A battery thermal runaway and safety threshold dynamic adjustment system and method

Abstract

The present invention discloses a battery thermal runaway and safety threshold dynamic adjustment system, comprising: a data acquisition module for collecting battery status parameters, environmental parameters, and vehicle operating conditions; a data preprocessing module for performing preprocessing operations on the collected data; an initial safety threshold setting module for combining a large amount of collected historical sample data and using statistical rules to calculate the initial risk threshold of battery thermal runaway as a judgment standard in the initial deployment phase; a dynamic threshold adjustment module for dynamically and individually adjusting the initial risk threshold based on a Bayesian optimization algorithm; and a thermal runaway prediction model module for combining traditional machine learning with deep learning models to predict the battery risk level for the next 24 hours and the battery temperature for the next hour. A battery thermal runaway and safety threshold dynamic adjustment method is also disclosed. The present invention realizes an early warning mechanism with high accuracy and high recall rate.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy vehicle battery management, and specifically relates to a battery thermal runaway and safety threshold dynamic adjustment system and method. Background Art

[0002] In new energy vehicle applications, power battery safety has always been a key factor affecting vehicle stability and user safety. Battery thermal runaway, in particular, can lead to serious consequences such as combustion and explosion.

[0003] Existing methods usually use fixed threshold settings to monitor thermal runaway, but they cannot cope with the nonlinear change trends of batteries under complex conditions such as aging, usage conditions, and environmental changes. They are prone to false alarms or missed alarms, affecting the reliability of the early warning system.

[0004] Therefore, to address the above issues, a battery thermal runaway and safety threshold dynamic adjustment system and method are provided. Summary of the Invention

[0005] To solve the above-mentioned problems existing in the prior art, the present invention provides a battery thermal runaway and safety threshold dynamic adjustment system and method, which realizes an early warning mechanism with high accuracy and high recall rate.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A battery thermal runaway and safety threshold dynamic adjustment system according to one embodiment of the present invention includes:

[0008] Data acquisition module, used to collect battery status parameters, environmental parameters and vehicle operating conditions;

[0009] Data preprocessing module, used to perform preprocessing operations on the collected data;

[0010] The initial safety threshold setting module is used to combine a large amount of historical sample data and use statistical rules to calculate the initial risk threshold of battery thermal runaway as a judgment standard in the initial deployment stage;

[0011] Dynamic threshold adjustment module, used to dynamically and individually adjust the initial risk threshold based on the Bayesian optimization algorithm;

[0012] Thermal runaway prediction model module, which combines traditional machine learning and deep learning models to predict the battery risk level over the next 24 hours and the battery temperature over the next hour.

[0013] The early warning strategy module is used to perform graded early warning responses based on prediction results and threshold judgments.

[0014] Preferably, the data acquisition module is deployed in the vehicle BMS (battery management system) system or a dedicated acquisition unit. The battery status parameters include single cell / whole pack temperature, single cell / whole pack voltage, charge and discharge current, SOC (state of charge), SOH (battery state of health) and battery internal resistance; environmental parameters include external environment temperature, humidity, air pressure, wind speed, road conditions and traffic congestion conditions; vehicle operating conditions include driving speed, acceleration, driving mode, charge and discharge power and rate of change of power.

[0015] Preferably, in the data preprocessing module, the preprocessing operation includes:

[0016] Outlier detection and denoising, including: temperature change rate exceeding limit, voltage or current drift and SOC mutation detection, among which,

[0017] Temperature change rate exceeds the limit: Calculate the temperature change rate per second and define a threshold. The defined threshold is that the temperature change rate exceeding 2 degrees / second is considered abnormal data;

[0018] Voltage or current drift: When charging or driving, compare the current value with the previous N frames of data. If the difference exceeds a certain threshold, it is considered abnormal data.

[0019] SOC mutation detection: When charging or driving, the SOC changes continuously. If there is a possibility that the continuous frame change exceeds a certain threshold, it will be considered an abnormal situation.

[0020] Preferably, after outlier detection and denoising, data alignment and unified sampling are performed: since the sampling frequencies of the sensors are different, linear interpolation plus spline resampling is used to unify the sampling frequency to one frame every 5 minutes, and a 24-hour historical sliding window is constructed for modeling; wherein,

[0021] The battery sampling frequency is one piece of data per second;

[0022] The sampling frequency of external temperature, humidity, and wind speed is one every 10 minutes;

[0023] The sampling frequency of speed, acceleration, and charge and discharge power is one record per second;

[0024] For one piece of data per second, only the data ending at 5 and 10 minutes are taken. If there is no data at a time point, the data at the closest time point is taken.

[0025] For data that is recorded every 10 minutes, temperature and humidity are interpolated using linear interpolation. The formula is as follows:

[0026] ;

[0027] Where, and Represents two pieces of data 10 minutes apart. and Represent two time points respectively;

[0028] Normalization and standardization: All features are uniformly normalized by Min-Max, where each feature column is calculated independently. and The same feature ensures that the same normalization parameters are used in the training, validation, and inference stages. The calculation formula is as follows:

[0029] ;

[0030] Where, and Represents the minimum and maximum value of the dimension respectively.

[0031] Preferably, in the initial safety threshold setting module, the initial risk threshold includes: a temperature threshold, a voltage threshold and an SOC threshold, wherein the calculation formulas are as follows:

[0032] ;

[0033] ;

[0034] ;

[0035] Where, 、 and They are temperature threshold, voltage threshold and SOC threshold respectively, is the average temperature, is the standard deviation of temperature, is the 99% quantile value, is the 98% quantile value.

