A power battery thermal runaway grading early warning and active cut-off control method

Abstract

The application discloses a power battery thermal runaway grading early warning and active cut-off control method, which sequentially executes the following steps: real-time collection of state parameter sets containing single cell voltage, battery module temperature, internal resistance change rate and characteristic gas concentration and delivery to a data preprocessing module; filtering and normalization processing of the state parameter sets, output of a feature vector and delivery to a thermal runaway grading early warning module and an active cut-off judgment module respectively; output of an early warning level signal according to the feature vector by a pre-training classification model built in the grading early warning module and synchronization to a man-machine interface and the cut-off judgment module; when the risk is determined to reach a preset threshold, the cut-off judgment module generates a cut-off instruction and controls an execution unit to disconnect a high-pressure main loop. The application realizes linkage of thermal runaway grading early warning and active cut-off, and improves the operation safety of the power battery.

Description

Technical Field

[0001] This invention belongs to the field of power batteries, specifically referring to a method for graded early warning and active disconnection control of thermal runaway in power batteries. Background Technology

[0002] Early warning and rapid disconnection of thermal runaway in power batteries are crucial for ensuring the safety of new energy vehicles. In related methods, the warning module and the disconnection control module typically operate independently: the warning module outputs alarm signals based on thresholds or simple models, while the disconnection decision still requires manual confirmation or relies on independent judgment logic. This information fragmentation leads to two prominent problems: first, the warning signal fails to directly and promptly drive the disconnection action, resulting in a delayed response; second, the disconnection decision does not fully utilize the risk evolution information contained in the tiered warning system, easily leading to premature or delayed disconnection. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, this invention proposes a graded early warning and active cut-off control method for thermal runaway of power batteries.

[0004] An embodiment of the present invention provides a graded early warning and active disconnection control method for thermal runaway of a power battery, comprising the following steps performed in sequence:

[0005] Step S1: Real-time acquisition of the state parameter set of the power battery, and transmission of the state parameter set to the data preprocessing module; the state parameter set includes at least single cell voltage, battery module temperature, internal resistance change rate and characteristic gas concentration;

[0006] Step S2: The data preprocessing module filters and normalizes the received state parameter set, and sends the processed feature vectors to the thermal runaway graded early warning module and the active cut-off judgment module, respectively.

[0007] Step S3: The thermal runaway graded early warning module has a built-in pre-trained classification model. The classification model outputs the corresponding early warning level signal according to the feature vector and sends the early warning level signal to the human-machine interface. At the same time, the early warning level signal is sent to the active cut-off judgment module.

[0008] Step S4: The active cut-off judgment module receives the feature vector and the warning level signal. When it determines that the current thermal runaway risk reaches the preset cut-off threshold, it generates an active cut-off command and sends it to the execution unit of the battery management system, which disconnects the high-voltage main circuit.

[0009] According to embodiments of the present invention, the method proposed by the present invention has the following beneficial effects:

[0010] The early warning and cut-off closed-loop linkage enables a faster response: In step S3, the early warning level signal output by the classification model is simultaneously sent to the human-machine interface and the active cut-off judgment module. In step S4, the cut-off module directly makes a judgment based on the signal without manual intervention or independent logic secondary confirmation, which significantly shortens the response delay from risk identification to high-voltage circuit disconnection.

[0011] Cut-off decision integrates multi-dimensional information for more accurate judgment: In step S2, the preprocessed feature vector is sent to the early warning module and the cut-off module respectively. In step S4, the cut-off module simultaneously receives the original feature vector and the early warning level signal, so that the cut-off conditions can comprehensively utilize the individual voltage, temperature, internal resistance change rate, gas concentration and risk evolution level, avoiding false cut-off or delayed cut-off caused by a single threshold or independent model.

[0012] The decoupling of tiered early warning and active cut-off responsibilities makes the system more reliable: Step S3 is only responsible for outputting multi-level early warning levels, while Step S4 independently determines whether the preset cut-off threshold has been reached. The two communicate with each other but are logically separated, which not only retains the flexible display function of tiered early warning, but also ensures that the cut-off action is triggered only when the threshold condition is met, thus reducing the risk of false actions.

[0013] Classification model drives multi-parameter fusion, making it more adaptable to working conditions: Step S3 uses a pre-trained classification model to replace the traditional fixed threshold method, which can dynamically output the risk level according to the changing trend of multi-dimensional state parameters, and maintain a low false alarm rate and false negative rate under complex working conditions.

[0014] According to some embodiments of the present invention, the classification model in step S3 is a gradient boosting tree model based on the XGBoost framework. The gradient boosting tree model takes the feature vector as input and outputs three warning levels, where the first warning level indicates the initial appearance of risk, the second warning level indicates the risk is aggravated, and the third warning level indicates the critical state of thermal runaway.

