A battery safety risk grade early warning and evaluation method

Abstract

The application discloses a battery safety risk grade early warning and evaluation method, relates to the technical field of lithium ion batteries, and comprises the following steps: collecting battery operation data, processing to obtain capacity attenuation characteristics and internal resistance growth characteristics, utilizing an Arrhenius model to obtain temperature-related degradation characteristic parameters, estimating capacity loss trends under different temperatures and times, inputting multi-source time series data into a long short-term memory network to predict capacity and internal resistance changes in the future for a plurality of cycles, combining life termination criteria, calculating remaining service life, and fusing degradation indexes of an aging model and degradation indexes of a capacity model to generate a safety state index, and outputting battery safety risk classification and corresponding early warning information. The application can provide overall safety risk evaluation for the whole battery system, accurately monitor and early warn single batteries or local modules, and improve the safety and reliability of battery operation.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery technology, specifically, to a method for early warning and assessment of battery safety risk levels. Background Technology

[0002] Driven by the global energy transition and the electrification of transportation, lithium-ion batteries have become the core support for electric vehicles and distributed energy storage systems. However, under complex operating conditions, batteries experience capacity decay and internal resistance growth over long periods. Influenced by multiple factors such as temperature, rate capability, charge / discharge strategies, and consistency differences, the degradation process exhibits distinct phased and nonlinear characteristics, and may rapidly deteriorate after an inflection point, inducing safety accidents such as thermal runaway, swelling, and internal short circuits. Therefore, how to provide interpretable, quantifiable, and forward-looking early warnings of potential risks is a key scientific and engineering problem for ensuring the safety and reliability of battery systems. Current engineering practices largely rely on battery management systems to monitor and alarm conventional parameters such as voltage, current, temperature, and SOC / SOH using threshold-based methods. Some solutions identify equivalent circuit parameters through experiments such as hybrid pulse power testing to assess the changing trends of polarization and internal resistance. There are also system-level safety assessment methods based on statistical or empirical thresholds, comprehensively judging macroscopic quantities such as ambient temperature and historical cycle data. While the aforementioned methods can reflect operational status to some extent, they have three limitations: First, the indicators are singular and highly a posteriori, relying heavily on threshold triggering, making it difficult to capture risk evolution before the degradation inflection point in a timely manner. Second, their extrapolation capabilities across temperature and operating conditions are insufficient; the accelerating effect of temperature on the decay rate is difficult to model uniformly, resulting in poor alarm stability and consistency across different scenarios. Third, their granularity and interpretability are limited; system-level assessments cannot drill down to the cell / module level, and the black-box data-driven method lacks mechanistic support, hindering operational decision-making and root cause analysis. Furthermore, practical applications also face challenges such as sparse data sampling, noise interference, uneven distribution of operating conditions, and high dispersion among cells. Judging solely by static thresholds or single-modal features often struggles to balance accuracy, timeliness, and generalization: setting thresholds too broadly may miss early risks, while setting them too strictly can lead to false alarms and unnecessary maintenance; the parameter snapshots obtained from a single static test are insufficient to characterize long-term time-series dependencies; and the lack of quantitative characterization of temperature acceleration mechanisms and phased degradation also limits the reliable prediction of key indicators such as remaining service life and safety status. Summary of the Invention

[0003] To address the aforementioned technical problems, the present invention aims to provide a method for early warning and assessment of battery safety risk levels, which balances the interpretability of mechanisms with the ability to learn time series: on the one hand, it is based on temperature-degradation kinetics to quantitatively characterize the temperature acceleration effect and stage inflection points; on the other hand, it integrates multi-source operating data to capture the nonlinear evolution and long-term dependence of capacity and internal resistance over time. While meeting engineering deployment constraints, it achieves graded, forward-looking, and executable safety early warning and assessment from the cell to the system level, providing a reliable basis for operation and maintenance strategies.

[0004] The present invention solves the above problems through the following technical solution:

[0005] A method for early warning and assessment of battery safety risk levels, comprising:

[0006] Step S1: Collect battery operating data and process it to obtain capacity decay characteristics and internal resistance growth characteristics. The battery operating data includes voltage U, current I, temperature T, capacity Q, and mixed pulse power test data.

[0007] Step S2: Obtain the capacity-cycle count curve based on the capacity decay characteristics. For the capacity-cycle count curve The analysis yielded the knee point and knee point value. Based on different temperature conditions, the knee point value Input the Arrhenius model to obtain temperature-related degradation characteristic parameters;

[0008] Step S3: Calculate the activation energy Ea and the pre-factor A of the battery capacity decay process according to the Arrhenius formula, and then estimate the capacity loss trend under different temperatures and times.

[0009] Step S2 starts from the knee point and extracts temperature-related degradation features. The Ea and A obtained in step S2 are Arrhenius fitting based on "knee point lifetime", which yields the degradation rate parameters corresponding to the knee point. It emphasizes the sensitivity of the battery aging curve inflection point to temperature changes and is used to identify the accelerated characteristics of degradation stages at different temperatures.

[0010] Step S3 involves using formula inversion to obtain kinetic parameters and establish a predictive model based on the complete capacity decay process. The Ea and A obtained in Step S3 are Arrhenius analyses of the entire capacity loss process, yielding rate parameters for the entire decay kinetic process. The aim is to predict the capacity decay trend under different temperature-time combinations. Step S3, to some extent, supplements and expands upon Step S2, but the data sources and fitting focuses differ between the two.

[0011] The "Arrhenius model" is a model that describes the change of reaction rate with temperature from a mechanistic perspective. The "Arrhenius equation" is the specific mathematical expression of this model.

[0012] Step S4: Using multi-source time-series data and Arrhenius feature parameters as input, construct a long short-term memory network to predict the capacity and internal resistance changes in the next few cycles; the multi-source time-series data includes capacity, internal resistance, temperature, state of charge and number of cycles;

[0013] In order for the Long Short-Term Memory Network to capture both inflection point features and learn the whole life cycle trend, it is usually necessary to use Ea and A (knee features) obtained in step S2 and Ea and A (whole process features) obtained in step S3 as feature inputs. They characterize the temperature acceleration effect of battery degradation from different perspectives, and together they can make the model more comprehensive.