[0036] Preferably, in the dynamic threshold adjustment module, dynamically and individually adjusting the initial risk threshold includes:

[0037] Objective function design:

[0038] ;

[0039] Where, is the threshold parameter set, is the false alarm rate, is the underreporting rate, and All are weights;

[0040] Optimization process: Using Gaussian process GP as the proxy model, using the collected historical data of one month as the training set, and using the expected boosting sampling strategy to find the optimal threshold, the parameter space is the voltage difference threshold, the temperature change rate threshold, and the SOC fluctuation threshold. The specific process is as follows:

[0041] Initialize the sampling points and sample the parameter space using random sampling;

[0042] Training GP model, fitting To the objective function;

[0043] Based on the expected improvement sampling strategy, find the next group ;

[0044] Evaluate the actual FP / FN and record the objective function;

[0045] Update GP and repeat the above steps until the maximum number of training rounds reaches 100, completing the optimization.

[0046] Personalized settings: Generate multiple sets of differentiated thresholds based on vehicle ID, battery supplier, battery pack number, and climate zone conditions.

[0047] Preferably, in the thermal runaway prediction model module, the traditional machine learning model is an XGBoost (an integrated machine learning algorithm based on a decision tree) model, which is used for static risk assessment. According to the input temperature mean, maximum / minimum value, 24h voltage change rate, SOC stability, maximum temperature difference, SOC standard deviation, charge and discharge times, current standard deviation, temperature difference mean, temperature difference standard deviation, pressure difference mean, and pressure difference standard deviation, three-level labels of high, medium, and low are obtained, wherein high risk refers to severe fluctuations and overheating, medium risk refers to slight abnormalities or trend risks, and low risk refers to stable and safe operation.

[0048] The deep learning model is an LSTM (recurrent neural network) model, which is used for trend prediction. Based on the input serialized 24-hour data, it obtains the temperature forecast for the next hour. If the future value approaches the current dynamic threshold, a second-level warning is triggered in advance.

[0049] Preferably, in the warning strategy module, the hierarchical warning response includes:

[0050] Level 1 warning: the current value exceeds the threshold;

[0051] Level 2 warning: LSTM predicts a trend breakout.

[0052] Level 3 warning: XGBoost is judged as high risk;

[0053] Combined triggering: When multiple warnings are met at the same time, the expert system is linked to recommend actions, including battery replacement, cooling, and diagnosis.

[0054] A second method of the present invention is a method for dynamically adjusting battery thermal runaway and safety thresholds, comprising:

[0055] Step S1, collecting battery status parameters, environmental parameters and vehicle operating conditions;

[0056] Step S2, performing preprocessing operations on the collected data;

[0057] Step S3: combining a large amount of collected historical sample data and using statistical rules to calculate the initial risk threshold of battery thermal runaway as a judgment standard for the initial deployment stage;

[0058] Step S4, dynamically and individually adjusting the initial risk threshold according to the Bayesian optimization algorithm;

[0059] Step S5: combining traditional machine learning and deep learning models to predict the battery risk level for the next 24 hours and the battery temperature for the next hour;

[0060] Step S6: Perform a graded warning response based on the prediction results and threshold judgment.

[0061] Preferably, in step S1, the battery status parameters include single cell / whole pack temperature, single cell / whole pack voltage, charge and discharge current, SOC, SOH and internal resistance of the battery; the environmental parameters include the temperature, humidity, air pressure, wind speed, road conditions and traffic congestion conditions of the external environment; and the vehicle operating conditions include driving speed, acceleration, driving mode, charge and discharge power and rate of change of charge.

[0062] Preferably, in step S2, the pre-processing operation includes:

[0063] Outlier detection and denoising, including: temperature change rate exceeding limit, voltage or current drift and SOC mutation detection, among which,

[0064] Temperature change rate exceeds the limit: Calculate the temperature change rate per second and define a threshold. The defined threshold is that the temperature change rate exceeding 2 degrees / second is considered abnormal data;

[0065] Voltage or current drift: When charging or driving, compare the current value with the previous N frames of data. If the difference exceeds a certain threshold, it is considered abnormal data.

[0066] SOC mutation detection: When charging or driving, the SOC changes continuously. If there is a possibility that the continuous frame change exceeds a certain threshold, it will be considered an abnormal situation.

[0067] Preferably, after outlier detection and denoising, data alignment and unified sampling are performed: since the sampling frequencies of the sensors are different, linear interpolation plus spline resampling is used to unify the sampling frequency to one frame every 5 minutes, and a 24-hour historical sliding window is constructed for modeling; wherein,

[0068] The battery sampling frequency is one piece of data per second;

[0069] The sampling frequency of external temperature, humidity, and wind speed is one every 10 minutes;

[0070] The sampling frequency of speed, acceleration, and charge and discharge power is one record per second;

[0071] For one piece of data per second, only the data ending at 5 and 10 minutes are taken. If there is no data at a time point, the data at the closest time point is taken.

[0072] For data that is recorded every 10 minutes, temperature and humidity are interpolated using linear interpolation. The formula is as follows:

[0073] ;

[0074] Where, and Represents two pieces of data 10 minutes apart. and Represent two time points respectively;

[0075] Normalization and standardization: All features are uniformly normalized by Min-Max, where each feature column is calculated independently. and The same feature ensures that the same normalization parameters are used in the training, validation, and inference stages. The calculation formula is as follows:

[0076] ;

[0077] Where, and Represents the minimum and maximum value of the dimension respectively.