[0015] According to some embodiments of the present invention, the training dataset of the classification model includes time-series segments of cell voltage, battery module temperature, internal resistance change rate, and characteristic gas concentration from historical thermal runaway events, with labels indicating the actual thermal runaway level at each time point.

[0016] According to some embodiments of the present invention, the data preprocessing module in step S2 uses a sliding window midpoint filtering algorithm to filter out impulse noise in the state parameter set, and uses the Z-score normalization method to uniformly map each parameter to a standard normal distribution.

[0017] According to some embodiments of the present invention, the thermal runaway classification early warning module in step S3, while outputting the early warning level signal, stores the feature vector of the early warning trigger time and the early warning level information into the historical database for the classification model to perform incremental learning and offline updates.

[0018] According to some embodiments of the present invention, after generating the active cut-off command, the active cut-off judgment module in step S4 simultaneously sends enable signals for prohibiting charging and discharging to the vehicle controller, and records the timestamp of the cut-off event and a snapshot of the current state parameters.

[0019] According to some embodiments of the present invention, the sampling frequency of the state parameter set collected in step S1 is not less than 10 Hz; when the thermal runaway graded early warning module outputs a level 2 or level 3 early warning signal, the sampling frequency is dynamically increased to not less than 50 Hz, and the increased sampling data is transmitted back to the data preprocessing module in real time.

[0020] According to some embodiments of the present invention, the triggering conditions for the first-level warning signal are: the single-cell voltage is below 2.8 volts or above 4.3 volts, or the battery module temperature is above 55 degrees Celsius;

[0021] The triggering conditions for the secondary warning signal are: a voltage drop rate greater than 0.5 volts per second, or a temperature rise rate greater than 5 degrees Celsius per second;

[0022] The triggering conditions for the Level 3 warning signal are: the rate of change of internal resistance is greater than 50% of the initial value, and the concentration of the characteristic gas is higher than 50 ppm.

[0023] According to some embodiments of the present invention, before outputting the active cut-off command, the active cut-off judgment module in step S4 performs redundancy verification: the feature vector is simultaneously input to an independent backup logic judgment unit, the output of the backup logic judgment unit is logically ANDed with the judgment result of the active cut-off judgment module, and when both are true, the active cut-off command is output.

[0024] According to some embodiments of the present invention, after the execution unit disconnects the high-voltage main circuit in step S4, a self-locking step is further included: the execution unit automatically locks its disconnected state, prohibits automatic reset, and restores the conduction state of the high-voltage main circuit through a manual reset operation.

[0025] Additional aspects and advantages 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

[0026] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0027] Figure 1 These are logic operation diagrams of some embodiments of the present invention;

[0028] Figure 2 This is a flowchart illustrating the implementation of step S3 in some embodiments of the present invention;

[0029] Figure 3 This is a signal transmission diagram of step S4 in some embodiments of the present invention.

[0030] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0031] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0032] The following is in conjunction with the appendix Figures 1-3 The present invention will be described clearly, completely and in detail.

[0033] refer to Figures 1-3 An embodiment of the present invention provides a graded early warning and active disconnection control method for thermal runaway of a power battery, comprising the following steps performed in sequence:

[0034] Step S1: Real-time acquisition of the state parameter set of the power battery, and transmission of the state parameter set to the data preprocessing module; the state parameter set includes at least single cell voltage, battery module temperature, internal resistance change rate and characteristic gas concentration;

[0035] Step S2: The data preprocessing module filters and normalizes the received state parameter set, and sends the processed feature vectors to the thermal runaway graded early warning module and the active cut-off judgment module, respectively.

[0036] Step S3: The thermal runaway graded early warning module has a built-in pre-trained classification model. The classification model outputs the corresponding early warning level signal according to the feature vector and sends the early warning level signal to the human-machine interface. At the same time, the early warning level signal is sent to the active cut-off judgment module.

[0037] Step S4: The active cut-off judgment module receives the feature vector and the warning level signal. When it determines that the current thermal runaway risk reaches the preset cut-off threshold, it generates an active cut-off command and sends it to the execution unit of the battery management system, which disconnects the high-voltage main circuit.

[0038] According to embodiments of the present invention, the method proposed by the present invention has the following beneficial effects:

[0039] The early warning and cut-off closed-loop linkage enables a faster response: In step S3, the early warning level signal output by the classification model is simultaneously sent to the human-machine interface and the active cut-off judgment module. In step S4, the cut-off module directly makes a judgment based on the signal without manual intervention or independent logic secondary confirmation, which significantly shortens the response delay from risk identification to high-voltage circuit disconnection.

