A big data-based electric vehicle battery failure early warning method

Abstract

The application relates to the technical field of electric vehicle battery management, and discloses an electric vehicle battery fault early warning method based on big data, which comprises the following steps: step 1, data acquisition and preprocessing, the running parameters of an electric vehicle battery are acquired through a battery management system, including a state of charge, a state of health, a voltage, a current and a temperature, the acquired data are cleaned, noise is removed through median filtering, and all parameters are aligned through a time synchronization method; step 2, a battery state space model is constructed, the state of charge, the voltage and the state of health of the battery are used as system state variables, and a state space equation is constructed. Through Bayesian optimization, a fault early warning threshold is dynamically adjusted, the self-adaptive optimization can be realized according to different working conditions, the alarm triggering condition can be automatically corrected under different battery aging states and use environments, the false alarm rate and the missed alarm rate are reduced, and the reliability of the fault early warning system is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric vehicle battery management, in particular to an electric vehicle battery fault early warning method based on big data. BACKGROUND

[0002] With the rapid development of the electric vehicle industry, the safety and reliability of electric vehicle batteries have attracted increasing attention. Electric vehicle battery failures can lead to reduced vehicle performance, reduced range, and even cause battery thermal runaway and safety accidents. Therefore, establishing an accurate, real-time, and adaptable battery fault early warning system is of great significance to improving the intelligence level of the battery management system and the safety and service life of electric vehicles.

[0003] Currently, battery fault early warning methods mainly include fixed threshold-based fault determination methods, physical model-based battery health evaluation methods, and data-driven machine learning methods. The fixed threshold method is simple and easy to implement, but due to battery aging and changes in the use environment, the fixed threshold cannot adapt to different working conditions, and false positives and false negatives are likely to occur. The physical model method estimates the battery state of health and state of charge based on battery equivalent circuit models or electrochemical models, but the model parameters may decrease in accuracy due to changes in use time. The data-driven method learns from historical data to predict faults, but traditional methods are mostly based on offline training and cannot adapt to changes in new data. In addition, they require a large amount of data and cannot achieve good results under small sample conditions.

[0004] To address the above problems, the present application proposes an electric vehicle battery fault early warning method based on big data, which dynamically adjusts the fault early warning threshold, uses a long short-term memory network for time series modeling, and combines incremental learning to improve the adaptability of the model, thereby improving the accuracy and robustness of fault prediction and adapting to different battery types, working conditions, and aging states.

[0005] Traditional fault early warning methods rely on fixed threshold settings, which can lead to false positives and false negatives when the battery state of health decreases and the environmental conditions change. The present application proposes a dynamic fault early warning threshold adjustment method based on Bayesian optimization, which dynamically optimizes the alarm trigger threshold by real-time evaluation of the fault probability and historical operating data, so that the early warning system can adapt to different working conditions, reduce the likelihood of false positives and false negatives, and improve the reliability of fault early warning.

[0006] Existing data-driven methods mostly rely on static feature extraction and cannot effectively model the time evolution of battery state. The present application uses a long short-term memory network for time series data modeling, which can learn the trend of battery operating state changes over time, capture the long-term dependence relationship of the battery aging process, and improve the prediction accuracy of battery faults. Compared with traditional machine learning methods, long short-term memory can adapt to complex battery operating environments and improve the accuracy of fault early warning.

[0007] The existing machine learning method usually relies on large-scale historical data for offline training, and cannot maintain high-precision prediction when the battery type, environmental conditions and use conditions change. The present application proposes a fault early warning model based on incremental learning, which dynamically adjusts the model parameters when new data arrives through online gradient update, so that the system can continuously learn new fault patterns, improve the adaptability to different battery types and working conditions, and reduce the retraining cost of the model.

[0008] In summary, the present application constructs an electric vehicle battery fault early warning method based on big data through dynamic fault early warning threshold adjustment, long short-term memory time series modeling and incremental learning optimization, providing a new technical solution for improving the intelligent level of the battery management system and the safety of the electric vehicle. SUMMARY

[0009] In view of the deficiencies of the prior art, the present application provides an electric vehicle battery fault early warning method based on big data to solve the problems raised in the background art.