[0078] Preferably, in step S3, the initial risk threshold includes: a temperature threshold, a voltage threshold, and an SOC threshold, wherein the calculation formulas are as follows:

[0079] ;

[0080] ;

[0081] ;

[0082] Where, 、 and They are temperature threshold, voltage threshold and SOC threshold respectively, is the average temperature, is the standard deviation of temperature, is the 99% quantile value, is the 98% quantile value.

[0083] Preferably, in step S4, dynamically and individually adjusting the initial risk threshold includes:

[0084] Objective function design:

[0085] ;

[0086] Where, is the threshold parameter set, is the false alarm rate, is the underreporting rate, and All are weights;

[0087] Optimization process: Using Gaussian process GP as the proxy model, using the collected historical data of one month as the training set, and using the expected boosting sampling strategy to find the optimal threshold, the parameter space is the voltage difference threshold, the temperature change rate threshold, and the SOC fluctuation threshold. The specific process is as follows:

[0088] Initialize the sampling points and sample the parameter space using random sampling;

[0089] Training GP model, fitting To the objective function;

[0090] Based on the expected improvement sampling strategy, find the next group ;

[0091] Evaluate the actual FP / FN and record the objective function;

[0092] Update GP and repeat the above steps until the maximum number of training rounds reaches 100, completing the optimization.

[0093] Personalized settings: Generate multiple sets of differentiated thresholds based on vehicle ID, battery supplier, battery pack number, and climate zone conditions.

[0094] Preferably, in step S5, the traditional machine learning model is an XGBoost model, which is used for static risk assessment, and obtains high, medium and low level labels according to the input temperature mean, maximum / minimum value, 24h voltage change rate, SOC stability, maximum temperature difference, SOC standard deviation, charge and discharge times, current standard deviation, temperature difference mean, temperature difference standard deviation, pressure difference mean and pressure difference standard deviation, wherein high risk refers to severe fluctuation and overheating, medium risk refers to slight abnormality or trend risk, and low risk refers to stable and safe operation;

[0095] The deep learning model is an LSTM model, which is used for trend prediction. Based on the input serialized 24-hour data, the temperature forecast for the next hour is obtained. If the future value approaches the current dynamic threshold, the second-level warning is triggered in advance.

[0096] Preferably, in step S6, the graded warning response includes:

[0097] Level 1 warning: the current value exceeds the threshold;

[0098] Level 2 warning: LSTM predicts a trend breakout.

[0099] Level 3 warning: XGBoost is judged as high risk;

[0100] Combined triggering: When multiple warnings are met at the same time, the expert system is linked to recommend actions, including battery replacement, cooling, and diagnosis.

[0101] Compared with the existing technology, the beneficial effects of the present invention are: through data collection, feature construction, intelligent prediction, and dynamic optimization, the present invention realizes early judgment and personalized warning of battery thermal runaway risks, improves the safety of the entire vehicle and prevents major battery accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0103] Figure 1 This is a module diagram of a battery thermal runaway and safety threshold dynamic adjustment system of the present invention;

[0104] Figure 2 This is a flow chart of a method for dynamically adjusting battery thermal runaway and safety thresholds according to the present invention. DETAILED DESCRIPTION

[0105] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0106] like Figure 1 As shown, a battery thermal runaway and safety threshold dynamic adjustment system includes: a data acquisition module 1, a data preprocessing module 2, an initial safety threshold setting module 3, a dynamic threshold adjustment module 4, a thermal runaway prediction model module 5 and an early warning strategy module 6.

[0107] The data acquisition module 1 is used to collect battery status parameters, environmental parameters and vehicle operating conditions.

[0108] In the embodiment, the data acquisition module 1 is deployed in the vehicle BMS system or a dedicated acquisition unit. The battery status parameters include the single cell / whole pack temperature, single cell / whole pack voltage, charge and discharge current, SOC, SOH and internal resistance of the battery; the environmental parameters include the temperature, humidity, air pressure, wind speed, road conditions and traffic congestion conditions of the external environment; the vehicle operating conditions include driving speed, acceleration, driving mode, charge and discharge power and rate of change of power.

[0109] The data preprocessing module 2 is used to perform preprocessing operations on the collected data.

[0110] In an embodiment, the pre-processing operation includes:

[0111] Outlier detection and denoising, including: temperature change rate exceeding limit, voltage or current drift and SOC mutation detection, among which,

[0112] Temperature change rate exceeds the limit: Calculate the temperature change rate per second and define a threshold. The defined threshold is that the temperature change rate exceeding 2 degrees / second is considered abnormal data;

[0113] Voltage or current drift: When in charging or driving mode, compare the current value with the previous N frames of data. If the difference exceeds a certain threshold, it is considered abnormal data. The definition of this invention is to compare with the data of the previous 10 frames, and the difference exceeds 15% of the previous 10 frames;

[0114] SOC mutation detection: Whether in charging or driving state, SOC changes continuously. If there is a continuous frame change that exceeds a certain threshold, it will be considered an abnormal situation. The amplitude of the continuous change frame defined by this invention is no more than 2% in charging state and no more than 1% in driving state;

[0115] After outlier detection and denoising, data alignment and unified sampling are performed: Due to the different sampling frequencies of each sensor, linear interpolation plus spline resampling is used to unify the sampling frequency to one frame every 5 minutes, and a 24-hour historical sliding window is constructed for modeling.

