Method and device for predicting service life of battery pack and storage medium

Abstract

The invention relates to the technical field of battery life prediction, and discloses a battery pack life prediction method and device and a storage medium, and the method comprises the steps: obtaining a standardized monitoring data matrix; generating a reconstructed electrochemical impedance spectroscopy data matrix according to the standardized monitoring data matrix and a preset impedance spectroscopy reconstruction algorithm; extracting a plurality of health factors according to a nonlinear manifold learning method; obtaining a predicted life end time point of the battery pack according to the target differential equation model; wherein the target differential equation model is obtained by performing constraint condition fusion, scale expansion, constraint condition intensity adjustment and control of balance of a known area and an unknown area on the initial neural differential equation model; the initial neural differential equation is used for describing the evolution dynamics of the health factors along with time. According to the method, high-precision, interpretable and high-adaptability battery life prediction is realized, and the method has important application value in the fields of electric automobiles, energy storage systems and electronic equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of battery life prediction, and more particularly, to a method, device and storage medium for battery pack life prediction. Background Art

[0002] With the rapid development of electric vehicles, large-scale energy storage devices and portable electronic devices, batteries, as key components of energy storage, have an important role to play in ensuring the safe operation of the system, optimizing maintenance plans, improving user experience and reducing operating costs.

[0003] The existing technology has the following problems: the internal health status of the battery is difficult to measure directly, especially measurement methods such as electrochemical impedance spectroscopy that reflect the internal status usually require special equipment and cannot be collected in real time, resulting in the model's lack of direct description of the internal degradation mechanism; the battery degradation process also includes multi-time scale characteristics ranging from millisecond electrochemical reactions to monthly structural aging, and existing models find it difficult to simultaneously capture the complex interactions between these characteristics; battery operating conditions and usage patterns are diverse, and the importance of health factors changes with the state. Traditional static models are difficult to adapt to changes in battery characteristics under different working conditions; obtaining a complete data set for the entire battery life cycle is costly and time-consuming, and data on different types of batteries and usage scenarios is scarce, making it difficult for the model to cope with new batteries or different application scenarios.

[0004] Therefore, it is necessary to develop a battery life prediction method that can accurately characterize the internal health status of the battery, capture the degradation process at multiple time scales, and have reasonable data requirements. Summary of the Invention

[0005] The present invention provides a method, device and storage medium for battery pack life prediction, which solve the technical problems in related technologies such as the difficulty in characterizing the internal health status of batteries, the difficulty in unified modeling of multi-time scale degradation processes, and the insufficient adaptability of static models.

[0006] In a first aspect of the present invention, a method for predicting battery life is provided, comprising: Collecting battery pack monitoring data and preprocessing it to obtain a standardized monitoring data matrix; wherein the standardized monitoring data matrix includes a standard voltage data sequence and a standard current data sequence; Generate a reconstructed electrochemical impedance spectrum data matrix based on the standard voltage data sequence, the standard current data sequence, and a preset impedance spectrum reconstruction algorithm; wherein the reconstructed electrochemical impedance spectrum matrix is ​​reconstructed and weightedly fused according to each preset stage of the battery pack charge and discharge process; Using a nonlinear manifold learning approach, several health factors are extracted from a standardized monitoring data matrix and a reconstructed electrochemical impedance spectroscopy data matrix; each health factor has a corresponding physicochemical degradation mechanism; and each health factor corresponds to at least one key battery pack aging parameter. According to the target differential equation model, the predicted end-of-life time point of the battery pack is obtained; wherein, the target differential equation model is obtained by incorporating constraints, expanding the scale, adjusting the constraint strength, and controlling the balance between known and unknown areas into the initial neural differential equation model; the initial neural differential equation is used to describe the evolution dynamics of the health factor over time.

[0007] In a preferred embodiment, the steps of incorporating constraints into the initial neural differential equation model, scaling the model, adjusting the constraint strength, and controlling the balance between known and unknown regions include: Incorporate battery electrochemical mechanism constraints into the initial neural differential equation model; The initial neural differential equation model is scaled up according to the health factor set obtained by classification according to the time scale to obtain the key neural differential equation model; wherein the key neural differential equation is a multi-time scale coupled dynamic equation model; Predicting uncertainty values ​​of key neural differential equation models at different states; Adjust the strength of the constraints on the battery electrochemical mechanism in the key neural differential equation model based on the uncertainty value; According to the preset balance strategy, the balance of the key neural differential equation model in the known and unknown areas is controlled.