[0014] Step S5: Based on the predicted capacity and internal resistance trajectory, combined with the lifespan termination criterion:

[0015] Q EOL ≤0.8Qn or R EOL ≥1.33Rn;

[0016] Among them, Q EOL R is the capacity at the end of its lifespan. EOL Rn is the internal resistance at the end of its lifespan; Qn is the nominal capacity, and Rn is the nominal internal resistance; when any criterion is met, the battery is determined to have entered the end-of-life state, and the remaining lifespan is calculated.

[0017] Step S6: Combine the remaining service life with the degradation index of the aging model and the degradation index of the capacity model to generate a safety status index, and output the battery safety risk classification and corresponding warning information according to the safety status level.

[0018] This invention uses the capacity decay characteristics, internal resistance growth characteristics, and temperature influence factor of lithium batteries as key parameters. It obtains the battery degradation mechanism characteristics through knee point identification and Arrhenius model calculation, and then combines this with a long short-term memory network to learn and predict multi-source time-series data, generating future capacity and internal resistance change trends for the battery. Based on the prediction results and a set lifespan termination threshold, the remaining lifespan is calculated and further mapped to a safety state index, thereby achieving battery risk classification and early warning.

[0019] Furthermore, the capacity Q includes the initial rated capacity of the battery. Measured capacity at time t Based on the initial rated capacity of the battery Measured capacity at time t Calculate capacity decay and capacity decay rate :

[0020] ;

[0021] ;

[0022] Capacity decay and capacity decay rate This constitutes the capacity decay characteristic.

[0023] Furthermore, the method for calculating the internal resistance growth characteristic is as follows:

[0024] Calculate the ohmic internal resistance using mixed pulse power test data. :

[0025] ;

[0026] in, This represents the instantaneous change in voltage at the lower terminal caused by the pulsed current excitation. The change is due to the current; by comparing the ohmic internal resistance at different times... The numerical values ​​are arranged according to the time series and the number of cycles to obtain the characteristics of internal resistance growth.

[0027] Furthermore, the capacity-cycle count curve... The analysis yielded the knee point and knee point value. The method is as follows: For the capacity-cycle count curve... Find the derivative, locate the inflection point of accelerated capacity decay, and determine the knee point and its value from the inflection point. .

[0028] Furthermore, the knee point value is determined by combining different temperature conditions. The method for obtaining temperature-related degradation feature parameters by inputting the Arrhenius model is as follows:

[0029] Temperature in different temperature-time combinations ,in, , Numbering at different temperature moments; extracting the associated temperature knee values Input Arrhenius model:

[0030] ;

[0031] in, For temperature-dependent reaction rate constants, It is the natural logarithm. Let R be the temperature and R be the gas constant. The rate of aging to the knee is defined as... Therefore, let ,get:

[0032] ;

[0033] By fitting the aging rate to the knee point at different temperatures, temperature-dependent degradation characteristic parameters, i.e., prefactors, are obtained. With activation energy .

[0034] Further, step S3 specifically includes:

[0035] Step S31: Obtain battery capacity loss under different temperature-time combinations through experiments or historical data. ,in, The temperature in the temperature-time combination; For time in the temperature-time combination;

[0036] Step S32: Based on Arrhenius thermodynamics and the time-cumulative characteristics of capacity loss, construct a model: temperature in the same temperature-time combination. Below, battery capacity loss Time in temperature-time combination It exhibits a power-law relationship, combined with the Arrhenius formula. ,get:

[0037] ;

[0038] Where k represents the decay rate and R is the gas constant; for nth power; Power;

[0039] Temperature in the same temperature-time combination Multiple groups below Take the natural logarithm:

[0040] ;

[0041] by x-axis The slope of the line fitted to the ordinate is the power. After eliminating the power-law effect of time accumulation, Normalization After normalization, its equation is:

[0042] ;

[0043] Taking the natural logarithm of the equation, we get:

[0044] ;

[0045] by x-axis By fitting a straight line to the ordinate, the activation energy can be obtained. With the previous factor ;

[0046] Step S33: Combine the temperatures in any temperature-time combination. Time in temperature-time combination Substituting the extended Arrhenius relation:

[0047] ;

[0048] in, This is the predicted value of capacity loss. The power is an adjustable factor, which yields the capacity loss trend at different temperatures T and times t. , Predicted value of capacity loss of; Time in a certain temperature-time combination The nth power.

[0049] Furthermore, step S4 specifically includes:

[0050] Constructing multi-source feature input vectors :

[0051] ;

[0052] in, This represents the current remaining capacity at time t. Let be the ohmic internal resistance at time t. Polarization resistor, Polarized capacitor, It reflects the polarization effect of the battery under rapid dynamic processes, such as electrochemical double layer charging and discharging, and interfacial charge transfer processes; It reflects the polarization effect of the battery in slow processes, such as concentration polarization and ion diffusion processes; It characterizes the rapid charge storage capacity at the electrode / electrolyte interface, corresponding to double-layer capacitance or charge transfer processes. Its variation is mainly affected by temperature and electrolyte interface properties; It characterizes the slow charge storage capacity within the electrode or under a concentration gradient, corresponding to the energy buffer of ion diffusion / concentration polarization. Its variation is mainly affected by the electrode material structure and ion diffusion resistance. Let t be the battery temperature. Let t be the state of charge. Let t be the number of charge-discharge cycles at time t. Let t be the capacity loss. This represents the number of iterations at the knee point corresponding to time t, used to help capture the decay inflection point;

[0053] The constructed multi-source feature input vector is fed into a long short-term memory network, and its gating units are used to achieve selective memorization and updating of temporal information.