[0116] The battery sampling frequency is one piece of data per second;

[0117] The sampling frequency of external temperature, humidity, wind speed and other data is 10 minutes;

[0118] The sampling frequency of speed, acceleration, and charge and discharge power is one record per second;

[0119] For one piece of data per second, only the data ending at 5 and 10 minutes are taken. If there is no data at a time point, the data at the closest time point is taken.

[0120] For data that is recorded every 10 minutes, temperature and humidity are interpolated using linear interpolation. The formula is as follows:

[0121] ;

[0122] Where, and Represents two pieces of data 10 minutes apart. and Represents two time points respectively, and the wind speed and other data are processed in the same way as for the 10 minutes;

[0123] Normalization and standardization: All features are uniformly normalized by Min-Max, where each feature column is calculated independently. and The same feature ensures that the same normalization parameters are used in the training, validation, and inference stages. The calculation formula is as follows:

[0124] ;

[0125] Where, and They represent the minimum and maximum values ​​of the dimension respectively, and all the data of all dimensions are uniformly processed into the range of [0,1].

[0126] The initial safety threshold setting module 3 is used to combine a large amount of collected historical sample data and use statistical rules to calculate the initial risk threshold of battery thermal runaway as a judgment standard in the initial deployment stage.

[0127] In the embodiment, the initial risk threshold includes: a temperature threshold, a voltage threshold, and an SOC threshold, wherein the calculation formulas are as follows:

[0128] ;

[0129] ;

[0130] ;

[0131] Where, 、 and They are temperature threshold, voltage threshold and SOC threshold respectively, is the average temperature, is the standard deviation of temperature, is the 99% quantile value, is the 98% quantile value.

[0132] The dynamic threshold adjustment module 4 is used to dynamically and individually adjust the initial risk threshold according to the Bayesian optimization algorithm.

[0133] In an embodiment, dynamically and individually adjusting the initial risk threshold includes:

[0134] Objective function design:

[0135] ;

[0136] Where, is the threshold parameter set, is the false alarm rate, is the underreporting rate, and All are weights, all equal to 1, but adjustable;

[0137] Optimization process: Using Gaussian process GP as the proxy model, using the collected historical data of one month as the training set, and using the expected boosting sampling strategy to find the optimal threshold, the parameter space is the voltage difference threshold, the temperature change rate threshold, and the SOC fluctuation threshold. The specific process is as follows:

[0138] Initialize the sampling points and sample the parameter space using random sampling;

[0139] Training GP model, fitting To the objective function;

[0140] Based on the expected improvement sampling strategy, find the next group ;

[0141] Evaluate the actual FP / FN and record the objective function;

[0142] Update GP and repeat the above steps until the maximum number of training rounds reaches 100, completing the optimization.

[0143] Personalized settings: Generate multiple sets of differentiated thresholds based on vehicle ID, battery supplier, battery pack number, and climate zone conditions.

[0144] In the embodiment, the biggest reason for this step is that even for the same car, different configurations and models correspond to different batteries, and the dynamic thresholds will be different in different environments, so it is necessary to consider different dimensions and train an independent threshold model for each combination.

[0145] Thermal runaway prediction model module 5 is used to combine traditional machine learning and deep learning models to predict the battery risk level in the next 24 hours and the battery temperature in the next hour.

[0146] In the embodiment, the traditional machine learning model is an XGBoost model, which is used for static risk assessment. According to the input temperature mean, maximum / minimum value, 24h voltage change rate, SOC stability, maximum temperature difference, SOC standard deviation, charge and discharge times, current standard deviation, temperature difference mean, temperature difference standard deviation, pressure difference mean, and pressure difference standard deviation, three-level labels of high, medium, and low are obtained. Among them, high risk refers to severe fluctuations and overheating, medium risk refers to minor abnormalities or trend risks, and low risk refers to stable and safe operation.

[0147] Data in historical records that trigger overheat protection or thermal runaway is defined as high risk; data in historical records that exceeds the threshold defined in step 4 is defined as medium risk, and other data is defined as low risk. This provides labels and input features, and the present invention then uses the XGBoost model for training, and the training process is a conventional machine learning step.

[0148] The deep learning model is an LSTM model, which is used for trend prediction. Based on the input serialized 24-hour data, it obtains the temperature forecast for the next hour. If the future value approaches the current dynamic threshold, a level 2 warning is triggered in advance.

[0149] According to the training model using LSTM, based on the serialized data obtained in the second step, there is one data every five minutes. The label combined with the first 12 data is the temperature of the 24th data, the label combined with the 2nd to 13th data is the temperature of the 25th data, and so on. The characteristics of each combination correspond to the temperature label one hour later. In this way, the characteristic data of a time series can be obtained. Using the LSTM time series model for training, the temperature one hour later can be predicted based on the characteristics of this one hour. If any point in the prediction exceeds the dynamic threshold, a secondary warning will be triggered.

[0150] The early warning strategy module 6 is used to perform graded early warning responses based on the prediction results and threshold judgments.

[0151] In the embodiment, the graded warning response includes:

[0152] Level 1 warning: the current value exceeds the threshold;

[0153] Level 2 warning: LSTM predicts a trend breakout.

