Battery adaptive safety protection method and system based on data driving

Abstract

The invention relates to the technical field of batteries, and discloses a battery self-adaptive safety protection method and system based on data driving, and the method comprises the steps: extracting parameter features of a transient segment, a steady segment and a fluctuation segment after segmentation, constructing a comprehensive feature matrix based on the parameter features, all environmental factors of the battery are screened based on the environmental influence model, the working condition of the battery is determined based on the working condition identification model, the environmental influence factors and the state feature sequence, the working condition is verified based on historical data, and a corresponding monitoring strategy is selected based on the verified working condition or the re-determined working condition. And the acquisition interval is adjusted based on the monitoring strategy, the risk direction of the battery is determined based on the relationship between the standard state feature sequence and the state feature sequence, and whether power-off protection is performed on the battery is judged based on the association relationship of all the risk directions. According to the invention, through data driving of historical data, the flexibility and reliability of battery safety protection are ensured.

Description

Technical Field

[0001] This invention relates to the field of battery technology, and more specifically, to a data-driven adaptive safety protection method and system for batteries. Background Technology

[0002] With the increasing popularity of energy storage systems, electric vehicles, power tools, and distributed power supply equipment, batteries are being used more and more frequently in high-rate charging and discharging and in complex environmental conditions. In actual operation, the voltage, current and other parameters of batteries are affected by load fluctuations and changes in ambient temperature, thus exhibiting multi-stage characteristics. Since there is a non-linear coupling relationship between these characteristics, it is difficult to effectively protect the safety of batteries by relying on fixed thresholds or single variables.

[0003] Chinese Patent Publication No. CN120978925A discloses a battery over-discharge protection method. The method includes: when the battery is detected to be in a static state, controlling the battery monitoring device to enter a power-off mode, the monitoring device being used to collect the battery's state parameters; determining a target time interval for collecting the battery's state parameters, wherein the target time interval refers to the time interval between two consecutive wake-ups of the monitoring device; and periodically waking up the monitoring device to enter a running mode based on the target time interval, so as to collect the battery's state parameters through the woken monitoring device to obtain the battery's state data, and sending it to the terminal corresponding to the battery. It is evident that when the battery is in a dynamic operating condition, safety protection will fail. Moreover, the interaction between battery parameters will further exacerbate the risk of over-discharge. It cannot adaptively adjust for different battery scenarios. Furthermore, when considering battery over-discharge during static operation, the comprehensive impact of the environment is not taken into account, leading to deviations in the judgment of the operating condition.

[0004] Therefore, it is necessary to design a data-driven adaptive battery safety protection method and system to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes a data-driven adaptive battery safety protection method and system, which aims to solve the above problems.

[0006] In one aspect, this invention proposes a data-driven adaptive battery safety protection method, comprising:

[0007] The battery's operating state parameters are collected and segmented. Parameter features are extracted from the segmented transient, steady-state, and fluctuating segments. A comprehensive feature matrix is ​​constructed based on the parameter features, and a state feature sequence is determined based on the comprehensive feature matrix.

[0008] Based on the environmental impact model, all environmental factors of the battery are screened to determine the environmental impact factors. Based on the operating condition identification model, the environmental impact factors and the state feature sequence, the operating condition of the battery is determined.

[0009] The working conditions are verified based on historical data. If the working conditions are determined to fail the verification, the working conditions of the battery are re-determined based on a clustering algorithm. A corresponding monitoring strategy is selected based on the verified working conditions or the re-determined working conditions, and the collection interval is adjusted based on the monitoring strategy.

[0010] If the verified or redefined operating condition is a risky condition, then the standard state feature sequence corresponding to the battery is obtained. Based on the relationship between the standard state feature sequence and the state feature sequence, the risk direction of the battery is determined. Based on the correlation of all risk directions, it is determined whether to perform power-off protection on the battery.

[0011] Furthermore, when segmenting the operating state parameters, extracting parameter features from the segmented transient, steady-state, and fluctuating segments, constructing a comprehensive feature matrix based on the parameter features, and determining the state feature sequence based on the comprehensive feature matrix, the process includes:

[0012] The operating status parameters include voltage, current, temperature, internal resistance change, polarization response, and load change;

[0013] The operating state parameters are noise-processed and divided into transient, steady-state, and fluctuating segments in chronological order. The impulse response, polarization recovery rate, and heating rate are extracted from the transient segments. The voltage balance and discharge stability are extracted from the steady-state segments. The internal resistance change trend and voltage descent trend are extracted from the fluctuating segments.

[0014] The parameter characteristics include the impact response, polarization recovery rate, heating rate, voltage balance, discharge stability, internal resistance change trend, and voltage descent trend.

[0015] The parameter features of different segments are time-aligned and normalized to determine the comprehensive feature matrix, and the comprehensive feature matrix is ​​mapped to the state feature sequence.

[0016] Furthermore, when screening all environmental factors of the battery based on the environmental impact model to determine the environmental impact factors, and determining the battery's operating conditions based on the operating condition identification model, the environmental impact factors, and the state feature sequence, the process includes:

[0017] The model parameters are determined by grid search and a gradient boosting tree model is built. An environmental dataset is obtained and divided into a model training set and a model test set according to the partitioning ratio. The gradient boosting tree model is trained based on the model training set and validated based on the model test set to determine the model accuracy.

[0018] If the accuracy of the currently trained gradient boosting tree model is lower than that of the previously trained gradient boosting tree model, then adjust the learning rate of the currently trained gradient boosting tree model and continue training.

[0019] If the model accuracy of the currently trained gradient boosting tree model is greater than or equal to the model accuracy of the previously trained gradient boosting tree model, then training is stopped, and the currently trained gradient boosting tree model is determined as the environmental impact model.