[0054] Input gate : via the sigmoid function The formula for controlling the input weights of new information is:

[0055] ;

[0056] in, The input weight matrix of the input gate is used to input the multi-source feature vector at the current time step. The weight matrix mapped to the input gate; The hidden state weight matrix for the input gate is used to capture the hidden state at the previous time step. Mapped to the input gate; This is the bias term for the input gate, used to compensate for the activation offset of the input gate;

[0057] Forgotten Gate : ;

[0058] Forgotten Gate Decide on the hidden state at the previous moment Information that needs to be retained, including: The input weight matrix for the forget gate is the multi-source feature input vector at the current time step. Mapped to the forget gate dimension, it is used to learn the regulatory weights of input features on the degree of retention of historical information; The hidden state weight matrix for the forget gate is used to capture the hidden state of the previous time step. Forgetting decision characteristics with time-dependent factors; This is the bias term for the forget gate, used to compensate for the mean shift between the input features and the hidden state, and to optimize the activation baseline of the forget gate.

[0059] Candidate Memory :pass The formula for generating potentially updated memory content is as follows:

[0060] ;

[0061] in, The input weight matrix for candidate memories is used to learn the current multi-source feature input vector. Contribution to the generation of new memories; Given the hidden state weight matrix of the candidate memory, we can mine the previous hidden state. New memory adaptation features in mid-temporal association; The bias term for candidate memories is adjusted to modify the activation benchmark of candidate memories and adapt to the nonlinear characteristics of battery aging.

[0062] Memory cell update: Combining the input gate and the forgetting gate, the state of memory cells is updated as follows:

[0063] ;

[0064] in, The cumulative battery aging information stored in the memory cell at time t-1 is ⊙, where ⊙ represents element-wise product.

[0065] Output gate : Controlling the state of memory cells to a hidden state The formula for its output is:

[0066] ;

[0067] in, The input weight matrix of the output gate is used to learn the current multi-source feature input vector. Regulation of memory output decisions (such as the filtering of decay information output by the SOC state). The hidden state weight matrix of the output gate captures the previous hidden state. Intermediate time-dependent output decision inertia The output gate bias term compensates for the activation offset of the output gate and optimizes the output accuracy in the hidden state;

[0068] The final output formula is:

[0069] ;

[0070] Based on the aforementioned gating mechanism, the Long Short-Term Memory (LSTM) network learns the dependencies of battery states in long-sequence inputs, enabling multi-step rolling predictions within a time window H for future prediction, and outputting trajectory predictions of capacity and internal resistance. It covers the short-term and medium-term evolution trends of battery aging process. The capacity is the predicted value within the time window length H of future forecasting; The predicted value of the internal resistance within the time window length H of future prediction.

[0071] Furthermore, it also includes employing a multi-objective loss function. Training the Long Short-Term Memory Network:

[0072] ;

[0073] in, This is the capacity forecast value. This is the predicted internal resistance value. This is the actual capacity value. Let θ be the true value of internal resistance, α be the weight of the capacity prediction task, β be the weight of the internal resistance prediction task, Θ be the parameters of the long short-term memory network model, λ be the coefficient of the L2 regularization term, and H be the length of the time window for future prediction.

[0074] Furthermore, the method for calculating the remaining useful life in step S5 is as follows:

[0075] Based on the trajectory prediction value of the output capacity and internal resistance Iterate through the data to find the first time that satisfies the capacity threshold condition. and the first moment when the internal resistance threshold condition is met. The smaller of these two times is taken as the actual end time of the battery's lifespan. Remaining service life (RUL):

[0076] RUL= ;in, This refers to the current moment.

[0077] Furthermore, it also includes defining a normalized capacity path termination lifetime index. With internal resistance path termination lifetime index ,have:

[0078] Capacity Path End-of-Life Index : ;

[0079] in, Nominal capacity The measured capacity at time t The physical meaning is the proportion of the current capacity decay to the total capacity decay at the end of its life;

[0080] Internal resistance path termination lifetime index : ;

[0081] in, As for the current internal resistance, The initial internal resistance, Used to quantify the proportion of current internal resistance growth to the total internal resistance growth at the end of life. That is, the degradation index of the aging model. That is, the degradation index of the capacity model;

[0082] right and The value is truncated to limit its range to [0-1] to prevent the exponent from going out of bounds due to abnormal data.

[0083] Further, step S6 specifically includes:

[0084] Construct a convergence risk score S, which integrates remaining lifetime risk, capacity aging risk, and internal resistance aging risk;

[0085] ;

[0086] in Used to eliminate negative risks. For reference lifespan, Reflects the deviation of remaining service life from reference service life; the closer the remaining service life is to the reference service life, the better. The higher this value, the higher the risk of remaining life expectancy; This is the capacity path termination lifetime index. As a weight for capacity aging risk, The larger the value, the closer the capacity decay is to the end-of-life threshold, and the higher the risk. This is the internal resistance path termination lifetime index. As the internal resistance aging risk weight; The larger the value, the closer the internal resistance increases to the end-of-life threshold, and the higher the risk. For task weights, the following must be satisfied: This is used to balance the impact of different risk factors;

[0087] Security Status Index Mapping: The fused risk score S is linearly mapped to the [0-8] interval to generate an intuitive and easily identifiable security status level. : ;

[0088] in Used to limit the range of values ​​for S, ensuring that the safe state after mapping is an integer between 0 and 8; This is a rounding function to make the safety status classification more in line with engineering decision-making habits.

[0089] Based on the safety status values, a tiered handling strategy is defined as "Normal—Attention—Power Limitation—Recommended Replacement—Forced Replacement / Isolation":

[0090] Safety status = 0~2: The battery is in normal safety status and can be continuously monitored.

[0091] Safety Status = 3~4: Entering the attention level, it is necessary to increase the frequency of inspections and focus on investigating the causes of abnormalities;

[0092] Safety status = 5~6: Trigger power limiting suggestion, reduce battery charging and discharging power to avoid accelerated aging;

[0093] Safety status = 7~8: Perform replacement / isolation operations, replace faulty batteries in a timely manner, and ensure the safe operation of the system.

[0094] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0095] (1) This invention can not only provide an overall safety risk assessment for the entire battery system, but also accurately monitor and provide early warning for individual cells or local modules, effectively preventing local failures from spreading to the whole system and improving the safety and reliability of battery operation.

