Lithium battery voltage runaway prevention method and system based on data scenario adaptation

Abstract

The application provides a lithium battery voltage runaway prevention method and system based on data scene self-adaptation, and belongs to the technical field of lithium battery voltage protection; the method comprises the following steps: collecting a plurality of parameter tuples related to cell self-discharge, and selecting a key judgment parameter set from the parameter tuples; mixing the abnormal calibration cells with normal cells to perform full-process testing, and generating a calibration data set through process data collection; constructing an initial judgment model according to the data scene of the calibration data set; using a monitoring model based on a time sequence function to monitor the parameter fluctuation of the cell self-discharge in real time, and training the initial judgment model when the parameter fluctuation reaches the alarm condition of the monitoring model, so as to obtain a judgment model for lithium battery voltage runaway prediction. The application can solve the risk of the collapse of the robustness of the model system for lithium battery voltage runaway prediction caused by the cumulative progress of data trends.

Description

Technical Field

[0001] This invention belongs to the field of lithium battery voltage protection technology, and particularly relates to a lithium battery voltage runaway prevention method and system based on data scenario adaptation. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] As the core power source for electric vehicles, the voltage consistency of lithium-ion batteries is crucial for system safety. Voltage runaway (i.e., uneven voltage among individual cells in a series-connected battery pack) can lead to overcharging / over-discharging of local cells, causing serious problems such as irreversible capacity decay, a sharp drop in cycle life, and deterioration in storage performance. Therefore, how to prevent voltage runaway in lithium-ion batteries is a very worthwhile research topic.

[0004] However, existing technologies for preventing voltage runaway in lithium batteries generally suffer from some technical problems, such as: (1) Existing methods mainly rely on single-dimensional data (such as static voltage threshold) or simple rule combinations to identify voltage runaway. Under such identification methods: on the one hand, production fluctuations may cause misjudgment (such as normal cells being mislabeled as runaway); on the other hand, they have poor adaptability to complex operating conditions, and there is a possibility of over-judgment (such as misjudging temporary fluctuations as runaway) and under-judgment (such as response delay to progressive micro short circuits).

[0005] (2) Existing methods also employ a one-time modeling strategy and combine multi-dimensional data for voltage runaway discrimination. Under this discrimination method: on the one hand, the mutual exclusion effect between multi-dimensional data may further reduce the reliability of the judgment, and when multi-dimensional parameters are introduced to improve accuracy, there is a lack of effective feature screening mechanism, and redundant parameters will interfere with the stability of the model; on the other hand, the static model cannot adapt to data drift during the manufacturing process (such as material batch differences and equipment aging), which leads to prediction distortion after long-term use.

[0006] (3) Existing methods also consider introducing an adaptive adjustment mechanism to enhance the stability of the discrimination model. However, this adaptive method relies on human experience to set adjustment rules and cannot respond in real time to temporal fluctuations in the production environment (such as seasonal temperature and humidity changes). Moreover, it lacks a closed-loop verification mechanism, which may cause secondary deviations during the correction process. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, this invention provides a lithium battery voltage runaway prevention method and system based on data scenario adaptation, which can solve the risk of robustness collapse of the model system used for lithium battery voltage runaway prediction caused by the cumulative data trend.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a method for preventing lithium battery voltage runaway based on data scenario adaptation.

[0009] A data-scenario-adaptive lithium battery voltage runaway prevention method includes: Collect multiple parameter tuples related to cell self-discharge, and select a set of key judgment parameters from them; Create calibrated abnormal battery cells and mix them with normal battery cells; collect process data corresponding to the key judgment parameter set and generate a calibration dataset. Based on the data scenario of the obtained calibration dataset, an initial discrimination model is constructed. Specifically: if the model is constructed for a small data volume scenario, an exponential kernel function is used to perform nonlinear dimensionality reduction on the normalized calibration dataset to construct the initial discrimination model; if the model is constructed for a large data volume scenario, the Kalman filter method is used to filter the normalized calibration dataset to construct the initial discrimination model. A time-series function-based monitoring model is used to monitor parameter fluctuations during cell self-discharge in real time. When parameter fluctuations reach the alarm conditions of the monitoring model, new data of the key judgment parameter set within the alarm time period are extracted to generate a corrected dataset, which is then input into the initial judgment model for training, in order to obtain a judgment model for predicting lithium battery voltage runaway.