[0008] In a preferred embodiment, obtaining the predicted end-of-life time point of the battery pack according to the target differential equation model includes: According to the target neural differential equation model, the future evolution trajectory of each health factor is predicted; Based on the future evolution trajectory of each health factor, the predicted end-of-life time of the battery pack is determined.

[0009] In a preferred embodiment, the scaling of the initial neural differential equation model based on the health factor set obtained by classification according to the time scale to obtain the key neural differential equation model includes: Several health factors are classified according to time scales to obtain several health factor sets; wherein each health factor set has a corresponding time scale; the time scales include a first change scale, a second change scale, and a third change scale; wherein the change speed corresponding to the first change scale is greater than the change speed corresponding to the second change scale; the change speed corresponding to the second change scale is greater than the change speed corresponding to the third change scale; each time scale has a corresponding step size; the step size corresponding to the first change scale is smaller than the step size corresponding to the second change scale; the step size corresponding to the second change scale is smaller than the step size corresponding to the third change scale; The corresponding health factor time scale change rate is solved according to the step size corresponding to the time scale of each health factor to obtain the key neural differential equation model.

[0010] In a preferred embodiment, the uncertainty of each health factor included in the key neural differential equation model is quantified based on a Bayesian neural network; Obtain the uncertainty value of each health factor under different conditions; The uncertainty values ​​of the key neural differential equation model in different states are obtained by weighted summing up the uncertainty values ​​of each health factor in different states and the preset uncertainty weights corresponding to each health factor.

[0011] In a preferred embodiment, after obtaining the predicted end-of-life time point of the battery pack according to the target differential equation model, the method further includes: Analyze the key factors affecting battery pack life and generate optimization strategies.

[0012] In a preferred embodiment, determining the predicted end-of-life time point of the battery pack based on the future evolution trajectory of each health factor includes: Determine the initial end-of-life time of the battery pack based on the future evolution trajectory of each health factor; The initial end-of-life time point is calibrated according to the historical prediction deviation and the uncertainty of each health factor to obtain the predicted end-of-life time point.

[0013] In a second aspect of the present invention, a device for predicting battery life is provided, the device comprising: An acquisition unit is used to acquire battery pack monitoring data and perform preprocessing to obtain a standardized monitoring data matrix; wherein the standardized monitoring data matrix includes a standard voltage data sequence and a standard current data sequence; A generating unit, configured to generate a reconstructed electrochemical impedance spectrum data matrix based on a standard voltage data sequence, a standard current data sequence, and a preset impedance spectrum reconstruction algorithm; wherein the reconstructed electrochemical impedance spectrum matrix is ​​obtained by reconstructing and weighted fusion according to each preset stage of the battery pack charge and discharge process; An extraction unit is configured to extract a plurality of health factors from a standardized monitoring data matrix and a reconstructed electrochemical impedance spectroscopy data matrix based on a nonlinear manifold learning method; wherein each health factor has a corresponding physicochemical degradation mechanism; and each health factor corresponds to at least one key battery pack aging parameter; The prediction unit is used to obtain the predicted end-of-life time point of the battery pack based on the target differential equation model; the target differential equation model is obtained by incorporating constraints, scaling, adjusting the constraint strength, and controlling the balance between known and unknown areas into the initial neural differential equation model; the initial neural differential equation is used to describe the evolution of health factors over time.

[0014] In a third aspect of the present invention, a non-transitory computer-readable storage medium is provided, in which at least one instruction or at least one program is stored. The at least one instruction or at least one program is loaded and executed by a processor to implement the aforementioned battery pack life prediction method.

[0015] The beneficial effects of the present invention are: By reconstructing the electrochemical impedance spectrum using conventional monitoring data, internal health status information can be obtained without the need for dedicated equipment, resolving the problem of direct measurement of internal status; then, through the scale expansion and constraint optimization of the neural differential equation model, a unified description of the multi-time scale degradation process is achieved, overcoming the limitation of traditional models that are difficult to capture complex interactions; at the same time, based on nonlinear manifold learning to extract health factors associated with physical degradation mechanisms, combined with the dynamic adjustment ability of the model, it adapts to the changes in the importance of health factors under changing working conditions; and the full-link design reduces dependence on complete data sets, enhances applicability in new batteries or data-scarce scenarios, and provides a better technical path for battery life prediction as a whole. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flowchart of a method for predicting battery life. DETAILED DESCRIPTION

[0017] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0018] At least one embodiment of the present invention discloses a method for predicting battery life. Figure 1As shown, the following steps are included: Step 1: Collect battery pack monitoring data and pre-process it to obtain a standardized monitoring data matrix; wherein the standardized monitoring data matrix includes a standard voltage data sequence and a standard current data sequence.