[0096] (2) This invention provides a safety risk assessment method that balances mechanistic interpretability and temporal learning capability: on the one hand, it uses temperature-degradation kinetics to quantitatively characterize the temperature acceleration effect and stage inflection points; on the other hand, it integrates multi-source operating data to capture the nonlinear evolution and long-term dependence of capacity and internal resistance over time. While meeting engineering deployment constraints, it achieves hierarchical, forward-looking, and executable safety early warning and assessment from the cell to the system level, providing a reliable basis for operation and maintenance strategies. This invention effectively makes up for the shortcomings of existing technologies in early identification, cross-scenario consistency, and decision interpretability, and improves the intrinsic safety level and operational reliability of power batteries throughout their entire life cycle.

[0097] (3) This invention combines the Arrhenius mechanism model with the Long Short-Term Memory Network (LSTM) time-series prediction model to jointly model and predict the capacity decay and internal resistance growth of power lithium batteries under multiple temperature and operating conditions, generating a battery safety assessment coefficient based on both physical mechanism and time-series learning. This safety assessment coefficient enables accurate calculation of the remaining battery life and maps the result to a safety state index, thus providing graded early warning for the battery. Through rolling prediction and early warning, local degradation problems can be effectively prevented from evolving into systemic safety hazards, improving the overall operational reliability.

[0098] (4) This invention combines the mechanism quantification of "knee point + Arrhenius" with the temporal learning of "long short-term memory network multi-step prediction" to achieve forward-looking safety warning and life assessment of the entire life cycle of power lithium batteries. The method can realize online monitoring and power limiting strategy linkage in vehicle battery management systems, and can also perform group health management and operation and maintenance scheduling in the battery management system / cloud platform of large-scale energy storage power stations. It has a clear implementation path and good engineering adaptability. Attached Figure Description

[0099] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0100] The present invention will be further described in detail below with reference to embodiments, but the implementation of the present invention is not limited thereto.

[0101] Example 1:

[0102] Combination Figure 1 As shown, a method for early warning and assessment of battery safety risk levels includes:

[0103] Step S1: Collect battery operating data, including voltage U, current I, temperature T, capacity Q, and hybrid pulse power HPPC test data; obtain battery capacity decay characteristics, internal resistance growth characteristics, and other time-series characteristics through data processing;

[0104] Step S2: Obtain the capacity-cycle count curve based on the capacity decay characteristics. For the capacity-cycle count curve Perform analysis to identify the knee point and extract its value. Knee point value An Arrhenius model was constructed based on different temperature conditions, and the knee value was... Input the Arrhenius model to obtain temperature-related degradation characteristic parameters;

[0105] Step S3: Calculate the activation energy Ea and pre-factor A of the battery capacity decay process according to the Arrhenius formula, obtain the mechanistic index of capacity decay, and predict the capacity loss trend at a given temperature and time.

[0106] Step S2 starts from the knee point and extracts temperature-related degradation features. The Ea and A obtained in step S2 are Arrhenius fitting based on "knee point lifetime", which yields the degradation rate parameters corresponding to the knee point. It emphasizes the sensitivity of the battery aging curve inflection point to temperature changes and is used to identify the accelerated characteristics of degradation stages at different temperatures.

[0107] Step S3 involves using formula inversion to obtain kinetic parameters and establish a predictive model based on the complete capacity decay process. The Ea and A obtained in Step S3 are Arrhenius analyses of the entire capacity loss process, yielding rate parameters for the entire decay kinetic process. The aim is to predict the capacity decay trend under different temperature-time combinations. Step S3, to some extent, supplements and expands upon Step S2, but the data sources and fitting focuses differ between the two.

[0108] The "Arrhenius model" is a model that describes the change of reaction rate with temperature from a mechanistic perspective. The "Arrhenius equation" is the specific mathematical expression of this model.

[0109] Step S4: Using the Arrhenius mechanism features and historical time-series data (capacity, internal resistance, temperature, etc.) as input, construct and train a Long Short-Term Memory (LSTM) network to achieve multi-step prediction of future capacity and internal resistance changes. That is, using multi-source time-series data and Arrhenius feature parameters as input, construct a LSTM network to predict capacity and internal resistance changes in the next few cycles. The multi-source time-series data includes capacity, internal resistance, temperature, state of charge, and number of cycles.

[0110] In order for the Long Short-Term Memory Network to capture both inflection point features and learn the whole life cycle trend, it is usually necessary to use Ea and A (knee features) obtained in step S2 and Ea and A (whole process features) obtained in step S3 as feature inputs. They characterize the temperature acceleration effect of battery degradation from different perspectives, and together they can make the model more comprehensive.

[0111] Step S5: Based on the predicted capacity and internal resistance trajectories, and combined with the lifespan termination criterion (capacity drops to 0.8Qn or internal resistance increases to 1.33Rn), calculate the remaining lifespan of the battery. Specifically:

[0112] Based on the predicted capacity and internal resistance trajectories, combined with the end-of-life criteria:

[0113] Q EOL ≤0.8Qn or R EOL ≥1.33Rn;

[0114] Among them, Q EOL R is the capacity at the end of its lifespan. EOL Rn is the internal resistance at the end of its lifespan; Qn is the nominal capacity; Rn is the initial internal resistance; when any criterion is met, the battery is determined to have entered the end-of-life state, and the remaining lifespan of the battery is calculated.

[0115] Step S6: Map the predicted remaining battery lifespan and lifespan expiration index (degradation index of aging model and degradation index of capacity model) to a safety status level SOS, output the battery safety risk level and corresponding warning information, and provide safety assurance for system operation.