[0010] Furthermore, abnormal cells are manufactured and mixed with normal cells, and process data corresponding to the key judgment parameter set are collected, including: mixing abnormal cells with normal cells for full process testing, applying physical markings to abnormal cells, and making them undergo formation, aging, and capacity testing processes synchronously with normal cells.

[0011] Furthermore, an exponential kernel function is used to perform nonlinear dimensionality reduction on the normalized calibration dataset to construct an initial discrimination model. This includes: first, dividing the calibration dataset into anomaly subsets and normal subsets, and then robustly normalizing the anomaly subsets and normal subsets; subsequently, nonlinear dimensionality reduction is performed using an exponential kernel function, and a classification function is constructed; the constructed classification function is used as the initial discrimination model.

[0012] Furthermore, the normalized calibration dataset is filtered using the Kalman filter method to construct an initial discrimination model. This includes: first, dividing the calibration dataset into anomaly subsets and normal subsets, and then performing Z-score normalization on both subsets; subsequently, using the normal subsets as the main data and the anomaly subsets as noise data, and segmenting them into multiple segments according to time sequence; finally, constructing the state equation and covariance update matrix based on the Kalman filter to build the initial discrimination model.

[0013] Furthermore, a monitoring model based on timing functions is used to monitor parameter fluctuations during cell self-discharge in real time. This includes: acquiring parameters of the fluctuation signal using timing functions, wherein the timing functions consist of a conventional timing monitoring and capture function based on an addition mode and a periodic change monitoring and capture function; wherein the conventional timing monitoring and capture function uses confirmation under infinitely continuous timing as the confirmation window, and the periodic change monitoring and capture function uses confirmation under periodic continuous timing as the confirmation window.

[0014] Furthermore, the alarm conditions of the monitoring model include: dividing the data of the key judgment parameter set into multiple modules, calculating the monitoring value and quantile interval of each module; when the monitoring value corresponding to the parameter fluctuation exceeds the quantile interval, the alarm conditions of the monitoring model are met.

[0015] Furthermore, in the process of obtaining the decision model for predicting lithium battery voltage runaway, the update of the decision model only modifies the weight parameters of the initial decision model, while the function structure and coefficients of the monitoring model remain fixed.

[0016] A second aspect of the present invention provides a lithium battery voltage runaway prevention system based on data scenario adaptation.

[0017] A data-driven, scenario-adaptive lithium battery voltage runaway prevention system includes: The key judgment parameter set determination module is configured to: collect multiple parameter tuples related to cell self-discharge and select the key judgment parameter set from them; The calibration module is configured to: create calibration abnormal cells and mix them with normal cells, collect process data corresponding to the key judgment parameter set, and generate a calibration dataset; The initial judgment model construction module is configured to: construct an initial judgment model based on the data scenario of the obtained calibration dataset. Specifically: if the model is constructed for a small data scenario, an exponential kernel function is used to perform nonlinear dimensionality reduction on the normalized calibration dataset to construct the initial judgment model; if the model is constructed for a large data scenario, the Kalman filter method is used to filter the normalized calibration dataset to construct the initial judgment model. The monitoring module is configured to use a time-series function-based monitoring model to monitor parameter fluctuations during cell self-discharge in real time. The lithium battery voltage runaway prevention module is configured to: when parameter fluctuations reach the alarm conditions of the monitoring model, extract new data of the key judgment parameter set within the alarm time period to generate a corrected dataset and input it into the initial judgment model for training, so as to obtain a judgment model for lithium battery voltage runaway prediction. A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the data scenario-adaptive lithium battery voltage runaway prevention method as described in the first aspect of the present invention.

[0018] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the data scenario-adaptive lithium battery voltage runaway prevention method as described in the first aspect of the present invention.

[0019] The above one or more technical solutions have the following beneficial effects: (1) This invention collects multiple parameter tuples related to cell self-discharge and calculates the covariance between each parameter tuple and the self-discharge data; it selects a set of key judgment parameters from the parameter tuples according to the magnitude of the covariance value, that is: based on covariance quantitative analysis, it screens out multiple core parameters strongly related to self-discharge, which can eliminate redundant dimension interference; subsequently, it uses the selected set of key judgment parameters to construct a sample library, which can ensure that the data used for subsequent training focuses more on effective features. Therefore, compared with the prior art, this invention can achieve multi-dimensional collaborative anti-interference judgment through this key judgment parameter screening mechanism, thereby avoiding the occurrence of problems such as misjudgment, over-judgment, and missed judgment.