[0020] All environmental factors are substituted into the environmental impact model to determine the environmental impact factors, and the environmental impact factors and state feature sequence are substituted into the operating condition identification model to determine the operating condition of the battery. The operating conditions include normal operating condition, fatigue operating condition, fluctuating operating condition and risk operating condition.

[0021] Furthermore, when verifying the operating conditions based on historical data, the following steps are included:

[0022] Based on the historical data, a historical sequence dataset with the same working condition type is determined. The historical sequence dataset includes several historical state feature sequences. The similarity between each historical state feature sequence and the state feature sequence is determined, and the minimum similarity is determined as the sequence similarity.

[0023] The sequence similarity and sequence similarity threshold are compared, and the working condition is verified based on the comparison results;

[0024] If the sequence similarity is greater than or equal to the sequence similarity threshold, the working condition is deemed to have passed the verification.

[0025] If the sequence similarity is less than the sequence similarity threshold, the working condition is deemed to have failed the verification.

[0026] Furthermore, if the operating condition is determined to fail verification, the operating condition of the battery is re-determined based on a clustering algorithm, including:

[0027] Based on the historical data, a historical sequence dataset for each working condition type is determined, and the historical sequence dataset is combined with the state feature sequence to establish a dataset to be clustered. The feature vector of each data in the dataset to be clustered is extracted, the expected number of clusters k is determined to be 4, and the parameters of the Gaussian distribution are initialized. The probability of each data in the dataset to be clustered belonging to each Gaussian distribution is obtained, the responsibility value is determined, and the cluster with the largest responsibility value is determined as the working condition of the battery.

[0028] Furthermore, selecting a corresponding monitoring strategy based on verified or redefined operating conditions, and adjusting the data collection interval based on the monitoring strategy, includes:

[0029] If the verified working condition or the redefined working condition is a normal working condition, the data acquisition interval is extended, and zero-point drift is detected within the data acquisition cycle. If zero-point drift exists, the normal working condition is switched to the fatigue working condition.

[0030] If the verified working condition or the redefined working condition is a fatigue condition, then the data acquisition interval should be shortened, and the temperature distribution uniformity, voltage consistency, and polarization response should be monitored.

[0031] If the verified or redefined operating condition is a fluctuating condition, the voltage change curve, transient current response, and temperature rise rate change are acquired based on the time window, and the acquisition interval is shortened based on the rate of change.

[0032] If the verified or redefined operating condition is a risky condition, the data acquisition interval is shortened and protective actions are triggered, including forced load reduction, limiting the maximum allowable current, and outputting a risk warning.

[0033] Furthermore, if the verified or redefined operating condition is a risky condition, then a standard state feature sequence corresponding to the battery is obtained. When determining the risk direction of the battery based on the relationship between the standard state feature sequence and the state feature sequence, the process includes:

[0034] A standard battery corresponding to the battery is identified, and a standard state feature sequence of the standard battery is determined. The state feature sequence is compared with the standard state feature sequence to determine the feature type that deviates from the standard parameter feature in the parameter feature. Based on the feature type, the risk direction of the battery is determined.

[0035] Furthermore, when determining whether to provide power-off protection for the battery based on the correlation of all risk directions, the process includes:

[0036] The Eclat algorithm generates several candidate item sets for all risk directions. Frequent itemsets are determined based on the support of the candidate item sets. The association results between each risk direction are determined based on the frequent itemsets. The number of associations is counted. Based on the number of associations, it is determined whether to perform power-off protection on the battery.

[0037] Furthermore, when determining whether to perform power-off protection on the battery based on the associated quantity, the process includes:

[0038] If the number of associated components is greater than or equal to the threshold number of associated components, it is determined that the battery should be powered off for protection.

[0039] If the number of associated components is less than the threshold number of associated components, it is determined that no power-off protection will be provided for the battery.

[0040] Compared with existing technologies, the advantages of this invention are as follows: By segmenting battery operating state parameters into transient, steady-state, and fluctuating segments and determining parameter characteristics, the complete evolution path of instantaneous risks, short-term fluctuations, and long-term fluctuations is captured, avoiding omissions and misjudgments caused by judging based on a single parameter. Based on an environmental impact model, environmental factors affecting battery performance are screened from all environmental factors, ensuring that the identification of operating conditions not only relies on electrical parameters but also integrates environmental information for comprehensive judgment. This avoids the risk of ignoring environmental interference and causing deviations in operating condition identification, thus improving the reliability of determining operating conditions. Data-driven verification of operating conditions based on historical data can filter out erroneous judgments caused by short-term noise or sudden changes. Furthermore, when faced with unseen state feature sequences, clustering algorithms ensure the flexibility and intelligence of the monitoring strategy. Dynamic adjustment of the acquisition interval based on the monitoring strategy enables intelligent allocation of monitoring resources. Analysis of correlations reveals whether multiple risk directions exhibit synergy, thus determining whether to trigger power-off protection and ensuring its activation. This achieves a full-scenario, cross-scale, and adaptive battery safety protection system, improving the safety, reliability, and intelligence of battery operation.

[0041] On the other hand, this application also provides a data-driven adaptive battery safety protection system for applying the data-driven adaptive battery safety protection method as described in any one of claims 1-9, comprising:

[0042] The acquisition and analysis module is configured to acquire the battery's operating state parameters, segment the operating state parameters, extract parameter features from the segmented transient segments, steady-state segments, and fluctuation segments, construct a comprehensive feature matrix based on the parameter features, and determine the state feature sequence based on the comprehensive feature matrix.

[0043] The operating condition determination module is configured to screen all environmental factors of the battery based on the environmental impact model, determine the environmental impact factors, and determine the operating condition of the battery based on the operating condition identification model, the environmental impact factors, and the state feature sequence.