[0010] To achieve the above purpose, the present application is implemented by the following technical scheme: an electric vehicle battery fault early warning method based on big data, comprising:

[0011] Step 1, data acquisition and preprocessing, obtaining the operating parameters of the electric vehicle battery through the battery management system, including the state of charge, the state of health, the voltage, the current and the temperature, and cleaning the obtained data, removing noise by median filtering, and aligning all parameters by using time synchronization method;

[0012] Step 2, constructing a battery state space model, using the state of charge, voltage and state of health of the battery as system state variables, constructing state space equations, the state equation describes the relationship between the state of charge and time, the health state equation describes the degradation trend of the battery health state, and the voltage equation describes the relationship between the battery terminal voltage and current and internal resistance, and combining the observation equation to determine the relationship between the measured data and the system state, providing a mathematical model for state estimation;

[0013] Step 3, using state estimation algorithm, based on the constructed state space model, using Kalman filter algorithm to estimate the state of charge and health state online, first using the prediction equation to predict the state at the next time, then combining the measurement data at the current time to update, to correct the state variable;

[0014] Step 4, establish an incremental learning model, based on the state of charge, state of health and voltage variables obtained by the state estimation algorithm, build an incremental learning model to model the battery failure risk, the incremental learning model uses long short-term memory neural network to process time series data, and when new data comes, the network parameters are optimized by online gradient update to adapt to different battery operating environments and different use stage failure mode changes;

[0015] Step 5, calculate the failure probability, use the incremental learning model to predict the battery state at future time, calculate the probability of battery failure within a set time window, the calculation process combines historical operation data, and considers the change trend of state of charge, state of health, current and temperature;

[0016] Step 6, set dynamic warning threshold, based on the calculated failure probability, use Bayesian optimization method to dynamically adjust the failure warning threshold, initialize the threshold and calculate the false alarm rate first, then determine the optimal threshold through iterative optimization, so that the triggering condition of failure warning adapts to the change of battery operating environment, and provides the optimal reference for the final warning decision;

[0017] Step 7, execute failure warning decision, according to the calculated failure probability and the set dynamic warning threshold, judge whether the battery enters the failure state, if the failure probability exceeds the dynamic threshold, trigger the failure alarm signal, and transmit the failure information to the battery management system to prompt the user to perform corresponding maintenance and replacement operation.

[0018] Preferably, in the data collection and preprocessing, the calculation of state of charge uses ampere hour integral method for correction, and the calculation formula is as follows:

[0019] ,

[0020] Among them, is the state of charge at time , is the state of charge at time , is the sampling time interval, is the rated capacity of the battery, is the battery current at time ,

[0021] is the error of state of charge calculation, with mean value 0 and variance of normal distribution.

[0022] Preferably, in the construction of battery state space model, the calculation of battery health state uses exponential decay model for description, and the calculation formula is as follows:

[0023] ,

[0024] wherein, is the battery state of health at time is the battery health degradation rate, is the charge-discharge impact coefficient, is the battery state of charge at time is the depth of discharge at time

[0025] ,

[0026] wherein, is the discharge amount at time is the maximum available capacity of the battery,

[0027] is the state of charge calculation error, with mean value 0 and variance of normal distribution.

[0028] Preferably, in the state estimation algorithm, the state transition matrix of Kalman filter adopts the following form:

[0029] ,

[0030] wherein, is the state transition matrix, is the time degradation coefficient of the state of health, is the sampling time interval, 0 represents the normal state of the battery, and 1 represents the fault state of the battery.

[0031] Preferably, in the process of establishing the incremental learning model, the incremental learning model adopts a long short-term memory neural network for time sequence feature extraction, and the hidden state update formula of the long short-term memory neural network unit is as follows:

[0032] ,

[0033] wherein, is the long short-term memory neural network hidden state vector at time

[0034] is the output gate vector of the long short-term memory neural network;

[0035] is the memory state of the long short-term memory neural network unit;

[0036] the hyperbolic tangent nonlinear transformation is performed on the memory unit state of the long short-term memory neural network the output gate is represented​​​​ element-wise weighting.

[0037] Preferably, the input of the incremental learning model when calculating the failure probability is the state variable at the time step, the input feature matrix is defined as follows:

[0038] ,

[0039] wherein, is the historical state data matrix within the time step .

[0040] is the battery terminal voltage sequence; is the current sequence;

[0041] is the temperature sequence; is the time window length;

[0042] represents the remaining capacity of the battery; represents the health degree of the battery.