[0020] (2) Based on the calibration dataset, this invention constructs an initial judgment model for voltage runaway by pre-setting judgment rules and loss functions; simultaneously, based on the key judgment parameter set, it constructs a monitoring model independent of the initial judgment model. The initial judgment model is used to process real-time classification, and the time series function is used to capture data drift based on the independent monitoring model. On the one hand, this invention no longer uses a one-time modeling model as the sole criterion, but continuously optimizes the judgment model after analyzing the time series data through the monitoring model. Therefore, even if multi-dimensional data is introduced, it will not affect the stability of the model. On the other hand, it also enables the judgment model to better adapt to data shift phenomena.

[0021] (3) When the parameter fluctuation reaches the alarm condition of the monitoring model, the present invention extracts new data of the key judgment parameter set within the alarm time state to generate a corrected dataset and inputs it into the initial judgment model for training, so as to obtain a judgment model for predicting lithium battery voltage runaway. The whole process realizes completely adaptive correction without human intervention, that is, through the additive mode fluctuation perception and quantile triggering mechanism, it responds to the time series fluctuation in real time and avoids secondary deviation.

[0022] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0024] Figure 1 This is a flowchart of the lithium battery voltage runaway prevention method based on data scenario adaptation in Embodiment 1 of the present invention.

[0025] Figure 2 This is a schematic diagram of the generation logic of the initial determination model in Embodiment 1 of the present invention.

[0026] Figure 3 This is a schematic diagram illustrating the generation logic of the monitoring model in Embodiment 1 of the present invention. Detailed Implementation

[0027] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0028] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0029] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0030] Example 1 This embodiment discloses a lithium battery voltage runaway prevention method based on data scenario adaptation.

[0031] like Figure 1 As shown, a data-scenario-adaptive lithium battery voltage runaway prevention method includes: Step S1: Collect multiple parameter tuples related to cell self-discharge and select the key judgment parameter set from them; Step S2: Create calibrated abnormal cells and mix them with normal cells; collect process data corresponding to the key judgment parameter set; and generate a calibration dataset. Step S3: Based on the data scenario of the obtained calibration dataset, construct an initial discrimination model. Specifically: if constructing for a small data scenario, use an exponential kernel function to perform nonlinear dimensionality reduction on the normalized calibration dataset to construct the initial discrimination model; if constructing for a large data scenario, use the Kalman filter method to filter the normalized calibration dataset to construct the initial discrimination model. Step S4: Use a time-series function-based monitoring model to monitor parameter fluctuations during cell self-discharge in real time; Step S5: When the parameter fluctuation reaches the alarm condition of the monitoring model, extract the new data of the key judgment parameter set within the alarm time period to generate a corrected dataset and input it into the initial judgment model for training, so as to obtain a judgment model for predicting lithium battery voltage runaway.

[0032] Based on the above process, this invention can solve the risk of robustness collapse of the model system used for predicting lithium battery voltage runaway caused by the cumulative effect of data trends. To facilitate understanding of the technical solution of this invention, the specific implementation methods of this invention will be further explained and described below.

[0033] In step S1, multiple parameter tuples related to cell self-discharge are collected, and the covariance between each parameter tuple and the self-discharge data is calculated; key decision parameter sets are selected from the parameter tuples according to the magnitude of the covariance value.

[0034] Collect multiple parameter tuples related to cell self-discharge, that is, collect parameters that can be collected automatically throughout the entire manufacturing process. By exploring the collection points, mark the parameters that may be related to cell self-discharge throughout the entire manufacturing process and name them accordingly. ;in, This indicates the measures taken related to cell self-discharge. A parameter tuple.

[0035] tuple of parameters Its corresponding self-discharge data Perform covariance calculation, i.e.: ; in, This indicates the calculation of covariance. Indicates the first The first parameter tuple in the nth parameter tuple The specific data point, that is, the parameter at the _th_ ... Observations at a sampling time or on a cell sample; Indicates the first Self-discharge data of individual battery cell samples.