[0019] Specifically, the monitoring data of the battery pack is obtained, including voltage and current time series data, and these data are preprocessed, including denoising, standardization and missing value processing, to generate a standardized monitoring data matrix ,in Indicates the number of time points, Represents the number of monitoring parameters; wherein, the standardized monitoring data matrix includes a standard voltage data sequence, a standard current data sequence, and may also include a temperature data sequence, etc.

[0020] Step 2: Generate a reconstructed electrochemical impedance spectrum data matrix based on the standard voltage data sequence, the standard current data sequence, and a preset impedance spectrum reconstruction algorithm; wherein the reconstructed electrochemical impedance spectrum matrix is ​​reconstructed and weightedly fused according to each preset stage of the battery pack charge and discharge process.

[0021] Through an innovative preset impedance spectrum reconstruction algorithm, conventional charge and discharge data are used to reconstruct electrochemical impedance spectrum information, without the need for dedicated electrochemical impedance spectroscopy (EIS) measurement equipment. The specific implementation process is as follows: Based on the standard voltage data sequence and the standard current data sequence, the battery pulse response function is calculated: ; in, represents the Fourier transform operator; represents the inverse Fourier transform operator; Indicates the battery time Voltage value at the moment; Indicates the battery time Current value at the moment; represents the impulse response function of the battery.

[0022] According to the pulse response function, the electrochemical impedance at different frequencies is calculated by the following equation : ; in, represents the angular frequency; represents an imaginary unit; Indicates the frequency sine wave; Indicates the angular frequency The electrochemical impedance value under represents the impulse response function of the battery.

[0023] Through the above calculations, a simulated electrochemical impedance spectroscopy (EIS) spectrum is generated, which contains real and imaginary data. , forming the reconstructed electrochemical impedance data matrix ,in, Indicates the number of frequency points; Indicates the angular frequency Electrochemical impedance value under represents the real part of impedance; represents the imaginary part of impedance; represents a complex field; Represents the reconstructed electrochemical impedance spectroscopy data matrix.

[0024] It should be noted that in order to improve the reconstruction accuracy, this algorithm adopts a segmented reconstruction strategy, which divides the battery charging and discharging process into multiple stages (such as constant current charging, constant voltage charging, static, discharging, etc.), reconstructs the impedance spectrum for each stage separately, and then obtains the complete impedance spectrum through weighted fusion.

[0025] For different frequency bands, the signal segment that best suits the frequency characteristics is selected for reconstruction: The low frequency band (0.01-0.1Hz) mainly uses long-term static and low-rate charge and discharge data; The mid-frequency band (0.1-100Hz) uses pulse charge and discharge data; The high frequency band (100-1000Hz) uses high frequency disturbance response data.

[0026] Taking lithium-ion batteries as an example, this algorithm can reconstruct an impedance spectrum with a frequency range of 0.01-1000 Hz, encompassing 25 frequency points, from a standard 1C charge-discharge curve. The reconstructed impedance spectrum can be used to identify various degradation mechanisms within the battery, such as SEI film growth (semicircular features in the low-frequency region), limited lithium-ion diffusion (slanted lines in the mid-frequency region), and increased charge transfer impedance (semicircular features in the high-frequency region).

[0027] Step 3: Extract several health factors from the standardized monitoring data matrix and the reconstructed electrochemical impedance spectroscopy data matrix based on a nonlinear manifold learning method; each health factor has a corresponding physicochemical degradation mechanism; and each health factor corresponds to at least one key parameter of battery pack aging.

[0028] Specifically, construct the enhanced feature matrix , combining the normalized monitoring data matrix with the reconstructed electrochemical impedance spectroscopy data matrix, where represents the standardized monitoring data matrix, Indicates the number of time points, Indicates the number of monitoring parameters, Represents the reconstructed electrochemical impedance spectroscopy data matrix, Indicates the number of frequency points.