[0043] Preferably, when setting the dynamic early warning threshold, the dynamic threshold of the failure early warning is solved by using the Bayesian optimization method, and the optimization objective function is defined as follows:

[0044] ,

[0045] wherein, is the dynamic failure early warning threshold at the time .

[0046] is the number of samples evaluated in the optimization process;

[0047] is the weight parameter; is the failure probability predicted by the model;

[0048] is an indicator function, which takes 1 if the predicted value exceeds the threshold, otherwise takes 0;

[0049] represents the optimal threshold found by optimization .

[0050] Preferably, when executing the failure early warning decision, the failure probability is calculated as follows:

[0051] ,

[0052] wherein, sigmoid activation function; model weight vector; state variable at the current time; bias term; failure probability.

[0053] Preferably, the dynamic threshold Adaptive adjustment is adopted using historical data, and the calculation formula is as follows:

[0054]

[0055] wherein, initial threshold value, adjustment coefficient, length of the historical sample window,

[0056] adjustment weight, historical failure indicator function.

[0057] Preferably, the electric vehicle battery failure warning method is applicable to different types of electric vehicle batteries, including but not limited to lithium iron phosphate batteries and ternary lithium batteries.

[0058] The present application provides an electric vehicle battery failure warning method based on big data. It has the following advantages:

[0059] 1. The present application dynamically adjusts the failure warning threshold value through Bayesian optimization, can adaptively optimize according to different working conditions, realizes automatic correction of alarm triggering conditions under different battery aging states and use environments, reduces the false positive rate and the false negative rate, and improves the reliability of the failure warning system.

[0060] 2. The present application processes battery historical operation data through a long short-term memory network, realizes long-term dependence modeling of the battery health state, improves the battery failure trend prediction accuracy, obtains accurate battery failure warning, and reduces the safety risk brought by sudden failure.

[0061] 3. The present application dynamically updates model parameters through an incremental learning strategy, realizes adaptive optimization under different battery models, use environments and workloads, obtains the effect of adapting to different battery characteristics without retraining, and improves the universality and generalization ability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 is a flowchart of the present application. DETAILED DESCRIPTION

[0063] ​In the following, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are part of, rather than all of the embodiments of the present application. Based on the embodiments in the present application, other embodiments obtained by those of ordinary skill in the art without creative effort should belong to the scope of the present application.

[0064] The present application will be described in detail in the following with reference to the accompanying drawings: Embodiments

[0065] Please refer to the accompanying Figure 1 The embodiments of the present application provide a big data-based electric vehicle battery fault early warning method, comprising:

[0066] Step 1, data acquisition and preprocessing, obtaining the operating parameters of the electric vehicle battery through the battery management system, including the state of charge, the state of health, the voltage, the current and the temperature, and cleaning the obtained data, removing noise by using median filtering, and aligning all parameters by using time synchronization method;

[0067] Step 2, constructing a battery state space model, using the state of charge, the voltage and the state of health of the battery as system state variables, constructing a state space equation, the state equation describes the relationship between the state of charge and time, the health state equation describes the degradation trend of the battery health state, and the voltage equation describes the relationship between the battery terminal voltage and the current and internal resistance, and combining the observation equation to determine the relationship between the measurement data and the system state, providing a mathematical model for state estimation;

[0068] Step 3, using state estimation algorithm, based on the constructed state space model, using Kalman filter algorithm to estimate the state of charge and the state of health online, first using the prediction equation to predict the state at the next time, then combining the measurement data at the current time to update, to correct the state variable;

[0069] Step 4, establishing an incremental learning model, based on the state of charge, the state of health and the voltage variables obtained by the state estimation algorithm, constructing an incremental learning model, modeling the battery fault risk, the incremental learning model uses long short-term memory neural network to process time series data, and optimizes the network parameters by online gradient update when new data comes, to adapt to the fault mode changes of different battery operating environments and different use stages;

[0070] Step 5, calculating the fault probability, using the incremental learning model to predict the battery state at the future time, calculating the probability of battery failure within a set time window, the calculation process combines historical operation data, considering the change trend of the state of charge, the state of health, the current and the temperature;

[0071] Step 6, set dynamic early warning threshold, based on the calculated failure probability, using Bayesian optimization method to dynamically adjust the failure warning threshold, first initialize the threshold and calculate the false alarm rate, and then determine the optimal threshold through iterative optimization, so that the triggering condition of failure warning adapts to the changes of battery operating environment, providing optimal reference for the final warning decision;

[0072] Step 7, execute failure warning decision, according to the calculated failure probability and the set dynamic warning threshold, judge whether the battery enters the failure state, if the failure probability exceeds the dynamic threshold, trigger the failure alarm signal, and transmit the failure information to the battery management system to prompt the user to perform corresponding maintenance and replacement operation.