[0036] After calculating the covariance between different parameter tuples and their corresponding self-discharge data, a set of key decision parameters is selected from the parameter tuples according to the magnitude of the covariance value. As an optional embodiment, the top 10 to 15 parameter tuples with the largest covariance values ​​can be selected as the set of key decision parameters.

[0037] In step S2, a calibrated abnormal battery cell is manufactured; the calibrated abnormal battery cell is mixed with normal battery cells and tested throughout the entire process, and process data corresponding to the key judgment parameter set is collected to generate a calibration dataset.

[0038] The process involves fabricating calibrated defective battery cells, which involves creating batches of defective samples based on the identified defect types of existing battery cells and marking them accordingly. In this embodiment, the fabricated defective battery cells mainly include various defect types such as diaphragm damage, electrode misalignment, and foreign object intrusion.

[0039] After the defective cells are manufactured, defect identification and data collection are required. The defective cells are then mixed with normal cells for full-process normal testing. The mixed cells are then used as normal products for the entire production line to obtain overall process parameters. Specifically, physical markings are applied to the defective cells, causing them to undergo formation, aging, and capacity testing processes simultaneously with normal cells, thus obtaining data for all cells. .

[0040] Based on step S2, this invention mixes abnormal cells with normal cells according to the production ratio, so that they can run synchronously in a series circuit. This can better capture the voltage fluctuation characteristics of abnormal cells in a real series environment. For example, after mixing abnormal cells with damaged separators, their voltage drop will be restrained by the adjacent normal cells. Unlike the "isolated fluctuation" when tested alone, this data can enable the model to more accurately identify runaway phenomena in actual production.

[0041] In step S3, an initial judgment model is constructed based on the data scenario of the obtained calibration dataset. Specifically: if the model is constructed for a small data volume scenario, an exponential kernel function is used to perform nonlinear dimensionality reduction on the normalized calibration dataset to construct the initial judgment model; if the model is constructed for a large data volume scenario, the Kalman filter method is used to filter the normalized calibration dataset to construct the initial judgment model. In actual implementation, if either the number of samples is less than 1000 or the dimension of the key judgment parameter is less than 10, it is considered to belong to a small data volume scenario; otherwise, it is considered to belong to a large data volume scenario. This can be achieved through the following methods: Step S3-1: When constructing the initial judgment model, if it is for a scenario with a small amount of data, firstly, the calibration dataset is divided into anomaly subsets. and normal subset Robust normalization is performed on abnormal and normal subsets; subsequently, nonlinear dimensionality reduction is performed using an exponential kernel function, and a classification function is constructed; the constructed classification function is used as the initial decision model (high discrimination capability), specifically as follows: Figure 2 As shown.

[0042] 1) Perform robust normalization on the calibration dataset, i.e.: ; in, This represents the value after robust normalization. Indicates the original value. This represents the median of a data series. It represents the interquartile range of a data series, i.e., the upper quartile minus the lower quartile.

[0043] 2) Subset of anomalies and normal subset To differentiate.

[0044] First, assume a plane, that is: ; in, Denotes the plane vector of the assumed plane. This represents the intercept.

[0045] Subsequently, based on the assumed plane, a data model is constructed, namely: .

[0046] 3) For abnormal subsets and normal subset Perform the transformation and determine the plane function based on the transformed parameters.

[0047] In actual production processes, most data is non-linear. Therefore, it is necessary to perform kernel function operations such as dimensionality increase and transformation on the data (abnormal subsets and normal subsets) to facilitate classification and solve the problem. ; ; in, This represents a constant used to confirm the coefficients of misallocation. In practical applications, the specific value of this constant can be set according to the actual situation. Indicates the first One interval floating parameter; This refers to the kernel function, which is a transformation function used for data transformation. The label represents the sample label. Abnormal cells are labeled with +1, and normal cells are labeled with -1. This label is used for classification training in supervised learning.

[0048] As an optional implementation, a polynomial or exponential scheme can be used when setting the transformation function, i.e.: A. Polynomial scheme ; in, This represents the multiplication factor, used to adjust the magnitude of the equation. This represents the constant term used for intercept adjustment. This represents the order of the polynomial used to control the complexity of the scheme.

[0049] B. Exponential Scheme .