[0029] Extract low-dimensional representations through nonlinear manifold learning algorithms: ; in, represents the nonlinear manifold learning function; represents the enhanced feature matrix; represents the target dimension; Represents the extracted health factor matrix.

[0030] The extracted health factors are interpreted physically. By analyzing the correlation with key parameters in the battery aging process (such as capacity, internal resistance, etc.), the physical meaning of each health factor is determined to form a health factor interpretation map. .

[0031] This embodiment uses an improved local linear embedding (LLE) algorithm to enhance the physical meaning preservation capability of manifold learning by introducing a weight matrix based on the physical correlation of the measured data. Specifically, the weight calculation formula is: ; in, Represents sample points With its neighbors The weight coefficient between Represents sample points and similarity between Indicates the number of neighbors considered for each sample point; A normalization factor for all neighbor similarities, ensuring that the sum of the weights is 1.

[0032] Similarity function Defined as: ; in, Sample points and similarity between Represents sample points and Squared Euclidean distance in feature space; is the feature space distance parameter; Represents sample points and The absolute value of the interval in time; is the time correlation parameter; Represents the exponential function.

[0033] Using commercial 18650 lithium-ion batteries as an example, this method was applied to monitoring data from a two-year service life, successfully extracting 10 health factors, including SEI film impedance factor, active material loss factor, lithium plating factor, electrolyte decomposition factor, and microcrack growth factor. These health factors correspond to different physical and chemical degradation mechanisms within the battery. Verification experiments show that compared to capacity estimation methods based solely on monitoring data, this method can detect changes in battery degradation trends 200-300 cycles in advance, providing a sufficient time window for preventive maintenance.

[0034] This embodiment outputs a set of health factors that characterize the internal state of the battery. In addition, the physical meaning mapping corresponding to each health factor can also be output , providing a basis for subsequent modeling.

[0035] Step 4: Obtain the predicted end-of-life time point of the battery pack based on the target differential equation model. The target differential equation model is obtained by incorporating constraints into the initial neural differential equation model, scaling it, adjusting the constraint strength, and controlling the balance between known and unknown regions. The initial neural differential equation is used to describe the dynamic evolution of health factors over time.

[0036] Specifically, the core dynamic equation of the initial neural differential equation model is: ; in, represents the health factor vector, is the health factor dimension; represents the external input vector; Indicates time; It is determined by the parameters Determined neural network function; represents the time derivative of the health factor vector.

[0037] Neural Networks A multi-layer perceptron structure is used, and the specific configuration is as follows: Input layer: receives health factors , external input and time , the dimension is ; Hidden layers: 2-3 fully connected layers, each using the hyperbolic tangent (tanh) activation function; Output layer: fully connected layer, outputs the rate of change of health factors , the dimension is .

[0038] In specific implementation, taking electric vehicle battery pack as an example, the health factor dimension can be selected , including core health indicators such as capacity attenuation rate, internal resistance growth rate, Coulomb efficiency, and temperature sensitivity; external input dimension , including charging current, discharging current, ambient temperature, depth of charge and number of cycles.

[0039] The neural network adopts the following structure: 16 neurons in the input layer, 64 neurons in the first hidden layer, 32 neurons in the second hidden layer, and 10 neurons in the output layer, all using L2 regularization to avoid overfitting.

[0040] During training, an adaptive learning rate optimizer is used, with an initial learning rate set to 0.001. The learning rate is halved after five consecutive epochs when the validation loss stops decreasing. The training dataset includes battery data from different operating conditions and aging stages to ensure model generalization. After model training, a numerical ordinary differential equation solver (such as the fourth-order Runge-Kutta method) is used to determine the time-dependent evolution of the battery health factor.

[0041] The initial neural differential equation model is subjected to constraint integration, scale expansion, constraint strength adjustment, and control of the balance between known and unknown regions, and the following steps are also included: Step 4.1: Incorporate the battery electrochemical mechanism constraints into the initial neural differential equation model.

[0042] Specifically, a set of constraint functions is constructed based on the electrochemical mechanism ,in, is the number of constraints, Indicates the A constraint function, represents the health factor vector, Indicates time.

[0043] Common constraints include: Monotonicity constraint: Certain health factors (such as capacity degradation, internal resistance growth, etc.) have monotonic variation characteristics over time ; in, represents the monotonicity constraint function; Indicates health factor the expected direction of change; Indicates health factor Partial derivatives with respect to time; Indicates health factor the expected direction of change; Indicates health factor Partial derivatives with respect to time; The function ensures that the constraint value is non-negative and is 0 when the health factor changes in the expected direction.