[0073] The benefits of data collection and preprocessing provide high-quality, complete, and usable data for subsequent modeling, ensuring that the failure warning system can analyze based on real and reliable battery status.

[0074] The benefits of building a battery state space model compared to traditional experience-based judgment methods are that state space modeling can associate battery operating conditions with measured data, enhancing the interpretability of the model and enabling accurate state estimation.

[0075] The benefits of using state estimation algorithms are that through the prediction-update cycle, the battery state variables can be continuously corrected as new data arrives, enabling the battery state estimation to adapt to real-time operating conditions and improving the accuracy of the failure warning system. Compared to static models, this method can dynamically respond to changes in battery state, improving the reliability of state of charge and state of health estimation.

[0076] The benefits of establishing an incremental learning model compared to traditional static training machine learning methods are that it can adaptively adjust the prediction ability without retraining the entire model, improving the long-term adaptability and generalization ability of the failure warning system.

[0077] The benefits of failure probability calculation are that it can use the incremental learning model to predict the battery state at future time, ensuring the accuracy and forward-looking nature of failure probability calculation, and providing a scientific and reliable basis for subsequent failure warning.

[0078] The benefits of setting dynamic early warning threshold are that compared to fixed threshold method, it can reduce false alarm rate and false negative rate, making the battery warning system adapt to complex operating conditions, improving the sensitivity and accuracy of failure detection.

[0079] The benefits of executing failure warning decision are that compared to traditional manual monitoring and fixed time maintenance, it realizes intelligent and automated battery failure warning, reduces the need for human intervention, and improves the safety and service life of the battery.

[0080] In data collection and preprocessing, the state of charge is calculated using the ampere-hour integral method, and the calculation formula is as follows:

[0081] ,

[0082] wherein, is the state of charge at time , is the state of charge at time , is the sampling time interval, is the rated capacity of the battery, is the battery current at time ,

[0083] is the state of charge calculation error, normally distributed with mean 0 and variance .

[0084] In constructing the battery state space model, the calculation of the battery health state is described by using an exponential decay model, and the calculation formula is as follows:

[0085] ,

[0086] wherein, is the battery health state at time , is the battery health decay rate, is the charge and discharge influence coefficient, is the discharge depth at time , defined as:

[0087] ,

[0088] wherein, is the discharge amount at time , is the maximum available capacity of the battery,

[0089] is the state of charge calculation error, normally distributed with mean 0 and variance .

[0090] Compared with the traditional ampere-hour integral method, the error correction term is introduced in the present application, which reduces the influence of current measurement error on the state of charge estimation and avoids the long-term accumulation of error leading to the deviation of state of charge calculation. Combined with the state estimation algorithm, the state of charge calculation value is dynamically corrected to approach the true value, thereby improving the reliability of the battery remaining capacity estimation.

[0091] The exponential decay model accurately reflects the change trend of the battery health state with the increase of the number of charge and discharge cycles, and avoids the low precision problem of the traditional linear model. The discharge capacity and the maximum available capacity of the battery are considered comprehensively, so that the battery health state estimation can adapt to different charge and discharge strategies, and the credibility of the prediction is improved. The error term is used to correct the battery health state estimation, reduce the influence of measurement noise, and make the battery health state estimation stable and reliable.

[0092] In summary, in the electric vehicle battery fault warning method, the improved state of charge calculation formula and the exponential decay battery health state model effectively improve the accuracy and reliability of the battery state estimation:

[0093] The ampere-hour integral method is used to correct the error, reduce the influence of measurement error on the state of charge estimation, and improve the accuracy of the battery remaining capacity.

[0094] The exponential decay model and the discharge depth influence factor are used to truly depict the change trend of the battery health state, and the long-term prediction accuracy is improved.