[0050] When selecting a kernel function, the exponential approach is preferred due to its flexibility and strong discriminative power; when computational resources are limited, the polynomial approach can be chosen. After parameter transformation based on the kernel function, the transformed parameters are substituted to determine the planar function.

[0051] By mapping low-dimensional nonlinear data to a high-dimensional linearly separable space using an exponential kernel function, hidden nonlinear correlations (such as the implicit relationship between voltage fluctuation rate and cell defects) can be uncovered in small samples, allowing previously overlapping normal / abnormal samples to form clear boundaries in the high-dimensional space. This nonlinear dimensionality reduction capability endows the model with "high resolution," solving the problem of "linear models failing to fit nonlinear relationships" in existing technologies. Simultaneously, a classification function is constructed based on a "key decision parameter set" (strongly correlated parameters screened through covariance in the early stages). Combined with the amplification effect of the exponential kernel function on core features, effective features (such as self-discharge voltage difference and internal resistance change rate) can be focused in a limited number of samples, strengthening the characteristic differences between normal cells and calibrated abnormal cells (such as diaphragm damage and electrode misalignment). This combination of "feature focusing - nonlinear amplification" enables the model to accurately distinguish subtle anomalies (such as small voltage fluctuations in early micro-short circuits) even with small amounts of data, thus solving the problem of "poor model stability due to mutual exclusion or redundancy of multi-dimensional parameters" in existing technologies.

[0052] Step S3-2: When constructing the initial judgment model, if it is for a large data volume scenario, firstly, the calibration dataset is divided into anomaly subsets. and normal subset The abnormal and normal subsets are then Z-score normalized. Subsequently, the normal subset is used as the main data, and the abnormal subset as noise data, and the data is segmented into multiple segments according to time sequence. Finally, a state equation and covariance update matrix are constructed based on Kalman filtering to build an initial decision model (with high anti-interference capability) based on large-scale non-metric parameter types. Specifically, as follows... Figure 2 As shown.

[0053] Changing or increasing the number of groups in the calibration dataset can easily lead to overfitting due to dimensionality explosion. Therefore, dimensionality selection is necessary, based on the confirmation of covariance correlation. By confirming correlation in advance and comprehensively considering dimensionality, data failure caused by crude methods such as single-dimensional or multi-dimensional approaches can be prevented. Furthermore, external changes may occur during cell data collection (such as routine 4M changes), and the entire system state is divided into a hidden state X (unknown) and an observed state z (the total set of data displayed). Based on this, the initial judgment model can be constructed using the following methods.

[0054] 1) Normalize the standard deviation of the calibration dataset, i.e.: ; in, This represents the value after standard deviation normalization. This represents the mean. It represents the standard deviation.

[0055] 2) Define and distinguish between abnormal subsets and normal subsets.

[0056] Define normal subsets The data is the main data Define an exception subset The data is noise data. Master data With noise data Together they form the total data .

[0057] Master data The main data is divided according to the chronological order of the initial data, that is, the master data is divided into... ;in, Indicates the first [number] after time sequence division One master data; For noisy data The data is segmented according to the time sequence of the initial data, that is, the noisy data is divided into segments sequentially. ;in, Indicates the first [number] after time sequence division One noise data point; For total data The data is divided according to the time sequence of the initial data, that is, the total data is divided sequentially into... ;in, Indicates the first [number] after time sequence division Total data.

[0058] 3) Construct the prediction function and prediction covariance.

[0059] according to Predicting states and control variables in time The state of time is used to construct the prediction function, i.e.: ; in, Indicates based on The state at any given moment The predicted state at a given time (such as the estimated value of the cell voltage), i.e., the prior estimate in Kalman filtering; Indicates the first Total data The state transition mode, i.e., the transformation matrix or vector state; express The external known transformation matrix under time conditions, express The true state of data in time.

[0060] The covariance prediction model used to predict covariance can be expressed as: ; in, express The covariance matrix of the predicted state at any given time is used to measure the uncertainty of the predicted value. The larger the matrix element, the greater the error fluctuation of the corresponding state. express For noisy data in time The covariance of the predicted values.

[0061] Based on this, the first Total data Represented as: ;in, express The observation matrix in time, i.e. the Gaussian distribution model corresponding to the master data.

[0062] 4) Construct the update function and predict the covariance.