[0040] Cut-off decision integrates multi-dimensional information for more accurate judgment: In step S2, the preprocessed feature vector is sent to the early warning module and the cut-off module respectively. In step S4, the cut-off module simultaneously receives the original feature vector and the early warning level signal, so that the cut-off conditions can comprehensively utilize the individual voltage, temperature, internal resistance change rate, gas concentration and risk evolution level, avoiding false cut-off or delayed cut-off caused by a single threshold or independent model.

[0041] The specific execution process of the data preprocessing module in step S2 includes the following sub-steps, clarifying the processing flow, processing standards, and feature vector transmission logic of the state parameter set:

[0042] S2.1 Status Parameter Set Reception

[0043] The data preprocessing module receives the power battery status parameter set collected in real time in step S1. The parameter set includes four core parameters: single cell voltage, battery module temperature, internal resistance change rate, and characteristic gas concentration. At the same time, it performs preliminary verification on the received parameter signals, and removes obviously abnormal invalid signals (such as parameter values ​​that exceed the reasonable range) to ensure the validity of the input data.

[0044] S2.2 Filtering Process

[0045] The verified state parameter set is filtered using a low-pass filtering algorithm to remove high-frequency noise (such as electromagnetic interference and fluctuations caused by acquisition equipment errors) generated during parameter acquisition, while preserving the true trend of parameter changes. This makes the parameter data more stable and provides an accurate basis for subsequent normalization processing.

[0046] S2.3 Normalization Processing

[0047] Normalization is performed on the filtered state parameters to map parameters with different dimensions and ranges (such as single-unit voltage in V and temperature in ℃) to the [0,1] interval, eliminating the influence of dimensional differences and giving each parameter equal weight, which facilitates the calculation and judgment of the subsequent classification model.

[0048] The decoupling of tiered early warning and active cut-off responsibilities makes the system more reliable: Step S3 is only responsible for outputting multi-level early warning levels, while Step S4 independently determines whether the preset cut-off threshold has been reached. The two communicate with each other but are logically separated, which not only retains the flexible display function of tiered early warning, but also ensures that the cut-off action is triggered only when the threshold condition is met, thus reducing the risk of false actions.

[0049] Classification model drives multi-parameter fusion, making it more adaptable to working conditions: Step S3 uses a pre-trained classification model to replace the traditional fixed threshold method, which can dynamically output the risk level according to the changing trend of multi-dimensional state parameters, and maintain a low false alarm rate and false negative rate under complex working conditions.

[0050] According to an embodiment of the present invention, the classification model in step S3 is a gradient boosting tree model based on the XGBoost framework. The gradient boosting tree model takes the feature vector as input and outputs three warning levels, where the first warning level indicates the initial appearance of risk, the second warning level indicates the risk is aggravated, and the third warning level indicates the critical state of thermal runaway.

[0051] The classification model used in step S3 is a gradient boosting tree model built on the XGBoost (Extreme Gradient Boosting) framework. This model takes the feature vectors extracted and processed in step S2 as input and outputs a three-level warning level through multiple rounds of gradient boosting iterations. The optimized sub-steps are as follows:

[0052] S3.1 Model Initialization

[0053] First, the initial configuration of the XGBoost gradient boosting tree model is completed, clarifying the core basic parameters of the model, including the total number of decision trees, the depth of a single decision tree, and the learning rate. Simultaneously, an initial prediction model is constructed, employing a constant prediction method, using the average warning level label of all training samples as the initial prediction result, laying the foundation for subsequent iterative training.

[0054] S3.2 Gradient Calculation and Residual Solution

[0055] In the model iteration phase, for each iteration, the deviation (i.e., residual) between the current model's prediction and the actual warning level label of the sample is calculated. This residual is quantified using the negative gradient of the loss function. This residual is used as the training target for the next decision tree, clarifying the direction of subsequent model optimization and ensuring that each iteration can specifically compensate for the prediction deviation of the previous round.

[0056] S3.3 Construct a decision tree and update the model

[0057] Using the residuals obtained from S3.2 as the training objective, a single CART decision tree is constructed using the input feature vectors as a base learner. A greedy algorithm is used to rationally partition the decision tree nodes, minimizing the residual fluctuations of samples within each node. Subsequently, a learning rate is introduced to control the contribution of this base learner to the overall model, avoiding overfitting. At the same time, the prediction results of this base learner are incorporated into the current model to update the model's predictive capabilities.