[0044] The adaptive protection module is configured to verify the operating conditions based on historical data. If the operating conditions are determined to fail the verification, the operating conditions of the battery are re-determined based on a clustering algorithm. The corresponding monitoring strategy is selected based on the verified operating conditions or the re-determined operating conditions, and the acquisition interval is adjusted based on the monitoring strategy.

[0045] The power failure protection module is configured to, if the verified operating condition or the redefined operating condition is a risky condition, acquire the standard state feature sequence corresponding to the battery, determine the risk direction of the battery based on the relationship between the standard state feature sequence and the state feature sequence, and determine whether to perform power failure protection on the battery based on the correlation of all risk directions.

[0046] It is understandable that the aforementioned data-driven adaptive battery safety protection method and system have the same beneficial effects, and will not be elaborated further here. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart illustrating a data-driven adaptive battery safety protection method provided in an embodiment of the present invention;

[0049] Figure 2 This is a functional block diagram of a data-driven adaptive battery safety protection system provided in an embodiment of the present invention. Detailed Implementation

[0050] 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.

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] See Figure 1 As shown in some embodiments of this application, a data-driven adaptive battery safety protection method includes:

[0053] S100: Collects the battery's operating state parameters, segments the operating state parameters, extracts parameter features from the segmented transient segments, steady-state segments, and fluctuation segments, constructs a comprehensive feature matrix based on the parameter features, and determines the state feature sequence based on the comprehensive feature matrix.

[0054] S200: Based on the environmental impact model, all environmental factors of the battery are screened to determine the environmental impact factors. Based on the operating condition identification model, environmental impact factors and state characteristic sequence, the operating condition of the battery is determined.

[0055] S300: Verify the operating conditions based on historical data. If the operating conditions are determined to fail the verification, the operating conditions of the battery are redefined based on the clustering algorithm. The corresponding monitoring strategy is selected based on the verified operating conditions or the redefined operating conditions, and the collection interval is adjusted based on the monitoring strategy.

[0056] S400: If the verified working condition or the redefined working condition is a risky condition, then obtain the standard state feature sequence corresponding to the battery, determine the risk direction of the battery based on the relationship between the standard state feature sequence and the state feature sequence, and determine whether to perform power-off protection on the battery based on the correlation of all risk directions.

[0057] Specifically, the system collects battery operating state parameters, which are multi-source time-series data characterizing the battery's real-time and long-term behavior, including voltage, current, and temperature. These operating state parameters are segmented according to time sequence to obtain transient, steady-state, and fluctuating segments. The transient segment reflects the short-term rapid response under sudden load changes or thermal shocks, the steady-state segment reflects the stable voltage performance, and the fluctuating segment reflects the trend of battery performance changes. Parameter features are extracted for each segment; these features are descriptive quantities extracted from the segment, describing characteristics such as numerical magnitude and numerical variation. A comprehensive feature matrix is ​​constructed based on parametric features. This comprehensive feature matrix is ​​a unified representation organized by time and feature dimensions, used to eliminate scale differences between segments. The comprehensive feature matrix then generates a state feature sequence, which can simultaneously carry data information on short-term mutations and long-term evolution. After having a unified state representation (state feature sequence), an environmental impact model is invoked to screen all environmental factors of the battery. Environmental factors include the battery's altitude, ultraviolet radiation intensity, dust concentration, temperature, humidity, air pressure, wind speed, and air density. The environmental impact model is a big data model used to identify the impact of environmental factors on the battery. By evaluating all environmental factors, it outputs environmental impact factors, which represent environmental factors that have a significant impact on the battery. Environmental influencing factors and state feature sequences are input into the operating condition identification model to obtain the battery's operating condition. The operating condition is a classification label for the current battery operating state. When processing inputs from different data dimensions, the operating condition identification model may misjudge. Therefore, the operating condition is verified based on historical data, which includes all operating conditions and corresponding data. This historical data helps determine the similarity between the operating condition and previous data. If the operating condition fails verification, it is re-identified using a clustering algorithm. Clustering is an algorithm that aggregates similar data, determining cluster affiliation based on the distance between feature vectors in the dataset to be clustered. The feature attributes of the cluster centers are then mapped to the operating condition to complete the re-determination of the operating condition. Based on the verified or re-determined operating condition, a corresponding monitoring strategy is selected. The monitoring strategy is a configuration scheme for the collection interval and corresponding monitoring items under different operating conditions. The collection interval is the time between two adjacent collections. Different monitoring strategies enable the resource investment in the data collection process to match the risk level of the battery, thereby achieving adaptive perception and safety protection for the entire battery life cycle and all scenarios. Furthermore, the reliability of the battery's adaptive safety protection is improved through historical data-driven verification.

[0058] Understandably, if the verified or redefined operating conditions are deemed risky, it indicates that the battery faces certain dangerous conditions and could evolve into a severe fault. Therefore, a standard state characteristic sequence corresponding to the battery is obtained. The standard state characteristic sequence corresponds to the state characteristic sequence. Using the standard state characteristic sequence as a benchmark, the risk direction of the battery is determined based on the relationship between the standard state characteristic sequence and the state characteristic sequence. The risk direction represents the deviation trend of data in the battery's parameter characteristics, including the deviation direction of temperature-related behavior, polarization-related behavior, voltage decay-related behavior, and internal resistance evolution-related behavior relative to the benchmark. The correlation of all risk directions reflects the co-occurrence patterns of different deviation directions in time, space, and causality. Determining whether to perform power-off protection for the battery based on the correlation of all risk directions avoids exacerbating the risk of over-discharge by relying on a single factor. The closed-loop mechanism of historical verification and clustering-based re-determination, along with the data-driven approach, ensures the reliability of battery safety protection. Furthermore, selecting the corresponding monitoring strategy according to the operating conditions ensures the flexibility and intelligence of safety protection.