[0063] according to Calculate the observed and predicted values ​​at time 1. The state estimate at time t, i.e.: ; in, express Weights of predicted and observed values ​​in different time periods.

[0064] After the covariance update, it can be represented as: ; ; in, express.

[0065] Therefore, the probability covariance matrix Updated weight values Optimal state estimation of the entire system at each time step It is completed through the superposition of prediction and update. Tense and Temporal Linkage. Through the temporal linking of "prediction-update," the Kalman filter ultimately outputs the optimal state estimates at each time step. Covariance Matrix These parameters together constitute the basic parameters of the initial discrimination model. Based on this, the model can achieve voltage runaway discrimination in the following way: For new cell time-series data, the state equation (a dynamic model based on temporal linkage) is used to predict the parameter evolution trend; the covariance matrix is ​​combined to determine whether the deviation between the actual observed value and the predicted value exceeds the "normal fluctuation range" (defined by the normal subset time-series pattern of the calibration dataset); if the deviation exceeds the range, it is determined to be a risk of voltage runaway. Therefore, Tense and The temporal linkage reflects the adaptability of the initial discriminant model to complex operating conditions.

[0066] Step S3-2, through normalization, time-series segmentation, and Kalman filter dynamic modeling adapted to the time-series characteristics of big data, can specifically address the core problems of existing technologies in big data scenarios, such as "parameter scale interference, neglect of time-series patterns, insufficient data drift adaptation, and weak noise resistance." Ultimately, it constructs an initial judgment model with high anti-interference capability, providing a reliable foundation for the accurate prevention of lithium battery voltage runaway.

[0067] In step S4, a monitoring model based on time-series functions is used to monitor the parameter fluctuations during cell self-discharge in real time.

[0068] By constructing a time-series function for monitoring, an automated adaptive model, namely the monitoring model, is built using the obtained key decision parameter set. The monitoring model is predicated on the perception of correction requirements; that is, by constructing a time-related function, process fluctuations are perceived, and the fluctuation characteristics are converted into correction signals to initiate corrections. These signals are then used to confirm alarms for subsequent data. After an alarm occurs, the data within the timeframe is automatically fitted, and the process returns to step S4 for re-determination. The monitoring model remains unchanged. The existence of the adaptive model avoids false interference from periodic, regular fluctuations, and the control of nodes further reduces the possibility of missed detections of truly problematic data.

[0069] The monitoring model uses time-series functions to monitor parameter fluctuations in real time, including: using time-series functions to collect parameters from fluctuation signals. The time-series functions consist of a conventional timed monitoring and capture function based on an addition mode and a periodic change monitoring and capture function. The conventional timed monitoring and capture function uses confirmation under infinitely continuous time series as the confirmation window, while the periodic change monitoring and capture function uses confirmation under periodic continuous time series as the confirmation window.

[0070] Therefore, the monitoring model is independent of the initial decision model, capturing data fluctuations only through time-series functions. When a correction is triggered, only the weights of the decision model are adjusted, without changing the structure of the monitoring function. For example, when the decision model drifts due to equipment aging, the monitoring model can still identify anomalies and trigger corrections through time-series fluctuations, forming a "monitoring-correction" closed loop. Compared to existing coupled models, the system is more robust.

[0071] The conventional timed monitoring capture function and the periodic change monitoring capture function based on the addition pattern can be expressed as: ; in, This represents the timing function in addition mode. This indicates a regular, timed monitoring and capture function. This represents a function for monitoring and capturing periodic changes.

[0072] Regular timed monitoring capture function The confirmation window is set to open under an infinitely continuous time sequence, i.e.: ; in, This represents the trigger function, which is set to 1 if the time exceeds the confirmation window time, and 0 otherwise. This represents the rate of change compared to the median value of the data in the previous iteration of the model.

[0073] Periodic change monitoring and capture function The confirmation window is set based on the periodic continuation time series, i.e.: ; in, Indicates harmonic frequency, Indicates the period value; in this embodiment, the harmonic frequency The cycle value is one year, that is... . , These represent the first and second digits of a periodic function. The cosine and sine coefficients of the second harmonic are obtained through least squares fitting and are used to quantify the amplitude and phase of the periodic fluctuations. Furthermore, , satisfy: .