[0095] When the state estimation algorithm is used, the state transition matrix of Kalman filter adopts the following form:

[0096] ,

[0097] Among them, is the state transition matrix, is the time decay coefficient of the health state, is the sampling time interval, 0 represents the normal state of the battery, and 1 represents the fault state of the battery.

[0098] The state transition matrix is used to establish a mathematical model of the change of the battery health state with time, so that the battery health state can be accurately estimated. By introducing the time decay coefficient λ of the health state, the aging process of the battery with time can be effectively described, and the low precision problem caused by simple linear interpolation is avoided.

[0099] The traditional method usually calculates the battery health state based on fixed historical data, and cannot adapt to the state change of the battery in long-term use. Through the prediction-update cycle of Kalman filter, this method can dynamically update the battery health state estimation value when new data arrives, so that the health state estimation is flexible and accurate.

[0100] The 0 / 1 state variable is used to distinguish whether the battery is in a fault state, so that the system can identify whether the battery enters a fault mode while the health state changes.

[0101] The traditional battery state of health estimation method is susceptible to measurement noise, resulting in large fluctuations in the calculation results. By using Kalman filtering, the measurement data can be smoothed using the system prediction model, effectively reducing the interference of noise on the battery state of health estimation, and making the results stable.

[0102] In the process of establishing the incremental learning model, the incremental learning model uses a long short-term memory neural network for time series feature extraction. The hidden state update formula of the long short-term memory neural network unit is as follows:

[0103] ,

[0104] Wherein, is the long short-term memory neural network hidden state vector at time .

[0105] is the output gate vector of the long short-term memory neural network;

[0106] is the memory state of the long short-term memory neural network unit;

[0107] The hyperbolic tangent nonlinear transformation is performed on the long short-term memory neural network memory unit state . The output gate is element-wise weighted.

[0108] Enhance the time series modeling capability, so that the system can fully learn the long-term trend of the battery state.

[0109] Avoid the problem of gradient disappearance, improve the modeling capability of long-term dependence, and ensure the prediction accuracy of the battery state of health.

[0110] Improve the nonlinear feature expression capability, so that the model can learn the complex relationship between battery state variables.

[0111] Optimize the information screening mechanism, reduce the interference of irrelevant data, and improve the reliability of fault prediction.

[0112] Support incremental learning, so that the model can dynamically update when new data arrives, adapt to different battery operating environments and aging patterns, and improve the generalization ability of the system.

[0113] Through the improvement, the battery fault early warning method of the application can accurately and real-timely predict the battery health condition, improve the intelligent level of the electric vehicle battery management system, reduce the safety hidden trouble caused by sudden failure, and improve the reliability and service life of the electric vehicle.

[0114] When calculating the fault probability, the input of the incremental learning model is​ The state variable of the time step, the input feature matrix is defined as follows:

[0115] ,

[0116] Wherein, is the historical state data matrix within the time step ;

[0117] is the battery terminal voltage sequence; is the current sequence;

[0118] is the temperature sequence; is the time window length;

[0119] represents the remaining capacity of the battery; represents the health degree of the battery.

[0120] The traditional fault prediction method usually relies on the data at the current time, and cannot capture the long-term evolution law of the fault, which is easy to cause misjudgment. The present application uses the historical state data within the time step, can identify the change trend of the battery state, and improves the accuracy of fault prediction.

[0121] Because the battery behaves differently in different working environments, the fixed rule fault detection method cannot adapt to different working conditions.

[0122] The present method inputs the data of multiple historical time points, so that the model can learn the battery characteristics under different working conditions, improve the adaptability, and reduce false positives and false negatives.

[0123] The traditional static machine learning method usually uses single time point state data, and cannot model the dynamic change trend of the battery.

[0124] The present application uses long short-term memory combined with time series data input, so that the model can learn the long-term change mode of the battery state, improve the prediction accuracy, and discover potential faults in advance.

[0125] By inputting continuous data within the time step, the present method can constantly update the model parameters, so that the system can adapt to new battery operation modes over time.

[0126] When executing the fault warning decision, the fault probability The calculation formula is as follows:

[0127] ,

[0128] Wherein, is a sigmoid activation function; is a model weight vector; is the state variable at the current time; is the bias term; is the failure probability.