[0059] In some embodiments of this application, when segmenting the operating state parameters, extracting parameter features from the segmented transient, steady-state, and fluctuating segments, constructing a comprehensive feature matrix based on the parameter features, and determining the state feature sequence based on the comprehensive feature matrix, the process includes: the operating state parameters include voltage, current, temperature, internal resistance change, polarization response, and load change; noise processing is performed on the operating state parameters, and they are segmented into transient, steady-state, and fluctuating segments in chronological order; impulse response, polarization recovery rate, and heating rate are extracted from the transient segments; voltage balance degree and discharge stability are extracted from the steady-state segments; and internal resistance change trend and voltage descent trend are extracted from the fluctuating segments. The parameter features include impulse response, polarization recovery rate, heating rate, voltage balance degree, discharge stability, internal resistance change trend, and voltage descent trend. The parameter features of different segments are time-aligned and normalized to determine the comprehensive feature matrix, and the comprehensive feature matrix is ​​mapped to the state feature sequence.

[0060] Specifically, by utilizing the response behavior of the battery at different operating stages, the operating state parameters are fragmented, deconstructed, characterized, and sequentially expressed to obtain a state feature sequence that comprehensively reflects the dynamic health state of the battery. The operating state parameters include voltage, current, temperature, internal resistance change, polarization response, and load change. Voltage reflects the battery's electrochemical energy conversion capability, current characterizes the actual output capability under load, temperature describes the battery's thermal state, internal resistance change reflects the changes in the battery's internal structure and material properties, polarization response reflects the battery's ability to recover equilibrium after disturbance, and load change represents the effect of external power demand on the battery. The operating status parameters are noise-processed to eliminate abnormal offsets caused by measurement equipment errors and transient interference, ensuring the reliability of subsequent segmentation. Then, the noise-processed operating status parameters are segmented into transient segments, steady-state segments, and fluctuation segments according to time sequence. The transient segment reflects short-term dynamic processes such as load changes and current surges in the battery. The steady-state segment reflects the voltage change process during the relatively stable phase of stable power supply or rest in the battery. The fluctuation segment reflects the operating process of the battery with non-uniform, periodic, or irregular load fluctuations. For different segments, parameter features characterizing their essential behavior are extracted. In the transient segment, impact response, polarization recovery rate, and heating rate are extracted. Impact response represents the voltage fluctuation amplitude when the battery is subjected to instantaneous current or load changes. Polarization recovery rate represents the battery's ability to recover from disturbance to steady state. Heating rate represents the intensity of thermal changes during the transient process. In the steady-state segment, voltage balance and discharge stability are extracted. Voltage balance represents the battery's equilibrium under steady state, while discharge stability measures the battery's output consistency during continuous power supply. In the fluctuating segment, internal resistance change trend and voltage descent trend are extracted. Internal resistance change trend reflects the evolution direction of the battery's internal impedance under unstable load conditions, and voltage descent trend reflects the voltage drop pattern during load fluctuations. The parameter features of different segments are time-aligned and normalized to ensure that each feature forms a uniform scale of comparison in chronological order. Normalization eliminates differences in the dimensions and units of different features, making the features comparable. A comprehensive feature matrix is ​​constructed based on time-aligned and normalized parameter features. This matrix integrates the change patterns of different segments and features on the time axis in a two-dimensional structure. Subsequently, the comprehensive feature matrix is ​​mapped into a state feature sequence. This sequence describes the battery's dynamic response mode, steady-state behavior, and fluctuation trends in a serialized manner. The state feature sequence not only captures the battery's transient impacts, steady-state equilibrium, and fluctuation trends, but also reveals the temporal logical relationships between features, thereby improving the accuracy of battery operating state identification and laying the foundation for subsequent adaptive safety protection.

[0061] In some embodiments of this application, when screening all environmental factors of the battery based on an environmental impact model to determine environmental impact factors, and determining the battery's operating conditions based on an operating condition identification model, environmental impact factors, and state feature sequences, the process includes: determining model parameters based on grid search and establishing a gradient boosting tree model; obtaining an environmental dataset; dividing the environmental dataset into a model training set and a model test set according to a partitioning ratio; training the gradient boosting tree model based on the model training set; and validating the trained gradient boosting tree model based on the model test set to determine the model accuracy. If the currently trained gradient boosting tree model is accurate... If the learning rate is less than the model accuracy of the gradient boosting tree model after the previous training, the learning rate of the currently trained gradient boosting tree model is adjusted, and training continues. If the model accuracy of the currently trained gradient boosting tree model is greater than or equal to the model accuracy of the previously trained gradient boosting tree model, training is stopped, and the currently trained gradient boosting tree model is identified as the environmental impact model. All environmental factors are substituted into the environmental impact model to determine the environmental impact factors. The environmental impact factors and state feature sequences are substituted into the operating condition identification model to determine the battery's operating conditions. The operating conditions include normal operating conditions, fatigue operating conditions, fluctuating operating conditions, and risk operating conditions.