[0074] In step S5, when the parameter fluctuation reaches the alarm condition of the monitoring model, new data of the key judgment parameter set within the alarm time period is extracted to generate a corrected dataset and input into the initial judgment model for training, so as to obtain a judgment model for predicting lithium battery voltage runaway.

[0075] The alarm conditions of the monitoring model include: dividing the data of the key judgment parameter set into multiple modules, calculating the monitoring value and quantile interval of each module; when the monitoring value corresponding to the parameter fluctuation exceeds the quantile interval, the alarm condition of the monitoring model is met. Therefore, quantile calculation no longer relies on assumptions such as Gaussian distribution, making it particularly suitable for asymmetric distributions of lithium battery parameters (such as the right-skewed distribution of self-discharge rate), thus improving robustness compared to fixed thresholds based on mean-standard deviation.

[0076] Specifically, by performing time-series segmentation on the original data used to form the model, the data is segmented into... Timing functions are applied to each module (30-50 modules). Sampling is used to form a trajectory, i.e.: ; in, Indicates the first The timing monitoring function value of each module is used to reflect the fluctuation trajectory of the module parameters over time.

[0077] For each predicted point, calculate the trajectory and fractional place, i.e.: ; ; ; in, This indicates the upper limit of the trigger correction signal. This indicates the lower limit for triggering the correction signal. For example... Figure 3 As shown, when the monitored value corresponding to the parameter fluctuation exceeds the quantile interval (i.e., after verification by the monitoring model, it exceeds the upper and lower limits of the trigger correction signal), it is considered that the trigger condition has been met. New data of the key judgment parameter set within the alarm time can be extracted to generate a correction dataset (i.e., current mass production data + labeled data) and input into the initial judgment model for training, so as to obtain a new judgment model for predicting lithium battery voltage runaway. That is, it can be judged according to the weight collection rules in the future. After the correction requirement is determined, the random judgment model formed by the trigger correction area data is used as the new judgment model.

[0078] It's also important to note that during the process of obtaining the judgment model for predicting lithium battery voltage runaway, updates to the judgment model only modify the weight parameters of the initial judgment model; the function structure and coefficients of the monitoring model remain fixed. Maintaining a fixed monitoring model ensures consistency in alarm criteria. For example, regardless of how the judgment model is updated, the monitoring model can always fit temperature and humidity fluctuations on an annual cycle, thus avoiding alarm logic confusion caused by model adjustments.

[0079] Example 2 This embodiment discloses a lithium battery voltage runaway prevention system based on data scenario adaptation.

[0080] A data-driven, scenario-adaptive lithium battery voltage runaway prevention system includes: The key decision parameter set determination module is configured to: collect multiple parameter tuples related to cell self-discharge, and calculate the covariance between each parameter tuple and the self-discharge data; select the key decision parameter set from the parameter tuples according to the magnitude of the covariance value. The calibration module is configured to: manufacture calibration abnormal cells; mix the calibration abnormal cells with normal cells and perform full-process testing, collect process data corresponding to the key judgment parameter set, and generate a calibration dataset; The initial judgment model construction module is configured to: construct an initial judgment model for voltage runaway based on the obtained calibration dataset and through preset judgment rules and loss functions; The monitoring model construction module is configured to: construct a monitoring model independent of the initial judgment model based on the obtained key judgment parameter set; the monitoring model uses a time series function to monitor parameter fluctuations in real time. The lithium battery voltage runaway prevention module is configured to: when parameter fluctuations reach the alarm conditions of the monitoring model, extract new data of the key judgment parameter set within the alarm time period to generate a corrected dataset and input it into the initial judgment model for training, so as to obtain a judgment model for lithium battery voltage runaway prediction. Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0081] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the data scenario-adaptive lithium battery voltage runaway prevention method as described in Embodiment 1 of this disclosure.

[0082] Example 4 The purpose of this embodiment is to provide an electronic device.

[0083] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the data scenario-adaptive lithium battery voltage runaway prevention method as described in Embodiment 1 of this disclosure.