[0129] The state transition matrix of Kalman filtering has significant advantages for battery state of health estimation:

[0130] By introducing the state transition matrix and time degradation coefficient , Kalman filtering can accurately capture the evolution trend of battery state of health, avoiding the low precision problem of traditional models.

[0131] Kalman filtering can dynamically adjust the battery state of health estimation according to real-time data, ensuring that the battery health can be continuously and accurately estimated by the use time and environmental changes.

[0132] By reducing noise interference, the timeliness and accuracy of battery failure detection are improved, providing stable and reliable data support for electric vehicle battery management systems.

[0133] Therefore, Kalman filtering optimizes battery state estimation and improves the performance of battery failure warning, with wide application potential, especially in the field of battery management of electric vehicles and energy storage systems.

[0134] When setting the dynamic warning threshold, the dynamic threshold of failure warning is solved by the Bayesian optimization method, and the optimization objective function is defined as follows:

[0135] ,

[0136] where is the dynamic failure warning threshold at time ;

[0137] is the number of samples evaluated in the optimization process;

[0138] is the weight parameter; is the failure probability predicted by the model;

[0139] is the indicator function, which takes 1 if the predicted value exceeds the threshold, otherwise it takes 0;

[0140] denotes the optimal threshold found by optimization .

[0141] The dynamic threshold is adaptively adjusted using historical data, and the calculation formula is as follows:

[0142] ,

[0143] wherein, is an initial threshold value, is an adjustment coefficient, is a historical sample window length,

[0144] is an adjustment weight, is a historical failure indication function.

[0145] The present application adopts the Bayesian optimization method to solve the dynamic fault early warning threshold, and successfully solves the limitations of traditional methods through adaptive adjustment of historical data:

[0146] Improve the accuracy of fault early warning, reduce false positives and false negatives, and make the system reliable.

[0147] Adapt to the change of battery health status, so that the system can monitor the battery status for a long time and is not affected by battery aging.

[0148] Optimize the early warning sensitivity, reduce false positives, and avoid false alarms caused by short-term fluctuations of the battery.

[0149] Enhance the generalization ability of the system, which can adapt to different types of batteries, use environments and working conditions.

[0150] Reduce manual intervention and improve intelligence. The threshold is automatically adjusted through historical data and Bayesian optimization to avoid the problem of hard-coded threshold.

[0151] The final effect is that the battery fault early warning system of the electric vehicle is intelligent and accurate, and can adapt to different use environments, improving the intelligent level of the battery management system and providing important protection for the safe operation of the electric vehicle.

[0152] The electric vehicle battery fault early warning method is applicable to different types of electric vehicle batteries, including but not limited to lithium iron phosphate batteries and ternary lithium batteries.

[0153] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for early warning of electric vehicle battery failure based on big data, characterized in that: include: Step 1: Data acquisition and preprocessing: The battery management system obtains the operating parameters of the electric vehicle battery, including state of charge, health status, voltage, current and temperature. The acquired data is cleaned, median filtering is used to remove noise, and all parameters are aligned using time synchronization methods. Step 2: Construct a battery state space model. Use the battery's state of charge, voltage, and health status as system state variables to construct a state space equation. The state equation describes the relationship between the state of charge and time, the health status equation describes the decline trend of the battery's health status, and the voltage equation describes the relationship between the battery terminal voltage, current, and internal resistance. Combined with the observation equation, the connection between the measurement data and the system state is determined to provide a mathematical model for state estimation. Step 3: Using the state estimation algorithm, based on the constructed state space model, the Kalman filter algorithm is used to perform online estimation of the state of charge and health state. The prediction equation is first used to predict the state at the next moment, and then the state is updated in combination with the measurement data at the current moment to correct the state variables. Step 4: Establish an incremental learning model. Based on the state of charge, health status, and voltage variables obtained by the state estimation algorithm, an incremental learning model is constructed to model the battery failure risk. The incremental learning model uses a long short-term memory neural network to process time series data and optimizes network parameters through online gradient updates when new data arrives to adapt to different battery operating environments and changes in failure modes at different usage stages. Step 5: Calculate the failure probability. Use the incremental learning model to predict the battery status at future times and calculate the probability of battery failure within a set time window. The calculation process incorporates historical operating data and considers the changing trends of state of charge, health status, current, and temperature. Step 6: Set a dynamic warning threshold. Based on the calculated fault probability, dynamically adjust the fault warning threshold using the Bayesian optimization method. Initialize the threshold and calculate the false alarm rate. Then, determine the optimal threshold through iterative optimization. This allows the triggering conditions for fault warnings to adapt to changes in the battery operating environment, providing the best reference for the final warning decision. Step 7: Execute the fault warning decision. Based on the calculated fault probability and the set dynamic warning threshold, determine whether the battery has entered a fault state. If the fault probability exceeds the dynamic threshold, a fault alarm signal is triggered and the fault information is transmitted to the battery management system to prompt the user to perform corresponding maintenance and replacement operations.