[0062] Specifically, the model parameters are determined based on grid search, and a gradient boosting tree model is established. The gradient boosting tree model is an ensemble model composed of multiple weak learners, which can continuously optimize the error through gradient iteration, enabling the model to capture the influence of various environmental factors on the battery's operating state. The environmental dataset includes environmental data such as ambient temperature, ambient humidity, atmospheric pressure, altitude, air velocity, ambient dust concentration, and corrosive gas concentration. The environmental dataset is divided into a model training set and a model test set in a ratio of 3:2 to ensure the model's generalization ability. The model training set is used to train the gradient boosting tree model, while the model test set is used to evaluate the performance of the trained model. The gradient boosting tree model is trained using data from the training set. During each training iteration, the model attempts to learn patterns and relationships from the training data to improve its prediction or classification capabilities. The model accuracy is determined based on the test set. If the accuracy of the current trained model is lower than the accuracy of the previous model, it indicates a performance decline. The learning rate is then adjusted using methods such as cosine annealing to control the magnitude of gradient updates, and training continues. If the accuracy of the current trained gradient boosting tree model is greater than or equal to the accuracy of the previous trained model, it indicates that the model's performance has improved or stabilized, and training can then be stopped. Training is stopped, and the currently trained gradient boosting tree model is designated as the environmental impact model. This model filters out environmental factors that have a real impact on the battery's operating state from all environmental factors, such as temperature, humidity, and air pressure, thereby eliminating irrelevant or weakly correlated environmental variables, such as ultraviolet radiation intensity. The environmental impact factors and state feature sequences are then input into the operating condition identification model, which is based on a random forest model and built using data on battery current, voltage, and load changes under various environmental conditions. The training process of the random forest model is consistent with that of the gradient boosting tree model and will not be repeated here. By filtering environmental factors through the environmental impact model, operating condition identification is unaffected by irrelevant environmental factors. Furthermore, the operating condition identification model captures the complex relationship between environmental factors and battery operation, improving the accuracy of operating condition identification and thus enhancing the reliability and intelligence of the battery's adaptive safety protection.

[0063] In some embodiments of this application, when verifying working conditions based on historical data, the process includes: determining a historical sequence dataset of the same type as the working condition based on the historical data, wherein the historical sequence dataset includes several historical state feature sequences; determining the similarity between each historical state feature sequence and the state feature sequence, and determining the minimum similarity as the sequence similarity; comparing the sequence similarity with a sequence similarity threshold; verifying the working condition based on the comparison result; if the sequence similarity is greater than or equal to the sequence similarity threshold, the working condition is determined to pass the verification; if the sequence similarity is less than the sequence similarity threshold, the working condition is determined to fail the verification.

[0064] In some embodiments of this application, if the working condition is determined to fail verification, the working condition of the battery is re-determined based on a clustering algorithm, including: determining a historical sequence dataset for each working condition type based on historical data, combining the historical sequence dataset with the state feature sequence to establish a dataset to be clustered, extracting the feature vector of each data in the dataset to be clustered, determining the expected number of clusters k to be 4, initializing the parameters of the Gaussian distribution, obtaining the probability that each data in the dataset to be clustered belongs to each Gaussian distribution, determining the responsibility value, and determining the working condition of the battery by the cluster with the largest responsibility value.

[0065] Specifically, the historical data consists of a dataset containing all working conditions and their corresponding historical sequences. All historical state feature sequences with the same working condition are selected from the historical data. The similarity between each historical state feature sequence and the current state feature sequence is determined. Similarity can be determined using methods such as Euclidean distance or cosine similarity. The minimum value among all similarities is defined as the sequence similarity, serving as the most conservative assessment of the current working condition. The sequence similarity threshold is set to 0.8. The sequence similarity is compared to the threshold. If the sequence similarity is greater than or equal to the threshold, it indicates that even the smallest similarity case is consistent with the historical working condition, and the working condition is deemed to have passed verification. Otherwise, the current working condition is considered to have deviated, and verification is deemed unsuccessful. By comparing with historical state feature sequences, deviations in working conditions can be accurately identified, avoiding errors caused by relying solely on the model. Furthermore, by continuously accumulating data and updating historical state feature sequences, the judgment capability and adaptability are gradually improved. Even when faced with new or unseen sequences of state characteristics, it can still provide reliable judgment results. The data-driven automatic judgment process improves the reliability of battery adaptive safety protection. Historical sequence datasets of operating conditions, namely historical sequence datasets corresponding to normal operating conditions, fatigue operating conditions, fluctuating operating conditions, and risky operating conditions, are used to redetermine the battery's operating conditions using clustering algorithms. This improves the battery's adaptive capability when facing errors introduced by the dependent model.

[0066] In some embodiments of this application, a corresponding monitoring strategy is selected based on a verified or redefined operating condition, and the acquisition interval is adjusted based on the monitoring strategy. This includes: if the verified or redefined operating condition is a normal operating condition, the acquisition interval is extended, and zero-point drift is detected within the acquisition period. If zero-point drift exists, the normal operating condition is switched to a fatigue operating condition. If the verified or redefined operating condition is a fatigue operating condition, the acquisition interval is shortened, and temperature distribution uniformity, voltage consistency, and polarization response are monitored. If the verified or redefined operating condition is a fluctuating operating condition, the voltage change curve, transient current response, and temperature rise rate change are acquired based on a time window, and the acquisition interval is shortened based on the rate of change. If the verified or redefined operating condition is a risky operating condition, the acquisition interval is shortened and protection behavior is triggered. Protection behavior includes forced load reduction, limiting the maximum allowable current, and outputting a risk warning.