[0116] Example 2:

[0117] Building upon Example 1, to accurately characterize the battery's operating state and performance evolution, it is necessary to systematically collect multi-dimensional operating data of the battery under different operating conditions. Voltage U reflects the potential difference in the battery's electrochemical process, current I reflects the rate characteristics of charge transfer, temperature T relates to the battery's internal thermodynamic processes, capacity Q characterizes the battery's ability to store charge, and mixed pulse power test data can capture the battery's dynamic power response characteristics. This constructs a comprehensive raw data set reflecting the battery's operating conditions. After data acquisition, the process proceeds to the preprocessing stage, the core objective of which is to extract key features characterizing battery performance degradation from the raw data.

[0118] The capacity Q includes the initial rated capacity of the battery. Measured capacity at time t Based on the initial rated capacity of the battery Measured capacity at time t Calculate capacity decay and capacity decay rate :

[0119] ;

[0120] ;

[0121] Capacity decay and capacity decay rate This constitutes the capacity decay characteristic, which reflects the degree to which the battery's storage capacity decreases over time or the number of cycles.

[0122] To address the characteristics of internal resistance growth, hybrid pulse power test data is used. Based on voltage transient response analysis and combined with Ohm's law and equivalent circuit model, the ohmic internal resistance is calculated from the voltage change ΔU and current change ΔI under pulse current excitation. :

[0123] ;

[0124] in, This represents the instantaneous change in voltage at the lower terminal caused by the pulsed current excitation. The change is due to the current; by comparing the ohmic internal resistance at different times... The internal resistance growth characteristics are extracted by comparing the internal resistance values ​​at different times. These characteristics are arranged according to time series and cycle number. The internal resistance growth characteristics reflect the evolution of the battery's internal electrochemical impedance. These fundamental parameters provide crucial inputs for advanced applications such as battery health status assessment and remaining life prediction, supporting in-depth analysis and accurate prediction of battery performance.

[0125] Example 3:

[0126] Based on Example 1 or 2, the capacity-cycle number curve is described. The analysis yielded the knee point and knee point value. The method is as follows: For the capacity-cycle count curve... Find the derivative, locate the inflection point of accelerated capacity decay, and determine the knee point and its value from the inflection point. .

[0127] In-depth analysis is conducted by focusing on the capacity-cycle-count curve. This curve contains crucial information about the battery aging process. The knee point is identified using the derivative variation pattern: by differentiating the capacity-cycle-count curve Q(t) (the first derivative dQ / dt reflects the rate of capacity change with cycle count, and the second derivative helps determine the inflection point of this rate change), a significant abrupt change in the rate of change of the derivative indicates the turning point where the curve transitions from a flat curve to rapid degradation; this is the knee point. The knee point value is extracted based on this identification logic. This value indicates the switching of the battery capacity degradation mode.

[0128] Furthermore, considering the significant impact of temperature on battery aging, the knee value was adjusted based on different temperature conditions. The method for obtaining temperature-related degradation feature parameters by inputting the Arrhenius model is as follows:

[0129] Temperature in different temperature-time combinations , It covers the temperature range that batteries may encounter during actual operation, among which, , Numbering at different temperature moments; extracting the associated temperature knee values Input Arrhenius model:

[0130] ;

[0131] in, For temperature-dependent reaction rate constants, It is the natural logarithm. Let R be the temperature and R be the gas constant. The rate of aging to the knee is defined as... Therefore, let ,get:

[0132] ;

[0133] By fitting the aging rate to the knee point at different temperatures, temperature-dependent degradation characteristic parameters, i.e., prefactors, are obtained. With activation energy This enables a quantitative analysis of the temperature-dependent aging mechanism of batteries, laying the foundation for temperature-degradation correlation in subsequent lifespan prediction.

[0134] Example 4:

[0135] Based on Example 1, Example 2, or Example 3, step S3 specifically includes:

[0136] Step S31: Obtain battery capacity loss under different temperature-time combinations through experiments or historical data. ,in, The temperature in the temperature-time combination; For time in the temperature-time combination;

[0137] Step S32: Based on the Arrhenius formula, derive the relationship between capacity decay rate and temperature, and then calculate the activation energy of the battery capacity decay process. (Reflecting the sensitivity of the capacity decay reaction to temperature; the higher the activation energy, the more significant the effect of temperature on the decay rate) and the pre-factor A, specifically:

[0138] Based on Arrhenius thermodynamics and the time-cumulative characteristics of capacity loss, a model is constructed: temperature in the same temperature-time combination. Below, battery capacity loss Time in temperature-time combination It exhibits a power-law relationship, combined with the Arrhenius formula. ,get:

[0139] ;

[0140] Where k represents the decay rate and R is the gas constant; for nth power; Power;

[0141] Temperature in the same temperature-time combination Multiple groups below Take the natural logarithm:

[0142] ;

[0143] by x-axis The slope of the line fitted to the ordinate is the power. After eliminating the power-law effect of time accumulation, Normalization After normalization, its equation is:

[0144] ;

[0145] Taking the natural logarithm of the equation, we get:

[0146] ;

[0147] by x-axis By fitting a straight line to the ordinate, the activation energy can be obtained. With the previous factor ;

[0148] Step S33: Combine the temperatures in any temperature-time combination. Time in temperature-time combination Substituting the extended Arrhenius relation:

[0149] ;

[0150] in, This is the predicted value of capacity loss. The power is an adjustable factor, set to 0.5 when dealing with simple diffusion control mechanisms or layered growth processes. Since the battery aging process is mainly affected by SEI layer growth, n=0.5 in the correlation equation, yielding the capacity loss trend at different temperatures T and times t. , Predicted value of capacity loss This collection enables quantitative prediction of battery capacity degradation under complex temperature-time conditions, providing crucial lifespan evolution data for battery health management. Time in a certain temperature-time combination The nth power, that is for To raise a specific value to the power of n.

[0151] Example 5:

[0152] Based on Example 4, step S4 specifically includes:

[0153] Constructing multi-source feature input vectors :

[0154] ;

[0155] in, This represents the measured capacity at time t. Let be the ohmic internal resistance at time t. Polarization resistor, Polarized capacitor, Let t be the battery temperature. Let t be the state of charge. Let t be the number of charge-discharge cycles at time t. Let t be the capacity loss. This represents the number of iterations at the knee point corresponding to time t, used to help capture the decay inflection point;

[0156] This approach comprehensively characterizes the static attributes and dynamic evolution trends of battery states. The constructed multi-source feature input vector is fed into a Long Short-Term Memory (LSTM) network, utilizing its gating units to achieve selective memorization and updating of temporal information.