[0154] Level 3 warning: XGBoost is judged as high risk;

[0155] Combined triggering: When multiple warnings are met at the same time, the expert system is linked to recommend actions, including battery replacement, cooling, and diagnosis.

[0156] like Figure 2 As shown, a method for dynamically adjusting battery thermal runaway and safety thresholds includes:

[0157] Step S1, collecting battery status parameters, environmental parameters and vehicle operating conditions.

[0158] In the embodiment, the battery status parameters include the single cell / whole pack temperature, single cell / whole pack voltage, charge and discharge current, SOC, SOH and internal resistance of the battery; the environmental parameters include the temperature, humidity, air pressure, wind speed, road conditions and traffic congestion conditions of the external environment; the vehicle operating conditions include driving speed, acceleration, driving mode, charge and discharge power and charge change rate.

[0159] Step S2: pre-processing the collected data.

[0160] In an embodiment, the pre-processing operation includes:

[0161] Outlier detection and denoising, including: temperature change rate exceeding limit, voltage or current drift and SOC mutation detection, among which,

[0162] Temperature change rate exceeds the limit: Calculate the temperature change rate per second and define a threshold. The defined threshold is that the temperature change rate exceeding 2 degrees / second is considered abnormal data;

[0163] Voltage or current drift: When charging or driving, compare the current value with the previous N frames of data. If the difference exceeds a certain threshold, it is considered abnormal data.

[0164] SOC mutation detection: When charging or driving, the SOC changes continuously. If there is a possibility that the continuous frame change exceeds a certain threshold, it will be considered an abnormal situation;

[0165] After outlier detection and denoising, data alignment and unified sampling are performed: Due to the different sampling frequencies of each sensor, linear interpolation plus spline resampling is used to unify the sampling frequency to one frame every 5 minutes, and a 24-hour historical sliding window is constructed for modeling.

[0166] The battery sampling frequency is one piece of data per second;

[0167] The sampling frequency of external temperature, humidity, wind speed and other data is 10 minutes;

[0168] The sampling frequency of speed, acceleration, and charge and discharge power is one record per second;

[0169] For one piece of data per second, only the data ending at 5 and 10 minutes are taken. If there is no data at a time point, the data at the closest time point is taken.

[0170] For data that is recorded every 10 minutes, temperature and humidity are interpolated using linear interpolation. The formula is as follows:

[0171] ;

[0172] Where, and Represents two pieces of data 10 minutes apart. and Represent two time points respectively;

[0173] Normalization and standardization: All features are uniformly normalized by Min-Max, where each feature column is calculated independently. and The same feature ensures that the same normalization parameters are used in the training, validation, and inference stages. The calculation formula is as follows:

[0174] ;

[0175] Where, and Represents the minimum and maximum value of the dimension respectively.

[0176] In step S3, a large amount of historical sample data is collected and statistical rules are used to calculate the initial risk threshold of battery thermal runaway as a judgment standard in the initial deployment stage.

[0177] In the embodiment, the initial risk threshold includes: a temperature threshold, a voltage threshold, and an SOC threshold, wherein the calculation formulas are as follows:

[0178] ;

[0179] ;

[0180] ;

[0181] Where, 、 and They are temperature threshold, voltage threshold and SOC threshold respectively, is the average temperature, is the standard deviation of temperature, is the 99% quantile value, is the 98% quantile value.

[0182] Step S4: Dynamically and individually adjust the initial risk threshold according to the Bayesian optimization algorithm.

[0183] In an embodiment, dynamically and individually adjusting the initial risk threshold includes:

[0184] Objective function design:

[0185] ;

[0186] Where, is the threshold parameter set, is the false alarm rate, is the underreporting rate, and All are weights;

[0187] Optimization process: Using Gaussian process GP as the proxy model, using the collected historical data of one month as the training set, and using the expected boosting sampling strategy to find the optimal threshold, the parameter space is the voltage difference threshold, the temperature change rate threshold, and the SOC fluctuation threshold. The specific process is as follows:

[0188] Initialize the sampling points and sample the parameter space using random sampling;

[0189] Training GP model, fitting To the objective function;

[0190] Based on the expected improvement sampling strategy, find the next group ;

[0191] Evaluate the actual FP / FN and record the objective function;

[0192] Update GP and repeat the above steps until the maximum number of training rounds reaches 100, completing the optimization.

[0193] Personalized settings: Generate multiple sets of differentiated thresholds based on vehicle ID, battery supplier, battery pack number, and climate zone conditions.

[0194] Step S5: Combine traditional machine learning and deep learning models to predict the battery risk level in the next 24 hours and the battery temperature in the next hour.

[0195] In the embodiment, the traditional machine learning model is an XGBoost model, which is used for static risk assessment. According to the input temperature mean, maximum / minimum value, 24h voltage change rate, SOC stability, maximum temperature difference, SOC standard deviation, charge and discharge times, current standard deviation, temperature difference mean, temperature difference standard deviation, pressure difference mean, and pressure difference standard deviation, three-level labels of high, medium, and low are obtained. Among them, high risk refers to severe fluctuations and overheating, medium risk refers to minor abnormalities or trend risks, and low risk refers to stable and safe operation.

[0196] The deep learning model is an LSTM model, which is used for trend prediction. Based on the input serialized 24-hour data, the temperature forecast for the next hour is obtained. If the future value approaches the current dynamic threshold, the second-level warning is triggered in advance.

[0197] Step S6: Perform a graded warning response based on the prediction results and threshold judgment.