[0067] Specifically, the initial interval for collecting battery operating status parameters is set to once every 30 seconds. This interval can be adjusted based on battery capacity and integration level. If the operating condition is normal, indicating stable parameters, slow temperature changes, and minimal load fluctuations, extending the collection interval reduces resource consumption. Typically, the extension time is proportional to the number of consecutive normal operating conditions; the more consecutive normal operating conditions, the longer the extension time. Zero-point drift detection is performed within the collection cycle. Zero-point drift indicates that when the input signal to the amplifier circuit is zero, the static operating point changes due to factors such as unstable power supply voltage, and this change is amplified and transmitted stage by stage, causing the output voltage to deviate. If zero-point drift is detected, it indicates a slight abnormality in the battery state, and the operating condition is switched to fatigue condition to improve the efficiency of data acquisition and monitoring. When the operating condition is fatigue-related, it indicates that the battery is in a stage of mild aging, performance degradation caused by repeated load impacts, or moderate heat. In this case, the sampling interval is shortened by 2 seconds to increase the monitoring density, meaning sampling is performed every 28 seconds. The specific reduction in time is adjusted according to the battery capacity and the degree of battery integration; generally, the larger the battery capacity and the more complex the battery integration, the greater the reduction in time. While shortening the sampling interval, the focus is on monitoring temperature distribution uniformity, voltage consistency, and polarization response. Temperature distribution uniformity reflects the difference in heat dissipation between cells, voltage consistency reflects the synchronicity of series-connected cells, and polarization response reflects the reversibility of ion movement within the battery. These parameters can provide early warning of the trend from fatigue to failure. If the operating condition is fluctuating, it indicates that the battery performance fluctuates discontinuously due to load changes. A voltage change curve is formed by arranging the continuously sampled voltages within a fixed-length time window in chronological order. Within the same time window, current is synchronously read and transient current response is extracted, i.e., the change in current at abrupt change points (rising or falling edges). First-order trend analysis is performed on the temperature within the time window to determine the rate of temperature increase over time. The respective rates of change (slope of voltage / current and rate of temperature rise) are determined separately. The values ​​of all rates of change (regardless of units) are weighted to determine a weighted sum. The acquisition interval is shortened based on the weighted sum; the larger the weighted sum, the more time is reduced. When the operating condition is a risky condition, the battery exhibits abnormalities such as excessive polarization or voltage instability. In this case, the acquisition interval is shortened by 5 seconds to increase monitoring density. The specific reduction time is also adjusted according to the battery capacity and battery integration level, and protection behaviors are triggered, such as forced load reduction (reducing power output to alleviate battery stress), limiting the maximum allowable current (preventing thermal runaway caused by current surges), and outputting a risk warning (informing the user through an alarm signal).The data acquisition interval is adaptively adjusted based on the stability, volatility, and potential risk levels of battery behavior under different operating conditions. This ensures safety while avoiding resource consumption caused by invalid data acquisition, thereby improving the battery's adaptability and the reliability of safety protection, and ultimately enhancing overall safety.

[0068] In some embodiments of this application, if the verified working condition or the redefined working condition is a risky working condition, then when obtaining the standard state feature sequence corresponding to the battery and determining the risk direction of the battery based on the relationship between the standard state feature sequence and the state feature sequence, the process includes: determining the standard battery corresponding to the battery and determining the standard state feature sequence of the standard battery; comparing the state feature sequence with the standard state feature sequence to determine the feature type that deviates from the standard parameter feature in the parameter features; and determining the risk direction of the battery based on the feature type.

[0069] Specifically, a standard battery corresponding to the current battery is identified, which is an unused battery of the same model as the current battery. The standard battery is then determined through experimental simulations and other methods to establish a standard state characteristic sequence. The state characteristic sequence and the standard state characteristic sequence are in one-to-one correspondence, thus ensuring that the characteristic types of the parameter characteristics are in one-to-one correspondence. That is, each parameter in the parameter characteristics corresponds to each standard parameter in the standard parameter characteristics. When one or more parameters in the parameter characteristics are found to deviate from the standard parameters during the comparison process, such as prolonged polarization recovery time or a continuous increase in internal resistance, these parameter deviations are recorded as risk directions, thereby ensuring the intelligence and reliability of battery safety protection.

[0070] In some embodiments of this application, when determining whether to perform power-off protection on the battery based on the correlation of all risk directions, the method includes: generating several candidate itemsets for all risk directions according to the Eclat algorithm, determining frequent itemsets based on the support of the candidate itemsets, determining the correlation results between each risk direction based on the frequent itemsets, counting the number of correlation results, and determining whether to perform power-off protection on the battery based on the number of correlations.

[0071] In some embodiments of this application, when determining whether to perform power-off protection on the battery based on the number of associated connections, the method includes: if the number of associated connections is greater than or equal to a threshold number of associated connections, then it is determined that power-off protection on the battery should be performed; if the number of associated connections is less than the threshold number of associated connections, then it is determined that power-off protection on the battery should not be performed.

[0072] Specifically, risk directions are treated as typified results of abnormal battery trends, such as voltage imbalance trends and polarization intensification trends. Each risk direction reflects the instability of the battery in a certain operational dimension. However, a single risk direction does not necessarily mean that the battery needs power-off protection. The Eclat algorithm is used to perform combined analysis on all risk directions to analyze whether multiple risk directions have parallel occurrence or mutual reinforcement relationships. By traversing all risk directions, several candidate itemsets are generated by combining simultaneously occurring risk directions. Based on the frequency of occurrence of candidate itemsets in all risk events, frequent itemsets are determined. Frequent itemsets represent the probability of synergistic relationships between risk directions in actual operation. For example, thermal instability trends are usually accompanied by polarization intensification trends. Based on these frequent itemsets, the correlation results between each pair or multiple risk directions are further derived, and these correlation results are statistically analyzed to determine the number of correlations. The correlation number is used to measure the state of overlapping battery risks. The correlation number threshold is inversely proportional to the degree of battery integration. The more complex the battery integration, the lower the correlation number threshold, thus strengthening the protection of delicate batteries. Conversely, the simpler the battery integration, meaning fewer structural layers and simpler components, such as a single cell, the higher the temperature of a single cell usually does not cause voltage imbalance or increased polarization time. Even if there is an abnormal trend in a certain aspect, it will not reach the standard for power-off. Therefore, the correlation number threshold is correspondingly higher. If the correlation number is greater than or equal to the correlation number threshold, it indicates a strong correlation or co-occurrence effect between multiple risk directions. At this time, the battery has entered a high-risk state, triggering power-off protection to prevent further evolution of risks. If the correlation number is less than the correlation number threshold, it indicates the existence of certain risk directions, but no synergistic deterioration has yet occurred. In this case, power-off protection is not required, thus improving the reliability of battery safety protection and capturing potential instability chain reactions, thereby enhancing the overall operational safety and stability of the battery.