2. The electric vehicle battery failure early warning method based on big data according to claim 1 is characterized in that: In the data collection and preprocessing, the state of charge is calculated using the ampere-hour integration method for correction, and the calculation formula is as follows: ， in, For the moment The state of charge, For the moment The state of charge, is the sampling time interval, is the rated capacity of the battery, For the moment The battery current, is the charge state calculation error, with a mean of 0 and a variance of Normal distribution.

3. The electric vehicle battery failure early warning method based on big data according to claim 2 is characterized in that: In the battery state space model, the battery health state is calculated using an exponential decay model, and the calculation formula is as follows: ， in, For the moment Battery health status, is the battery health degradation rate, is the charge and discharge influence coefficient, For the moment The depth of discharge is defined as: ， in, For the moment Discharge amount, is the maximum available capacity of the battery, is the charge state calculation error, with a mean of 0 and a variance of Normal distribution.

4. The electric vehicle battery failure early warning method based on big data according to claim 1 is characterized in that: When the state estimation algorithm is used, the state transfer matrix of the Kalman filter adopts the following form: ， in, is the state transition matrix, is the time decay coefficient of the health state, is the sampling time interval, 0 represents the normal state of the battery, and 1 represents the fault state of the battery.

5. The electric vehicle battery failure early warning method based on big data according to claim 1 is characterized in that: In the process of establishing the incremental learning model, the incremental learning model uses a long short-term memory neural network to extract temporal features. The hidden state update formula of the long short-term memory neural network unit is as follows: ， in, For the moment The hidden state vector of the long short-term memory neural network; is the output gate vector of the long short-term memory neural network; is the memory state of the long short-term memory neural network unit; Memory unit state of long short-term memory neural network Perform hyperbolic tangent nonlinear transformation; Represents the output gate right Element-wise weighting of .

6. The electric vehicle battery failure early warning method based on big data according to claim 1 is characterized in that: When calculating the failure probability, the input of the incremental learning model is The state variables of the time step, the input feature matrix is ​​defined as follows: ， in, is the time step The historical state data matrix within; is the battery terminal voltage sequence; is the current sequence; is the temperature series; is the time window length; Indicates the remaining battery power; Indicates the health of the battery.

7. The electric vehicle battery failure early warning method based on big data according to claim 1 is characterized in that: When setting the dynamic warning threshold, the dynamic threshold of the fault warning is solved using the Bayesian optimization method, and the optimization objective function is defined as follows: ， in, For the moment Dynamic fault warning threshold; is the number of samples evaluated during the optimization process; is the weight parameter; The failure probability predicted by the model; is an indicator function, which takes 1 if the predicted value exceeds the threshold, otherwise it takes 0; Indicates finding the optimal threshold through optimization .

8. The electric vehicle battery failure early warning method based on big data according to claim 1 is characterized in that: When executing the fault warning decision, the fault probability The calculation formula is as follows: ， in, is the sigmoid activation function; is the model weight vector; is the state variable at the current moment; is the bias term; is the failure probability.

9. The electric vehicle battery failure early warning method based on big data according to claim 7, characterized in that: The dynamic threshold Adaptive adjustment is made using historical data. The calculation formula is as follows: ， in, is the initial threshold, is the adjustment coefficient, is the historical sample window length, To adjust the weights, is the historical fault indication function.

10. The electric vehicle battery failure early warning method based on big data according to claim 1, characterized in that: The electric vehicle battery failure early warning method is applicable to different types of electric vehicle batteries, including but not limited to lithium iron phosphate batteries and ternary lithium batteries.

Images