[0044] Change rate constraints: The rate of change of health factors is usually physically limited: ; in, represents the rate-of-change constraint function; Indicates health factor The absolute value of the rate of change; Indicates health factor The maximum permissible rate of change; The function ensures that the constraint value is non-negative.

[0045] Relationship constraints between health factors: Physical relationships between different health factors: ; in, Represents the relationship constraint function between health factors; Represents the physical relationship function that should be satisfied between health factors; Indicates the absolute value of the deviation of the physical relationship from the ideal value 0; Integrate these constraints into the loss function of the initial neural differential equation: ; in, represents the constraint loss function, is the weight coefficient of each constraint; Indicates that all The weighted sum of constraint functions, is the total number of constraint functions, Indicates the The constraint function is in the health factor and time The value below.

[0046] Step 4.2, scale-up the initial neural differential equation model according to the health factor set obtained by classification according to the time scale to obtain the key neural differential equation model; wherein the key neural differential equation is a multi-time-scale coupled dynamic equation model.

[0047] Here, first, several health factors are classified according to the time scale to obtain several health factor sets; wherein each health factor set has a corresponding time scale; the time scale includes a first change scale, a second change scale and a third change scale; wherein the change speed corresponding to the first change scale is greater than the change speed corresponding to the second change scale; the change speed corresponding to the second change scale is greater than the change speed corresponding to the third change scale; each time scale has a corresponding step size; the step size corresponding to the first change scale is smaller than the step size corresponding to the second change scale; the step size corresponding to the second change scale is smaller than the step size corresponding to the third change scale.

[0048] Finally, the time scale change rate of each health factor is solved according to the step size corresponding to the time scale of each health factor to obtain the key neural differential equation model.

[0049] Specifically, the health factor set is decomposed into subsets of different time scales: ,in, 、 、 They represent the health factors of fast changes (first change scale), medium changes (second change scale) and slow changes (third change scale) respectively.

[0050] Then, the multi-time-scale coupled dynamic equations are constructed: ; ; ; in, represents a subset of health factors that change rapidly; A subset of health factors representing moderately rapid changes; A subset of health factors representing slow changes; represents the external input vector; represents the time variable; 、 and They are the neural network functions corresponding to fast changes, medium changes, and slow changes respectively; 、 and Represents the rate of change of fast change, medium change and slow change respectively.

[0051] To improve computational efficiency, different step sizes are used for solving different time scales: fast-changing factors use small step sizes (such as seconds), medium-changing factors use medium step sizes (such as hours), and slow-changing factors use large step sizes (such as days) for numerical integration, thereby significantly reducing the computational complexity while ensuring computational accuracy.

[0052] The output of this step is the key neural differential equation model, including the neural network parameters and constraint set , providing a basis for subsequent uncertainty perception learning and lifespan prediction.

[0053] Step 4.3, predict the uncertainty value of the key neural differential equation model under different states.

[0054] First, the uncertainty of each health factor included in the key neural differential equation model is quantified based on the Bayesian neural network.

[0055] Afterwards, the uncertainty value of each health factor under different states is obtained; Finally, the uncertainty value of each health factor in different states and the preset uncertainty weight corresponding to each health factor are weighted summed to obtain the uncertainty value of the key neural differential equation model in different states.

[0056] Specifically, the Bayesian neural network algorithm is used to quantify the uncertainty of the dynamic evolution model and the neural network parameters Introducing prior distribution , and obtain the posterior distribution through variational inference ,in, represents the set of parameters of the neural network, Representation parameters The prior distribution of Representation parameters The posterior distribution of is approximated, A set of parameters representing a variational distribution.

[0057] For each health factor In state and time Calculate the uncertainty of the following prediction: ; in, Indicates health factor In state and time uncertainty in forecasts under Represents the posterior distribution of the parameter The variance calculated below; Represents the neural network's response to health factor The predicted output of Indicates time Health factor state vector at time ; Indicates time The external input vector at time .

[0058] The uncertainty of the overall state is defined as the weighted sum of the uncertainties of each health factor: ; in, Indicates status In time A measure of the overall uncertainty of Indicates the total number of health factors; Indicates health factor Importance weight of Indicates that all The weighted sum of the health factors.

[0059] In step 4.4, adjust the strength of the battery electrochemical mechanism constraints in the key neural differential equation model based on the uncertainty value.