[0157] Input gate : via the sigmoid function The formula for controlling the input weights of new information is:

[0158] ;

[0159] in, The input weight matrix of the input gate is used to input the multi-source feature vector at the current time step. The weight matrix mapped to the input gate; The hidden state weight matrix for the input gate is used to capture the hidden state at the previous time step. Mapped to the input gate; This is the bias term for the input gate, used to compensate for the activation offset of the input gate after weighted summation;

[0160] Forgotten Gate : ;

[0161] Forgotten Gate Decide on the hidden state at the previous moment Information that needs to be retained, including: The input weight matrix for the forget gate is the multi-source feature input vector at the current time step. Mapped to the forget gate dimension, it is used to learn the regulatory weights of input features on the degree of retention of historical information; The hidden state weight matrix for the forget gate is used to capture the hidden state of the previous time step. Forgetting decision characteristics with time-dependent characteristics (such as memory inertia with a decaying trend under continuous cycles); This is the bias term for the forget gate, used to compensate for the mean shift between the input features and the hidden state, and to optimize the activation baseline of the forget gate.

[0162] Candidate Memory :pass The formula for generating potentially updated memory content is as follows:

[0163] ;

[0164] in, The input weight matrix for candidate memories is used to learn the current multi-source feature input vector. Contribution to the generation of new memories (e.g., the number of charge-discharge cycles at time t) (Incremental impact on capacity loss); Given the hidden state weight matrix of the candidate memory, we can mine the previous hidden state. The "new memory fit" feature of mid-temporal association (such as the constraint of historical decay rate on current potential memory); It serves as a bias term for candidate memories, compensating for the activation baseline of candidate memories and adapting to the nonlinear characteristics of battery aging (such as abrupt changes at the decay inflection point).

[0165] Memory cell update: Combining the input gate and the forgetting gate, the state of memory cells is updated as follows:

[0166] ;

[0167] in, The cumulative battery aging information stored in the memory cell at time t-1 is ⊙, where ⊙ represents element-wise product.

[0168] Output gate : Controlling the state of memory cells to a hidden state The formula for its output is:

[0169] ;

[0170] in, The input weight matrix of the output gate is used to learn the current multi-source feature input vector. Regulation of memory output decisions (such as the filtering of decay information output by the SOC state). The hidden state weight matrix of the output gate captures the previous hidden state. Medium-time dependent output decision inertia (such as the continuous output characteristics of attenuation information under continuous high load), The output gate bias term compensates for the activation offset of the output gate and optimizes the output accuracy in the hidden state;

[0171] The final output formula is:

[0172] ;

[0173] Based on the aforementioned gating mechanism, the Long Short-Term Memory (LSTM) network learns the dependencies of battery states in long-sequence inputs, enabling multi-step rolling predictions within a time window H for future prediction, and outputting trajectory predictions of capacity and internal resistance. It covers the short-term and medium-term evolution trends of battery aging process. The capacity is the predicted value within the time window length H of future forecasting; The predicted value of the internal resistance within the time window length H of future prediction.

[0174] Furthermore, it also includes employing a multi-objective loss function. Training the Long Short-Term Memory Network:

[0175] ;

[0176] in, This is the capacity forecast value. This is the predicted internal resistance value. This is the actual capacity value. Let θ be the true internal resistance value, α be the weight of the capacity prediction task, β be the weight of the internal resistance prediction task, Θ be the parameters of the Long Short-Term Memory network model, λ be the coefficient of the L2 regularization term, and H be the length of the future prediction time window. By minimizing this loss function, the model can learn the battery state evolution law while ensuring prediction accuracy and generalization ability, providing multi-step time-series prediction support for battery health status assessment and remaining life prediction.

[0177] Example 6:

[0178] Based on Example 5, the method for calculating the remaining service life in step S5 is as follows:

[0179] Based on the trajectory prediction value of the output capacity and internal resistance Iterate through the data to find the first time that satisfies the capacity threshold condition. and the first moment when the internal resistance threshold condition is met. The smaller of these two times is taken as the actual end time of the battery's lifespan. Remaining service life (RUL):

[0180] RUL= ;in, This refers to the current moment.

[0181] Furthermore, it also includes defining a normalized capacity path termination lifetime index. With internal resistance path termination lifetime index ,have:

[0182] Capacity Path End-of-Life Index : ;

[0183] in, Nominal capacity The measured capacity at time t The physical meaning is the proportion of the current capacity decay to the total capacity decay at the end of its life;

[0184] Internal resistance path termination lifetime index : ;

[0185] in, As for the current internal resistance, The initial internal resistance, Used to quantify the proportion of current internal resistance growth to the total internal resistance growth at the end of life. That is, the degradation index of the aging model. That is, the degradation index of the capacity model;

[0186] right and The values ​​are truncated and limited to the range of [0-1] to avoid the exponents going out of bounds due to abnormal data. This allows the two exponents to reflect the aging process of battery capacity and internal resistance more stably and intuitively, providing a basic indicator for subsequent safety status assessment.

[0187] Example 7:

[0188] Based on Example 6, step S6 specifically includes:

[0189] Construct a convergence risk score S, which integrates remaining lifetime risk, capacity aging risk, and internal resistance aging risk;

[0190] ;

[0191] in This is used to eliminate negative risks (ensuring that the risk score is non-negative). The reference lifespan can be selected from battery design lifespan, preset maintenance prediction cycle, etc. Reflects the deviation of remaining service life from reference service life; the closer the remaining service life is to the reference service life, the better. The higher this value, the higher the risk of remaining life expectancy; This is the capacity path termination lifetime index. As a weight for capacity aging risk, The larger the value, the closer the capacity decay is to the end-of-life threshold, and the higher the risk. This is the internal resistance path termination lifetime index. As the internal resistance aging risk weight; The larger the value, the closer the internal resistance increases to the end-of-life threshold, and the higher the risk. For task weights, the following must be satisfied: This is used to balance the impact of different risk factors;

[0192] Security Status Index Mapping: The fused risk score S is linearly mapped to the [0-8] interval to generate an intuitive and easily identifiable security status level. : ;

[0193] in, Used to limit the range of values ​​for S, ensuring that the safe state after mapping is an integer between 0 and 8; This is a rounding function to make the safety status classification more in line with engineering decision-making habits.