[0073] In summary, the beneficial effects of this invention are as follows: By segmenting battery operating state parameters into transient, steady-state, and fluctuating segments and determining parameter characteristics, the complete evolution path of instantaneous risks, short-term fluctuations, and long-term fluctuations is captured, avoiding omissions and misjudgments caused by relying on a single parameter. Based on an environmental impact model, environmental factors affecting battery performance are screened from all environmental factors, ensuring that the identification of operating conditions not only relies on electrical parameters but also integrates environmental information for comprehensive judgment. This avoids the risk of ignoring environmental interference and causing deviations in operating condition identification, thus improving the reliability of determining operating conditions. Data-driven verification of operating conditions based on historical data can filter out erroneous judgments caused by short-term noise or sudden changes. Furthermore, when faced with unseen state feature sequences, clustering algorithms ensure the flexibility and intelligence of the monitoring strategy. Dynamic adjustment of the acquisition interval based on the monitoring strategy enables intelligent allocation of monitoring resources. Analysis of correlations reveals whether multiple risk directions exhibit synergy, thus determining whether to trigger power-off protection and ensuring its activation. This achieves a full-scenario, cross-scale, and adaptive battery safety protection system, improving the safety, reliability, and intelligence of battery operation.

[0074] In another preferred embodiment based on the above embodiments, see [reference] Figure 2 As shown, this embodiment provides a data-driven adaptive battery safety protection system for applying the above-described data-driven adaptive battery safety protection method, including:

[0075] The data acquisition and analysis module is configured to acquire the battery's operating state parameters, segment the operating state parameters, extract parameter features from the segmented transient segments, steady-state segments, and fluctuation segments, construct a comprehensive feature matrix based on the parameter features, and determine the state feature sequence based on the comprehensive feature matrix.

[0076] The operating condition determination module is configured to screen all environmental factors of the battery based on the environmental impact model, determine the environmental impact factors, and determine the battery's operating condition based on the operating condition identification model, environmental impact factors, and state feature sequence.

[0077] The adaptive protection module is configured to verify the operating conditions based on historical data. If the operating conditions are determined to fail the verification, the operating conditions of the battery are redefined based on a clustering algorithm. The corresponding monitoring strategy is selected based on the verified operating conditions or the redefined operating conditions, and the acquisition interval is adjusted based on the monitoring strategy.

[0078] The power failure protection module is configured to acquire the standard state feature sequence corresponding to the battery if the verified working condition or the redefined working condition is a risky condition, determine the risk direction of the battery based on the relationship between the standard state feature sequence and the state feature sequence, and determine whether to perform power failure protection on the battery based on the correlation of all risk directions.

[0079] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A data-driven adaptive battery safety protection method, characterized in that, include: The battery's operating state parameters are collected and segmented. Parameter features are extracted from the segmented transient, steady-state, and fluctuating segments. A comprehensive feature matrix is ​​constructed based on the parameter features, and a state feature sequence is determined based on the comprehensive feature matrix. Based on the environmental impact model, all environmental factors of the battery are screened to determine the environmental impact factors. Based on the operating condition identification model, the environmental impact factors and the state feature sequence, the operating condition of the battery is determined. The working conditions are verified based on historical data. If the working conditions are determined to fail the verification, the working conditions of the battery are re-determined based on a clustering algorithm. A corresponding monitoring strategy is selected based on the verified working conditions or the re-determined working conditions, and the collection interval is adjusted based on the monitoring strategy. If the verified or redefined operating condition is a risky condition, then the standard state feature sequence corresponding to the battery is obtained. Based on the relationship between the standard state feature sequence and the state feature sequence, the risk direction of the battery is determined. Based on the correlation of all risk directions, it is determined whether to perform power-off protection on the battery.

2. The data-driven adaptive battery safety protection method according to claim 1, characterized in that, When segmenting the operating state parameters, extracting parameter features from the segmented transient, steady-state, and fluctuating segments, constructing a comprehensive feature matrix based on the parameter features, and determining the state feature sequence based on the comprehensive feature matrix, the process includes: The operating status parameters include voltage, current, temperature, internal resistance change, polarization response, and load change; The operating state parameters are noise-processed and divided into transient, steady-state, and fluctuating segments in chronological order. The impulse response, polarization recovery rate, and heating rate are extracted from the transient segments. The voltage balance and discharge stability are extracted from the steady-state segments. The internal resistance change trend and voltage descent trend are extracted from the fluctuating segments. The parameter characteristics include the impact response, polarization recovery rate, heating rate, voltage balance, discharge stability, internal resistance change trend, and voltage descent trend. The parameter features of different segments are time-aligned and normalized to determine the comprehensive feature matrix, and the comprehensive feature matrix is ​​mapped to the state feature sequence.