[0198] In the embodiment, the graded warning response includes:

[0199] Level 1 warning: the current value exceeds the threshold;

[0200] Level 2 warning: LSTM predicts a trend breakout.

[0201] Level 3 warning: XGBoost is judged as high risk;

[0202] Combined triggering: When multiple warnings are met at the same time, the expert system is linked to recommend actions, including battery replacement, cooling, and diagnosis.

[0203] The following will illustrate the technical details of the present invention through several embodiments.

[0204] Example 1: Overall structure of battery thermal runaway prediction system

[0205] This embodiment proposes a device-cloud collaborative battery thermal runaway risk prediction system, which mainly consists of the following modules:

[0206] (1) Data acquisition module (end side)

[0207] Installed in BMS or independent data acquisition unit.

[0208] Real-time collection of battery data (such as temperature, voltage, current, SOC), vehicle status (speed, power) and environmental information (external temperature, humidity).

[0209] Data is uploaded to the cloud using the MQTT / HTTP protocol with a configurable period (default 1s).

[0210] (2) Data preprocessing module (cloud)

[0211] Clean the raw data, including:

[0212] Abnormal detection: such as temperature jump (ΔT>30°C / s), voltage abnormality (V<2.5V or V>4.2V);

[0213] Filtering and denoising: using sliding average and Kalman filtering;

[0214] Time alignment: linear interpolation of data of different frequencies;

[0215] Window processing: extract historical data within 24 hours as input features.

[0216] (3) Initial safety threshold setting module

[0217] Use the statistical characteristics of historical data to calculate the initial threshold for thermal runaway determination:

[0218] The initial risk thresholds include: temperature threshold, voltage threshold, and SOC threshold, where the calculation formulas are as follows:

[0219] ;

[0220] ;

[0221] ;

[0222] (4) Dynamic threshold adjustment module

[0223] Based on Bayesian optimization algorithm:

[0224] Construct false positive rate / missing negative rate as the objective function;

[0225] Use Gaussian processes to model the current data distribution;

[0226] Retrain the optimal threshold every 30 days;

[0227] Supports personalized settings (fine-tuning for vehicle models, battery pack numbers, etc.).

[0228] (5) Thermal runaway prediction model module

[0229] XGBoost:

[0230] The training data is statistical features (maximum, minimum, mean, slope, etc.) within a 24-hour window;

[0231] Output: high, medium, low risk level.

[0232] LSTM:

[0233] Input: serialized historical temperature, voltage, and SOC data;

[0234] Output: temperature or voltage prediction for the next hour;

[0235] If the predicted value approaches the threshold → early warning.

[0236] (6) Early warning strategy module

[0237] The triggering rules are as follows:

[0238] The current value exceeds the dynamic threshold → Level 1 warning;

[0239] LSTM prediction trend exceeds the threshold → Level 2 warning;

[0240] XGBoost judges high risk → High-level risk warning;

[0241] Information can be fed back through the vehicle computer / APP / BMS panel; linked to the expert system to recommend maintenance measures (such as coolant inspection, battery pack replacement, etc.).

[0242] Example 2: Threshold Dynamic Adjustment Process (Bayesian Optimization)

[0243] 1. Define the objective function design:

[0244] ;

[0245] Where, is the threshold parameter set, is the false alarm rate, is the underreporting rate, and All are weights;

[0246] 2. Collect real alarm data model prediction results;

[0247] 3. Count FP and FN every day and fit the Gaussian process;

[0248] 4. Use the EI (Expected Improvement) strategy to select the next threshold to be evaluated;

[0249] 5. After multiple rounds of iteration, the optimal threshold is determined after convergence.

[0250] Example 3: Model training process

[0251] 1. Data collection: 12 months of data from 3,000 vehicles;

[0252] 2. Training set division: 90% of the data is used for training and 10% for testing;

[0253] 3. LSTM configuration: Input: past 24 hours of data; Number of layers: 2; Hidden layer dimension: 128; Time steps: 288 (every 5 minutes, for a total of 24 hours);

[0254] 4. XGBoost configuration: maximum depth: 6; learning rate: 0.1; number of trees: 200; input features: mean, variance, maximum rate of change, etc.;

[0255] 5. Integrated logic:

[0256] If LSTM predicts that future data approaches the threshold and XGBoost determines it to be high risk, the highest level warning is triggered.

[0257] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A battery thermal runaway and safety threshold dynamic adjustment system, characterized in that: include: Data acquisition module, used to collect battery status parameters, environmental parameters and vehicle operating conditions; Data preprocessing module, used to perform preprocessing operations on the collected data; The initial safety threshold setting module is used to combine a large amount of historical sample data and use statistical rules to calculate the initial risk threshold of battery thermal runaway as a judgment standard in the initial deployment stage; Dynamic threshold adjustment module, used to dynamically and individually adjust the initial risk threshold based on the Bayesian optimization algorithm; Thermal runaway prediction model module, which combines traditional machine learning and deep learning models to predict the battery risk level over the next 24 hours and the battery temperature over the next hour. The early warning strategy module is used to perform graded early warning responses based on prediction results and threshold judgments; In the dynamic threshold adjustment module, the dynamic personalized adjustment of the initial risk threshold includes: Objective function design: ； Where θ is the threshold parameter set, FP is the false alarm rate, FN is the missed alarm rate, ω1 and ω2 are weights; Optimization process: Using Gaussian process GP as the proxy model, using the collected historical data of one month as the training set, and using the expected boosting sampling strategy to find the optimal threshold, the parameter space is the voltage difference threshold, the temperature change rate threshold, and the SOC fluctuation threshold. The specific process is as follows: Initialize the sampling points and sample the parameter space using random sampling; Train the GP model and fit θ to the target function; Based on the expected improvement sampling strategy, find the next group of θ; Evaluate the actual FP / FN and record the objective function; Update GP and repeat the above steps until the maximum number of training rounds reaches 100, completing the optimization. Personalized settings: Generate multiple sets of differentiated thresholds based on vehicle ID, battery supplier, battery pack number, and climate zone conditions.