[0058] S3.4 Iteration Termination and Prediction Output

[0059] Repeat the iterative process from S3.2 to S3.3 until the preset termination condition is met—either the set total number of decision trees is reached, or the model prediction error is reduced to below a preset threshold. At this point, stop the iteration and determine the final model. Input the feature vector of the sample to be detected into the final model, and output the warning prediction score corresponding to the sample through model calculation.

[0060] S3.5 Warning Level Determination

[0061] Based on the predicted scores output by the model, a reasonable range of grading thresholds is preset, and the predicted scores are mapped to the corresponding three-level warning levels. The specific judgment criteria are as follows:

[0062] Level 1 Warning (Initial Risk): The predicted score is in the lowest threshold range. At this time, the number of abnormal indicators reflected by the feature vector is small, the risk is in the nascent stage, and it has not had a significant impact on the operation of the system.

[0063] Level 2 Warning (Risk Escalation): The predicted score is in the middle threshold range. At this time, the number of abnormal indicators increases and the risk shows a continuous escalation trend. Preliminary prevention and control measures need to be taken in a timely manner.

[0064] Level 3 Warning (Critical State of Thermal Runaway): The predicted score is at or above the highest threshold range. At this point, the core abnormal indicators have reached the critical value, and the risk is extremely explosive. Emergency response procedures must be initiated immediately.

[0065] Note: The grading thresholds corresponding to the warning levels are determined through professional methods such as ROC curve analysis and cross-validation of training samples to ensure the accuracy and reliability of the warning judgment and to meet the needs of actual application scenarios.

[0066] According to an embodiment of the present invention, the training dataset of the classification model includes time-series segments of cell voltage, battery module temperature, internal resistance change rate, and characteristic gas concentration from historical thermal runaway events, with labels indicating the actual thermal runaway level at each time point.

[0067] According to an embodiment of the present invention, the data preprocessing module in step S2 uses a sliding window midpoint filtering algorithm to filter out impulse noise in the state parameter set, and uses the Z-score normalization method to uniformly map each parameter to a standard normal distribution.

[0068] According to an embodiment of the present invention, the thermal runaway classification early warning module in step S3, while outputting the early warning level signal, stores the feature vector of the early warning trigger time and the early warning level information into the historical database for the classification model to perform incremental learning and offline updates.

[0069] According to an embodiment of the present invention, after generating the active cut-off command, the active cut-off judgment module in step S4 simultaneously sends enable signals for prohibiting charging and discharging to the vehicle controller, and records the timestamp of the cut-off event and a snapshot of the current state parameters.

[0070] The specific execution process of the active cut-off judgment module in step S4 includes the following sub-steps, clarifying the complete process of signal transmission and event recording:

[0071] S4.1 Signal Reception and Risk Assessment

[0072] The active cut-off judgment module synchronously receives the feature vector output in step S2 and the warning level signal output in step S3, and combines it with the preset thermal runaway risk cut-off threshold to make a comprehensive judgment on the current thermal runaway risk of the power battery.

[0073] S4.2 Active disconnection command generation

[0074] When the current thermal runaway risk is determined to reach the preset cutoff threshold, the active cutoff judgment module immediately generates an active cutoff command and sends the command to the execution unit of the battery management system to trigger the high-voltage main circuit disconnection operation.

[0075] S4.3 Vehicle Controller Signal Transmission

[0076] While generating the active cut-off command, the active cut-off judgment module sends dual enable signals to the vehicle controller, namely a charging prohibition enable signal and a discharging prohibition enable signal, to ensure that the vehicle temporarily stops charging and discharging operations, further avoiding safety risks.

[0077] S4.4 Cut off event information recording

[0078] The active disconnection judgment module synchronously records key information of this disconnection event, including the timestamp of the disconnection event (accurate to the millisecond level) and a snapshot of the power battery's state parameters at the moment of disconnection (covering core parameters such as single cell voltage, battery module temperature, internal resistance change rate, and characteristic gas concentration), providing data support for subsequent fault investigation and risk review.

[0079] According to an embodiment of the present invention, the sampling frequency of the state parameter set collected in step S1 is not less than 10 Hz; when the thermal runaway graded early warning module outputs a level 2 or level 3 early warning signal, the sampling frequency is dynamically increased to not less than 50 Hz, and the increased sampling data is transmitted back to the data preprocessing module in real time.

[0080] According to an embodiment of the present invention, the triggering condition for the first-level warning signal is: the single cell voltage is lower than 2.8 volts or higher than 4.3 volts, or the battery module temperature is higher than 55 degrees Celsius;

[0081] The triggering conditions for the secondary warning signal are: a voltage drop rate greater than 0.5 volts per second, or a temperature rise rate greater than 5 degrees Celsius per second;

[0082] The triggering conditions for the Level 3 warning signal are: the rate of change of internal resistance is greater than 50% of the initial value, and the concentration of the characteristic gas is higher than 50 ppm.