[0194] Based on the safety status values, a tiered handling strategy is defined as "Normal—Attention—Power Limitation—Recommended Replacement—Forced Replacement / Isolation":

[0195] Safety status = 0~2: The battery is in normal safety status and can be continuously monitored.

[0196] Safety Status = 3~4: Entering the attention level, it is necessary to increase the frequency of inspections and focus on investigating the causes of abnormalities;

[0197] Safety status = 5~6: Trigger power limiting suggestion, reduce battery charging and discharging power to avoid accelerated aging;

[0198] Safety status = 7~8: Perform replacement / isolation operations, replace faulty batteries in a timely manner, and ensure the safe operation of the system.

[0199] This invention starts from the state of a single battery cell and supports hierarchical early warning systems that extend upwards to the module system level, combined with the mechanistic parameters in steps S1-S5. With real-time running characteristics The synergistic drive ensures consistency across operating conditions and improves the prediction accuracy in the nonlinear degradation stage.

[0200] Although the present invention has been described herein with reference to illustrative embodiments, the above embodiments are merely preferred embodiments of the present invention, and the implementation of the present invention is not limited to the above embodiments. It should be understood that those skilled in the art can devise many other modifications and implementations, which will fall within the scope and spirit of the principles disclosed in this application.

Claims

1. A method for early warning and assessment of battery safety risk levels, characterized in that, include: Step S1: Collect battery operating data and process it to obtain capacity decay characteristics and internal resistance growth characteristics. The battery operating data includes voltage U, current I, temperature T, capacity Q, and mixed pulse power test data. Step S2: Obtain the capacity-cycle count curve based on the capacity decay characteristics. For the capacity-cycle count curve The analysis yielded the knee point and knee point value. Based on different temperature conditions, the knee point value Input the Arrhenius model to obtain temperature-related degradation characteristic parameters; Step S3: Calculate the activation energy Ea and the pre-factor A of the battery capacity decay process according to the Arrhenius formula, and then estimate the capacity loss trend under different temperatures and times. Step S4: Using multi-source time-series data and Arrhenius feature parameters as input, construct a long short-term memory network to predict the capacity and internal resistance changes in the next few cycles; the multi-source time-series data includes capacity, internal resistance, temperature, state of charge and number of cycles; Step S5: Based on the predicted capacity and internal resistance trajectory, combined with the lifespan termination criterion: Q EOL ≤0.8Qn or R EOL ≥1.33Rn; Among them, Q EOL R is the capacity at the end of its lifespan. EOL Rn is the internal resistance at the end of its lifespan; Qn is the nominal capacity, and Rn is the nominal internal resistance; when any criterion is met, the battery is determined to have entered the end-of-life state, and the remaining lifespan is calculated. Step S6: Fuse the remaining useful life with the degradation index of the aging model and the degradation index of the capacity model to generate a safety status index, specifically including: Construction fusion risk score S: ; in Used to eliminate negative risks. For reference lifespan, Reflects the deviation of remaining service life from reference service life; the closer the remaining service life is to the reference service life, the better. The higher this value, the higher the risk of remaining life expectancy; This is the capacity path termination lifetime index. As a weight for capacity aging risk, The larger the value, the closer the capacity decay is to the end-of-life threshold, and the higher the risk. This is the internal resistance path termination lifetime index. As the internal resistance aging risk weight; The larger the value, the closer the internal resistance increases to the end-of-life threshold, and the higher the risk. For task weights, the following must be satisfied: This is used to balance the impact of different risk factors; Security Status Index Mapping: The fused risk score S is linearly mapped to the [0-8] interval to generate a security status level. : ; in Used to limit the range of values ​​for S, ensuring that the safe state after mapping is an integer between 0 and 8; The rounding function makes the safety status classification more in line with engineering decision-making habits; It also outputs battery safety risk classification and corresponding warning information according to the safety status level.

2. The battery safety risk level early warning and assessment method according to claim 1, characterized in that, The capacity Q includes the initial rated capacity of the battery. Measured capacity at time t Based on the initial rated capacity of the battery Measured capacity at time t Calculate capacity decay and capacity decay rate : ； ； Capacity decay and capacity decay rate This constitutes the capacity decay characteristic.

3. The battery safety risk level early warning and assessment method according to claim 1, characterized in that, The method for calculating the internal resistance growth characteristic is as follows: Calculate the ohmic internal resistance using mixed pulse power test data. : ； in, This represents the instantaneous change in voltage at the lower terminal caused by the pulsed current excitation. The change is due to the current; by comparing the ohmic internal resistance at different times... The numerical values ​​are arranged according to the time series and the number of cycles to obtain the characteristics of internal resistance growth.

4. The battery safety risk level early warning and assessment method according to claim 1, characterized in that, The capacity-cycle count curve The analysis yielded the knee point and knee point value. The method is as follows: For the capacity-cycle count curve... Find the derivative, locate the inflection point of accelerated capacity decay, and determine the knee point and its value from the inflection point. .

5. The battery safety risk level early warning and assessment method according to claim 1, characterized in that, The knee point value is determined by combining different temperature conditions. The method for obtaining temperature-related degradation feature parameters by inputting the Arrhenius model is as follows: Temperature in different temperature-time combinations ,in, , Numbering at different temperature moments; extracting the associated temperature knee values Input Arrhenius model: ； in, For temperature-dependent reaction rate constants, It is the natural logarithm. Let R be the temperature and R be the gas constant. The rate of aging to the knee is defined as... Therefore, let ,get: ； By fitting the aging rate to the knee point at different temperatures, temperature-dependent degradation characteristic parameters, i.e., prefactors, are obtained. With activation energy .