2. A battery thermal runaway and safety threshold dynamic adjustment system according to claim 1, characterized in that: The data acquisition module is deployed in the vehicle's BMS system or a dedicated acquisition unit. The battery status parameters include single cell / whole pack temperature, single cell / whole pack voltage, charge and discharge current, SOC, SOH and battery internal resistance; environmental parameters include external environment temperature, humidity, air pressure, wind speed, road conditions and traffic congestion conditions; vehicle operating conditions include driving speed, acceleration, driving mode, charge and discharge power and charge change rate.

3. The battery thermal runaway and safety threshold dynamic adjustment system according to claim 1, characterized in that: In the data preprocessing module, the preprocessing operation includes: Outlier detection and denoising, including: temperature change rate exceeding limit, voltage or current drift and SOC mutation detection, among which, Temperature change rate exceeds the limit: Calculate the temperature change rate per second and define a threshold. The defined threshold is that the temperature change rate exceeding 2 degrees / second is considered abnormal data; Voltage or current drift: When charging or driving, compare the current value with the previous N frames of data. If the difference exceeds a certain threshold, it is considered abnormal data. SOC mutation detection: When charging or driving, the SOC changes continuously. If there is a possibility that the continuous frame change exceeds a certain threshold, it will be considered an abnormal situation.

4. A battery thermal runaway and safety threshold dynamic adjustment system according to claim 3, characterized in that: After outlier detection and denoising, data alignment and unified sampling are performed: Due to the different sampling frequencies of each sensor, linear interpolation plus spline resampling is used to unify the sampling frequency to one frame every 5 minutes, and a 24-hour historical sliding window is constructed for modeling. The battery sampling frequency is one piece of data per second; The sampling frequency of external temperature, humidity, and wind speed is one every 10 minutes; The sampling frequency of speed, acceleration, and charge and discharge power is one record per second; For one piece of data per second, only the data ending at 5 and 10 minutes are taken. If there is no data at a time point, the data at the closest time point is taken. For data that is recorded every 10 minutes, temperature and humidity are interpolated using linear interpolation. The formula is as follows: ； Where x0 and x1 represent two pieces of data 10 minutes apart, and t0 and t1 represent two time points. Normalization and standardization: All features are uniformly normalized by Min-Max, where each feature column is calculated independently, x min and x max The same feature ensures that the same normalization parameters are used in the training, validation, and inference stages. The calculation formula is as follows: ； Where x min and x max Represents the minimum and maximum value of the dimension respectively.

5. The battery thermal runaway and safety threshold dynamic adjustment system according to claim 1, characterized in that: In the initial safety threshold setting module, the initial risk threshold includes: temperature threshold, voltage threshold and SOC threshold, wherein the calculation formulas are as follows: ； ； ； Where, T threshold 、V threshold and SOC threshold are temperature threshold, voltage threshold and SOC threshold respectively, μ T is the average temperature, σ T is the standard deviation of temperature, P 99 (V) is the 99% quantile value, P 98 (SOC) is the 98th percentile value.

6. The battery thermal runaway and safety threshold dynamic adjustment system according to claim 1, characterized in that: In the thermal runaway prediction model module, the traditional machine learning model is the XGBoost model, which is used for static risk assessment. According to the input temperature mean, maximum / minimum value, 24h voltage change rate, SOC stability, maximum temperature difference, SOC standard deviation, charge and discharge times, current standard deviation, temperature difference mean, temperature difference standard deviation, pressure difference mean, and pressure difference standard deviation, three-level labels of high, medium, and low are obtained. Among them, high risk refers to severe fluctuations and overheating, medium risk refers to minor abnormalities or trend risks, and low risk refers to stable and safe operation. The deep learning model is an LSTM model, which is used for trend prediction. Based on the input serialized 24-hour data, a temperature forecast for the next hour is obtained. If the future value approaches the current dynamic threshold, a secondary warning is triggered in advance.

7. The battery thermal runaway and safety threshold dynamic adjustment system according to claim 1, characterized in that: In the warning strategy module, the hierarchical warning response includes: Level 1 warning: the current value exceeds the threshold; Level 2 warning: LSTM predicts a trend breakout. Level 3 warning: XGBoost is judged as high risk; Combined triggering: When multiple warnings are met at the same time, the expert system is linked to recommend actions, including battery replacement, cooling, and diagnosis.