3. The data-driven adaptive battery safety protection method according to claim 2, characterized in that, When screening all environmental factors of the battery based on an environmental impact model to determine the environmental impact factors, and determining the battery's operating conditions based on an operating condition identification model, the environmental impact factors, and a state feature sequence, the process includes: The model parameters are determined by grid search and a gradient boosting tree model is built. An environmental dataset is obtained and divided into a model training set and a model test set according to the partitioning ratio. The gradient boosting tree model is trained based on the model training set and validated based on the model test set to determine the model accuracy. If the accuracy of the currently trained gradient boosting tree model is lower than that of the previously trained gradient boosting tree model, then adjust the learning rate of the currently trained gradient boosting tree model and continue training. If the model accuracy of the currently trained gradient boosting tree model is greater than or equal to the model accuracy of the previously trained gradient boosting tree model, then training is stopped, and the currently trained gradient boosting tree model is determined as the environmental impact model. All environmental factors are substituted into the environmental impact model to determine the environmental impact factors, and the environmental impact factors and state feature sequence are substituted into the operating condition identification model to determine the operating condition of the battery. The operating conditions include normal operating condition, fatigue operating condition, fluctuating operating condition and risk operating condition.

4. The data-driven adaptive battery safety protection method according to claim 3, characterized in that, When verifying the operating conditions based on historical data, the following are included: Based on the historical data, a historical sequence dataset with the same working condition type is determined. The historical sequence dataset includes several historical state feature sequences. The similarity between each historical state feature sequence and the state feature sequence is determined, and the minimum similarity is determined as the sequence similarity. The sequence similarity and sequence similarity threshold are compared, and the working condition is verified based on the comparison results; If the sequence similarity is greater than or equal to the sequence similarity threshold, the working condition is deemed to have passed the verification. If the sequence similarity is less than the sequence similarity threshold, the working condition is deemed to have failed the verification.

5. The data-driven adaptive battery safety protection method according to claim 4, characterized in that, If the operating condition is determined to fail verification, the operating condition of the battery is re-determined based on a clustering algorithm, including: Based on the historical data, a historical sequence dataset for each working condition type is determined, and the historical sequence dataset is combined with the state feature sequence to establish a dataset to be clustered. The feature vector of each data in the dataset to be clustered is extracted, the expected number of clusters k is determined to be 4, and the parameters of the Gaussian distribution are initialized. The probability of each data in the dataset to be clustered belonging to each Gaussian distribution is obtained, the responsibility value is determined, and the cluster with the largest responsibility value is determined as the working condition of the battery.

6. The data-driven adaptive battery safety protection method according to claim 5, characterized in that, Selecting a corresponding monitoring strategy based on verified or redefined operating conditions, and adjusting the data collection interval based on the monitoring strategy, includes: If the verified working condition or the redefined working condition is a normal working condition, the data acquisition interval is extended, and zero-point drift is detected within the data acquisition cycle. If zero-point drift exists, the normal working condition is switched to the fatigue working condition. If the verified working condition or the redefined working condition is a fatigue condition, then the data acquisition interval should be shortened, and the temperature distribution uniformity, voltage consistency, and polarization response should be monitored. If the verified or redefined operating condition is a fluctuating condition, the voltage change curve, transient current response, and temperature rise rate change are acquired based on the time window, and the acquisition interval is shortened based on the rate of change. If the verified or redefined operating condition is a risky condition, the data acquisition interval is shortened and protective actions are triggered, including forced load reduction, limiting the maximum allowable current, and outputting a risk warning.

7. The data-driven adaptive battery safety protection method according to claim 6, characterized in that, If the verified or redefined operating condition is a risky condition, then a standard state feature sequence corresponding to the battery is obtained. When determining the risk direction of the battery based on the relationship between the standard state feature sequence and the state feature sequence, the following steps are taken: A standard battery corresponding to the battery is identified, and a standard state feature sequence of the standard battery is determined. The state feature sequence is compared with the standard state feature sequence to determine the feature type that deviates from the standard parameter feature in the parameter feature. Based on the feature type, the risk direction of the battery is determined.

8. The data-driven adaptive battery safety protection method according to claim 7, characterized in that, When determining whether to provide power-off protection for the battery based on the correlation of all risk directions, the following are included: The Eclat algorithm generates several candidate item sets for all risk directions. Frequent itemsets are determined based on the support of the candidate item sets. The association results between each risk direction are determined based on the frequent itemsets. The number of associations is counted. Based on the number of associations, it is determined whether to perform power-off protection on the battery.

9. The data-driven adaptive battery safety protection method according to claim 8, characterized in that, When determining whether to perform power-off protection on the battery based on the associated quantity, the following are included: If the number of associated components is greater than or equal to the threshold number of associated components, it is determined that the battery should be powered off for protection. If the number of associated components is less than the threshold number of associated components, it is determined that no power-off protection will be provided for the battery.

10. A data-driven adaptive battery safety protection system, used to apply the data-driven adaptive battery safety protection method as described in any one of claims 1-9, characterized in that, include: The acquisition and analysis module is configured to acquire the battery's operating state parameters, segment the operating state parameters, extract parameter features from the segmented transient segments, steady-state segments, and fluctuation segments, construct a comprehensive feature matrix based on the parameter features, and determine the state feature sequence based on the comprehensive feature matrix. The operating condition determination module is configured to screen all environmental factors of the battery based on the environmental impact model, determine the environmental impact factors, and determine the operating condition of the battery based on the operating condition identification model, the environmental impact factors, and the state feature sequence. The adaptive protection module is configured to verify the operating conditions based on historical data. If the operating conditions are determined to fail the verification, the operating conditions of the battery are re-determined based on a clustering algorithm. The corresponding monitoring strategy is selected based on the verified operating conditions or the re-determined operating conditions, and the acquisition interval is adjusted based on the monitoring strategy. The power failure protection module is configured to, if the verified operating condition or the redefined operating condition is a risky condition, acquire the standard state feature sequence corresponding to the battery, determine the risk direction of the battery based on the relationship between the standard state feature sequence and the state feature sequence, and determine whether to perform power failure protection on the battery based on the correlation of all risk directions.

Images