6. The battery safety risk level early warning and assessment method according to claim 5, characterized in that, Step S3 specifically involves: Step S31: Obtain battery capacity loss under different temperature-time combinations through experiments or historical data. ,in, The temperature in the temperature-time combination; For time in the temperature-time combination; Step S32: Based on Arrhenius thermodynamics and the time-cumulative characteristics of capacity loss, construct a model: temperature in the same temperature-time combination. Below, battery capacity loss Time in temperature-time combination It exhibits a power-law relationship, combined with the Arrhenius formula. ,get: ； Where k represents the decay rate; for nth power; Power; Temperature in the same temperature-time combination Multiple groups below Take the natural logarithm: ； by x-axis The slope of the line fitted to the ordinate is the power. After eliminating the power-law effect of time accumulation, Normalization After normalization, its equation is: ； Taking the natural logarithm of the equation, we get: ； by x-axis By fitting a straight line to the ordinate, the activation energy can be obtained. With the previous factor ; Step S33: Combine the temperatures in any temperature-time combination. Time in temperature-time combination Substituting the extended Arrhenius relation: ； in, The predicted capacity loss values ​​are obtained, showing the capacity loss trends at different temperatures T and times t. , Predicted value of capacity loss A set; Time in a certain temperature-time combination The nth power.

7. The battery safety risk level early warning and assessment method according to claim 6, characterized in that, Step S4 specifically includes: Constructing multi-source feature input vectors : ； in, The measured capacity at time t is represented as . Let be the ohmic internal resistance at time t. Polarization resistor, Polarized capacitor, Let t be the battery temperature. Let t be the state of charge. Let t be the number of charge-discharge cycles at time t. Let t be the capacity loss. This represents the number of iterations at the knee point corresponding to time t, used to help capture the decay inflection point; The constructed multi-source feature input vector is fed into a long short-term memory network, and its gating units are used to achieve selective memorization and updating of temporal information. Input gate : via the sigmoid function The formula for controlling the input weights of new information is: ； in, The input weight matrix of the input gate is used to input the multi-source feature vector at the current time step. The weight matrix mapped to the input gate; The hidden state weight matrix for the input gate is used to capture the hidden state at the previous time step. Mapped to the input gate; This is the bias term for the input gate, used to compensate for the activation offset of the input gate; Forgotten Gate : ; Forgotten Gate Decide on the hidden state at the previous moment Information that needs to be retained, including: The input weight matrix for the forget gate is the multi-source feature input vector at the current time step. Mapped to the forget gate dimension, it is used to learn the regulatory weights of input features on the degree of retention of historical information; The hidden state weight matrix for the forget gate is used to capture the hidden state of the previous time step. Forgetting decision characteristics with time-dependent factors; This is the bias term for the forget gate, used to compensate for the mean shift between the input features and the hidden state, and to optimize the activation baseline of the forget gate. Candidate Memory :pass The formula for generating potentially updated memory content is as follows: ； in, The input weight matrix for candidate memories is used to learn the current multi-source feature input vector. Contribution to the generation of new memories; Given the hidden state weight matrix of the candidate memory, we can mine the previous hidden state. New memory adaptation features in mid-temporal association; It serves as a bias term for candidate memories, compensating for the activation baseline of candidate memories and adapting to the nonlinear characteristics of battery aging. Memory cell update: Combining the input gate and the forgetting gate, the state of memory cells is updated as follows: ； in, The cumulative battery aging information stored in the memory cell at time t-1 is ⊙, where ⊙ represents element-wise product. Output gate : Controlling the state of memory cells to a hidden state The formula for its output is: ； in, The input weight matrix of the output gate is used to learn the current multi-source feature input vector. Regulation of memory output decisions, The hidden state weight matrix of the output gate captures the hidden state at the previous time step. Intermediate time-dependent output decision inertia The output gate bias term compensates for the activation offset of the output gate and optimizes the output accuracy in the hidden state; The final output formula is: ； Based on the aforementioned gating unit, the Long Short-Term Memory (LSTM) network learns the dependencies of battery states in long-sequence inputs, enabling multi-step rolling predictions within a time window length H for future prediction, and outputting trajectory predictions of capacity and internal resistance. It covers the short-term and medium-term evolution trends of battery aging process. The capacity is the predicted value within the time window length H of future forecasting; The predicted value of the internal resistance within the time window length H of future prediction.

8. The battery safety risk level early warning and assessment method according to claim 7, characterized in that, It also includes using a multi-objective loss function. Training the Long Short-Term Memory Network: ； in, This is the capacity forecast value. This is the predicted internal resistance value. This is the actual capacity value. Let θ be the true value of internal resistance, α be the weight of the capacity prediction task, β be the weight of the internal resistance prediction task, Θ be the parameters of the long short-term memory network model, λ be the coefficient of the L2 regularization term, and H be the length of the time window for future prediction.

9. The battery safety risk level early warning and assessment method according to claim 7, characterized in that, The method for calculating the remaining useful life in step S5 is as follows: Based on the trajectory prediction value of the output capacity and internal resistance Iterate through the data to find the first time that satisfies the capacity threshold condition. and the first moment when the internal resistance threshold condition is met. The smaller of these two times is taken as the actual end time of the battery's lifespan. Remaining service life (RUL): RUL= ;in, The current moment; It also includes defining a normalized capacity path termination lifetime index. With internal resistance path termination lifetime index ,have: Capacity Path End-of-Life Index : ; in, Nominal capacity; The measured capacity at time t; The physical meaning is the proportion of the current capacity decay to the total capacity decay at the end of its life; Internal resistance path termination lifetime index : ; in, As for the current internal resistance, The initial internal resistance, Used to quantify the proportion of current internal resistance growth to the total internal resistance growth at the end of life. right and The value is truncated to limit its range to [0-1] to prevent the exponent from going out of bounds due to abnormal data.

Images