8. A method for dynamic adjustment of battery thermal runaway and safety threshold, characterized in that: include: Step S1, collecting battery status parameters, environmental parameters and vehicle operating conditions; Step S2, performing preprocessing operations on the collected data; Step S3: combining a large amount of collected historical sample data and using statistical rules to calculate the initial risk threshold of battery thermal runaway as a judgment standard for the initial deployment stage; Step S4, dynamically and individually adjusting the initial risk threshold according to the Bayesian optimization algorithm; Step S5: combining traditional machine learning and deep learning models to predict the battery risk level for the next 24 hours and the battery temperature for the next hour; Step S6: Perform a graded warning response based on the prediction results and threshold judgment; In step S4, dynamically and individually adjusting the initial risk threshold includes: Objective function design: ； Where θ is the threshold parameter set, FP is the false alarm rate, FN is the missed alarm rate, ω1 and ω2 are weights; Optimization process: Using Gaussian process GP as the proxy model, using the collected historical data of one month as the training set, and using the expected boosting sampling strategy to find the optimal threshold, the parameter space is the voltage difference threshold, the temperature change rate threshold, and the SOC fluctuation threshold. The specific process is as follows: Initialize the sampling points and sample the parameter space using random sampling; Train the GP model and fit θ to the target function; Based on the expected improvement sampling strategy, find the next group of θ; Evaluate the actual FP / FN and record the objective function; Update GP and repeat the above steps until the maximum number of training rounds reaches 100, completing the optimization. Personalized settings: Generate multiple sets of differentiated thresholds based on vehicle ID, battery supplier, battery pack number, and climate zone conditions.

9. The method for dynamic adjustment of battery thermal runaway and safety threshold according to claim 8, characterized in that: In step S1, the battery status parameters include the cell / whole pack temperature, cell / whole pack voltage, charge and discharge current, SOC, SOH and internal resistance of the battery; the environmental parameters include the external environment temperature, humidity, air pressure, wind speed, road conditions and traffic congestion conditions; and the vehicle operating conditions include driving speed, acceleration, driving mode, charge and discharge power and charge change rate.

10. The method for dynamic adjustment of battery thermal runaway and safety threshold according to claim 8, characterized in that: In step S2, the pre-processing operation includes: Outlier detection and denoising, including: temperature change rate exceeding limit, voltage or current drift and SOC mutation detection, among which, Temperature change rate exceeds the limit: Calculate the temperature change rate per second and define a threshold. The defined threshold is that the temperature change rate exceeding 2 degrees / second is considered abnormal data; Voltage or current drift: When charging or driving, compare the current value with the previous N frames of data. If the difference exceeds a certain threshold, it is considered abnormal data. SOC mutation detection: When charging or driving, the SOC changes continuously. If there is a possibility that the continuous frame change exceeds a certain threshold, it will be considered an abnormal situation.

11. A battery thermal runaway and safety threshold dynamic adjustment method according to claim 10, characterized in that: In step S2, after outlier detection and denoising, data alignment and unified sampling are performed: due to the different sampling frequencies of the sensors, linear interpolation plus spline resampling is used to unify the sampling frequency to one frame every 5 minutes, and a 24-hour historical sliding window is constructed for modeling; wherein, The battery sampling frequency is one piece of data per second; The sampling frequency of external temperature, humidity, and wind speed is one every 10 minutes; The sampling frequency of speed, acceleration, and charge and discharge power is one record per second; For one piece of data per second, only the data ending at 5 and 10 minutes are taken. If there is no data at a time point, the data at the closest time point is taken. For data that is recorded every 10 minutes, temperature and humidity are interpolated using linear interpolation. The formula is as follows: ； Where x0 and x1 represent two pieces of data 10 minutes apart, and t0 and t1 represent two time points. Normalization and standardization: All features are uniformly normalized by Min-Max, where each feature column is calculated independently, x min and x max The same feature ensures that the same normalization parameters are used in the training, validation, and inference stages. The calculation formula is as follows: ； Where x min and x max Represents the minimum and maximum value of the dimension respectively.

12. The method for dynamic adjustment of battery thermal runaway and safety threshold according to claim 8, characterized in that: In step S3, the initial risk thresholds include: a temperature threshold, a voltage threshold, and an SOC threshold, wherein the calculation formulas are as follows: ； ； ； Where, T threshold 、V threshold and SOC threshold are temperature threshold, voltage threshold and SOC threshold respectively, μ T is the average temperature, σ T is the standard deviation of temperature, P 99 (V) is the 99% quantile value, P 98 (SOC) is the 98th percentile value.

13. The method for dynamic adjustment of battery thermal runaway and safety threshold according to claim 8, characterized in that: In step S5, the traditional machine learning model is an XGBoost model, which is used for static risk assessment. According to the input temperature mean, maximum / minimum value, 24h voltage change rate, SOC stability, maximum temperature difference, SOC standard deviation, charge and discharge times, current standard deviation, temperature difference mean, temperature difference standard deviation, pressure difference mean, and pressure difference standard deviation, three-level labels of high, medium, and low are obtained. Among them, high risk refers to severe fluctuations and overheating, medium risk refers to minor abnormalities or trend risks, and low risk refers to stable and safe operation. The deep learning model is an LSTM model, which is used for trend prediction. Based on the input serialized 24-hour data, a temperature forecast for the next hour is obtained. If the future value approaches the current dynamic threshold, a secondary warning is triggered in advance.

14. The method for dynamic adjustment of battery thermal runaway and safety threshold according to claim 8, characterized in that: In step S6, the graded warning response includes: Level 1 warning: the current value exceeds the threshold; Level 2 warning: LSTM predicts a trend breakout. Level 3 warning: XGBoost is judged as high risk; Combined triggering: When multiple warnings are met at the same time, the expert system is linked to recommend actions, including battery replacement, cooling, and diagnosis.

Images