[0084] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0085] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0086] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A lithium battery voltage runaway prevention method based on data scenario adaptation, characterized in that, include: Collect multiple parameter tuples related to cell self-discharge, and select a set of key judgment parameters from them; The abnormal battery cells are manufactured and mixed with normal battery cells. Process data corresponding to the key judgment parameter set are collected to generate a calibration dataset. Specifically, the abnormal battery cells are mixed with normal battery cells and tested throughout the entire process. Physical markings are applied to the abnormal battery cells so that they undergo the formation, aging and capacity testing processes synchronously with normal battery cells. Based on the data scenario of the obtained calibration dataset, an initial judgment model is constructed. Specifically: if the model is constructed for a small data volume scenario, an exponential kernel function is used to perform nonlinear dimensionality reduction on the normalized calibration dataset to construct the initial judgment model; if the model is constructed for a large data volume scenario, the Kalman filter method is used to filter the normalized calibration dataset to construct the initial judgment model. A monitoring model based on time-series functions is used to monitor parameter fluctuations during cell self-discharge in real time. Specifically, a time-series function is used to collect parameters from the fluctuation signal. The time-series function consists of a conventional timing monitoring and capture function based on an addition mode and a periodic change monitoring and capture function. The conventional timing monitoring and capture function uses the confirmation period under an infinitely continuous time sequence as the confirmation window, while the periodic change monitoring and capture function uses the confirmation period under a periodic continuous time sequence as the confirmation window. When parameter fluctuations reach the alarm conditions of the monitoring model, new data of the key judgment parameter set within the alarm time period are extracted to generate a corrected dataset and input into the initial judgment model for training, so as to obtain a judgment model for predicting lithium battery voltage runaway; in the process of obtaining the judgment model for predicting lithium battery voltage runaway, the update of the judgment model only corrects the weight parameters of the initial judgment model, and the function structure and coefficients of the monitoring model remain fixed. The alarm conditions of the monitoring model include: dividing the data of the key judgment parameter set into multiple modules, calculating the monitoring value and quantile interval of each module; when the monitoring value corresponding to the parameter fluctuation exceeds the quantile interval, the alarm conditions of the monitoring model are met.

2. The lithium battery voltage runaway prevention method based on data scenario adaptation as described in claim 1, characterized in that, An exponential kernel function is used to perform nonlinear dimensionality reduction on the normalized calibration dataset to construct an initial decision model. This includes: first, dividing the calibration dataset into anomaly subsets and normal subsets, and performing robust normalization on the anomaly subsets and normal subsets; then, performing nonlinear dimensionality reduction using an exponential kernel function and constructing a classification function; and finally, using the constructed classification function as the initial decision model.

3. The lithium battery voltage runaway prevention method based on data scenario adaptation as described in claim 1, characterized in that, The normalized calibration dataset is filtered using the Kalman filter method to construct an initial decision model. This process includes: first, dividing the calibration dataset into anomaly subsets and normal subsets, and then performing Z-score normalization on both subsets; subsequently, using the normal subsets as the main data and the anomaly subsets as noise data, and segmenting them into multiple segments according to time sequence; finally, constructing the state equation and covariance update matrix based on the Kalman filter to build the initial decision model.

4. A lithium battery voltage runaway prevention system based on data scenario adaptation, employing the lithium battery voltage runaway prevention method as described in any one of claims 1-3, characterized in that, include: The key judgment parameter set determination module is configured to: collect multiple parameter tuples related to cell self-discharge and select the key judgment parameter set from them; The calibration module is configured to: create calibration abnormal cells and mix them with normal cells, collect process data corresponding to the key judgment parameter set, and generate a calibration dataset; The initial judgment model construction module is configured to: construct an initial judgment model based on the data scenario of the obtained calibration dataset. Specifically: if the model is constructed for a small data scenario, an exponential kernel function is used to perform nonlinear dimensionality reduction on the normalized calibration dataset to construct the initial judgment model; if the model is constructed for a large data scenario, the Kalman filter method is used to filter the normalized calibration dataset to construct the initial judgment model. The monitoring module is configured to: use a time-series function-based monitoring model to monitor parameter fluctuations during cell self-discharge in real time; The lithium battery voltage runaway prevention module is configured to: when parameter fluctuations reach the alarm conditions of the monitoring model, extract new data of the key judgment parameter set within the alarm time period to generate a corrected dataset and input it into the initial judgment model for training, so as to obtain a judgment model for lithium battery voltage runaway prediction.

5. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the data scenario-adaptive lithium battery voltage runaway prevention method as described in any one of claims 1-3.

6. An electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the data scenario-adaptive lithium battery voltage runaway prevention method as described in any one of claims 1-3.

Images