition is triggered. Alternatively, after traversing the first 50% of the state classification tree, if the total number of abnormal features minus the number of identified abnormal features is less than a preset threshold, then the adjustment condition is triggered.

2. The method as described in claim 1, characterized in that, The determination of abnormal characteristics of the battery based on the energy distribution specifically includes: The energy distribution of the low-frequency component is normalized using the energy distribution of the high-frequency component. The normalized energy distribution is input into a pre-established anomaly feature recognition model, and the number of anomaly features is output.

3. The method as described in claim 1, characterized in that, The construction of the state classification tree based on the amplitude changes of high-frequency components and a preset threshold is specifically as follows: Calculate the ratio of the maximum amplitude of the high-frequency component to a preset threshold. If the number of abnormal features is greater than or equal to 50% of the maximum amplitude, the number of branches in the state classification tree is the ratio; otherwise, the number of branches is the integer part of the ratio.

4. The method as described in claim 1, characterized in that, The modification of the signal acquisition sampling frequency is specifically as follows: Adjust the sampling frequency to 1.5 times the original frequency; Alternatively, merge branches at adjacent levels in the state classification tree and adjust the segmentation strategy of sampling frequency based on the merging result.

5. A battery health status detection system, characterized in that, The system includes: The feature extraction module is used to collect the battery's voltage and current signals when the battery is in a charge-discharge cycle, perform wavelet transform processing on the collected signals to obtain the low-frequency and high-frequency components of the signals, calculate the energy distribution of the low-frequency components and the energy distribution of the high-frequency components, and determine the abnormal characteristics of the battery based on the energy distribution. The state analysis module is used to reconstruct the original signal of the battery based on the low-frequency component using sparse representation if the number of abnormal features is less than a preset threshold; otherwise, it constructs a state classification tree based on the amplitude change of the high-frequency component and the preset threshold. If the number of levels in the classification tree is greater than 2, it adjusts the traversal strategy of the state classification tree according to the number of identified abnormal features and the total number of abnormal features. When the adjustment conditions are met, it modifies the sampling frequency of signal acquisition. The health assessment module is used to acquire the reconstructed battery signal and generate corresponding health status indicators in the health status assessment system. The strategy for adjusting the traversal of the state classification tree based on the number of identified abnormal features and the total number of abnormal features is as follows: If, after traversing a certain level of the state classification tree, the total number of abnormal features minus the number of identified abnormal features is less than a preset threshold, then the adjustment condition is triggered. Alternatively, after traversing the first 50% of the state classification tree, if the total number of abnormal features minus the number of identified abnormal features is less than a preset threshold, then the adjustment condition is triggered.

6. The system as described in claim 5, characterized in that, The determination of abnormal characteristics of the battery based on the energy distribution specifically includes: The energy distribution of the low-frequency component is normalized using the energy distribution of the high-frequency component. The normalized energy distribution is input into a pre-established anomaly feature recognition model, and the number of anomaly features is output.

7. The system as described in claim 5, characterized in that, The construction of the state classification tree based on the amplitude changes of high-frequency components and a preset threshold is specifically as follows: Calculate the ratio of the maximum amplitude of the high-frequency component to a preset threshold. If the number of abnormal features is greater than or equal to 50% of the maximum amplitude, the number of branches in the state classification tree is the ratio; otherwise, the number of branches is the integer part of the ratio.

8. The system as described in claim 5, characterized in that, The modification of the signal acquisition sampling frequency is specifically as follows: Adjust the sampling frequency to 1.5 times the original frequency; Alternatively, merge branches at adjacent levels in the state classification tree and adjust the segmentation strategy of sampling frequency based on the merging result.

Images