[0083] According to an embodiment of the present invention, before outputting the active cut-off command, the active cut-off judgment module in step S4 performs redundancy verification: the feature vector is simultaneously input to an independent backup logic judgment unit, the output of the backup logic judgment unit is logically ANDed with the judgment result of the active cut-off judgment module, and when both are true, the active cut-off command is output.

[0084] According to an embodiment of the present invention, after the execution unit disconnects the high-voltage main circuit in step S4, a self-locking step is further included: the execution unit automatically locks its disconnected state, prohibits automatic reset, and restores the conduction state of the high-voltage main circuit through a manual reset operation.

[0085] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for graded early warning and active disconnection control of thermal runaway in power batteries, characterized in that, The steps are executed in the following order: Step S1: Real-time acquisition of the state parameter set of the power battery, and transmission of the state parameter set to the data preprocessing module; the state parameter set includes at least single cell voltage, battery module temperature, internal resistance change rate and characteristic gas concentration; Step S2: The data preprocessing module filters and normalizes the received state parameter set, and sends the processed feature vectors to the thermal runaway graded early warning module and the active cut-off judgment module, respectively. Step S3: The thermal runaway graded early warning module has a built-in pre-trained classification model. The classification model outputs the corresponding early warning level signal according to the feature vector and sends the early warning level signal to the human-machine interface. At the same time, the early warning level signal is sent to the active cut-off judgment module. Step S4: The active cut-off judgment module receives the feature vector and the warning level signal. When it determines that the current thermal runaway risk reaches the preset cut-off threshold, it generates an active cut-off command and sends it to the execution unit of the battery management system, which disconnects the high-voltage main circuit.

2. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 1, characterized in that, The classification model in step S3 is a gradient boosting tree model based on the XGBoost framework. The gradient boosting tree model takes the feature vector as input and outputs three warning levels, where the first warning level indicates the initial appearance of risk, the second warning level indicates the risk is aggravated, and the third warning level indicates the critical state of thermal runaway.

3. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 2, characterized in that, The training dataset for the classification model includes time-series segments of cell voltage, battery module temperature, internal resistance change rate, and characteristic gas concentration from historical thermal runaway events, with labels indicating the actual thermal runaway level at each time point.

4. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 1, characterized in that, The data preprocessing module in step S2 uses a sliding window midpoint filtering algorithm to filter out impulse noise in the state parameter set, and uses the Z-score normalization method to uniformly map each parameter to a standard normal distribution.

5. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 1, characterized in that, In step S3, the thermal runaway graded early warning module outputs an early warning level signal and stores the feature vector of the early warning trigger time and the early warning level information into a historical database for incremental learning and offline updates by the classification model.

6. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 1, characterized in that, In step S4, after generating the active cut-off command, the active cut-off judgment module simultaneously sends enable signals for prohibiting charging and discharging to the vehicle controller, and records the timestamp of the cut-off event and a snapshot of the current state parameters.

7. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 1, characterized in that, In step S1, the sampling frequency of the state parameter set is not less than 10 Hz; when the thermal runaway graded early warning module outputs a level 2 or level 3 early warning signal, the sampling frequency is dynamically increased to not less than 50 Hz, and the increased sampling data is sent back to the data preprocessing module in real time.

8. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 2, characterized in that, The triggering conditions for the first-level warning signal are: the voltage of a single cell is lower than 2.8 volts or higher than 4.3 volts, or the temperature of the battery module is higher than 55 degrees Celsius; The triggering conditions for the secondary warning signal are: a voltage drop rate greater than 0.5 volts per second, or a temperature rise rate greater than 5 degrees Celsius per second; The triggering conditions for the Level 3 warning signal are: the rate of change of internal resistance is greater than 50% of the initial value, and the concentration of the characteristic gas is higher than 50 ppm.

9. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 1, characterized in that, Before outputting the active cut-off command, the active cut-off judgment module in step S4 performs redundancy verification: the feature vector is simultaneously input into an independent backup logic judgment unit, and the output of the backup logic judgment unit is logically ANDed with the judgment result of the active cut-off judgment module. When both are true, the active cut-off command is output.

10. The method for graded early warning and active disconnection control of thermal runaway in a power battery according to claim 1, characterized in that, After the execution unit disconnects the high-voltage main circuit in step S4, a self-locking step is also included: the execution unit automatically locks its disconnected state, prohibits automatic reset, and restores the conduction state of the high-voltage main circuit through manual reset operation.

Images