[0060] Specifically, design the constraint strength adjustment function: ; in, Indicates that the status and time Dynamically adjusted constraint strength under ; Indicates the strength of the basic constraint; represents the adjustment coefficient; Indicates that the status and time The uncertainty measure of the prediction under .

[0061] During model training or inference, the dynamically adjusted constraint strength is applied to the loss function: ; in, represents the total loss function; represents the data fitting loss; represents the sum of weighted constraints; Indicates the total number of constraints; Indicates the The dynamically adjusted weights of the constraints; Indicates the A constraint function.

[0062] In step 4.5, according to the preset balance strategy, the balance of the key neural differential equation model in the known area and the unknown area is controlled.

[0063] Specifically, during the training of the key neural differential equation model, the data sampling strategy is adjusted to balance the model's performance in known areas and its exploration of unknown areas: ; in, Indicates the selection of data points The probability of training; Indicates proportionality to the relationship; represents the exponential function; represents the parameter that controls the degree of exploration; Indicates that the status and time The forecast uncertainty.

[0064] Exploration and utilization strategy in the inference phase: When predicting battery life, uncertainty information is taken into account and prediction results with different confidence levels are generated: ; ; in, Indicates a point in the future Predicted value of health factors; Indicates the current time point Health factor status; Indicates from the current time To the future time The integral of the change in health factors; Indicated by the parameter A determined neural network function that describes the rate of change of the health factor; Indicates a point in the future 95% confidence interval of ; and represent the lower and upper bounds of the confidence interval, respectively.

[0065] In specific application scenarios, this model is particularly well-suited for predicting changes in battery operating conditions or new battery models. For example, for electric vehicle battery packs, when driving habits change (e.g., from highway commuting to short-distance urban use), traditional fixed-constraint models struggle to adapt. However, this method automatically detects this change in operating conditions and appropriately relaxes physical constraints in the new operating range, allowing the model to learn new degradation patterns from the data.

[0066] The strategy also dynamically adjusts the sampling frequency, automatically increasing the data collection frequency when it detects that the battery has entered a rapid aging stage or the operating conditions have changed significantly, and otherwise reducing the sampling frequency to save computing resources.

[0067] In an exemplary embodiment of the present application, obtaining a predicted end-of-life time point of a battery pack according to a target differential equation model further includes the following steps: First, based on the target neural differential equation model, the future evolution trajectory of each health factor is predicted.

[0068] Here, given the current time point Health factor status and forecast of usage conditions in the future , by numerically solving the neural differential equation, we can obtain the trajectory of health factors in the future time period: ; in, Indicates the current time point; Indicates the health factor status at the current time point; Indicates the time span of the forecast; Indicates time Operating conditions; Indicates time At the current time To the future time within the scope of Indicates time Health factor status; Indicates that the parameter is Neural network function; Represents a numerical solver for differential equations.

[0069] Taking into account the evolution characteristics of health factors at different time scales, a multi-scale solution strategy is adopted: Rapidly changing health factors , using a smaller step size for fine solution; Health factors for moderate changes , using a medium step size to solve; Slowly changing health factors , using a larger step size to solve.

[0070] Secondly, based on the future evolution trajectory of each health factor, the predicted end-of-life time of the battery pack is determined.

[0071] First, the initial end-of-life time point of the battery pack is determined based on the future evolution trajectory of each health factor. Finally, the initial end-of-life time point is calibrated based on historical prediction deviations and the uncertainty of each health factor to obtain the predicted end-of-life time point.

[0072] Here, the battery life end judgment criteria are established , common judgment criteria include: Capacity degradation criterion: When the capacity health factor Drop to below 80% of rated capacity; Internal resistance growth criterion: when the internal resistance health factor Increase to more than 150% of the initial value; Comprehensive performance criteria: A weighted combination of indicators based on multiple health factors.

[0073] Prediction-based health factor trajectory , find the earliest time point that meets the end-of-life criterion: ; in, Indicates time The health factor state vector of Indicates the current time point; Indicates the time span of the forecast; Indicates the time interval from the current time to the predicted end point; Represents the end-of-life judgment function, which returns true when the battery health status reaches the preset termination condition; Indicates the minimum operation; Indicates the predicted remaining battery life.

[0074] If the end of life is not reached within the forecast timeframe, the forecast timeframe will need to be extended.

[0075] The calibration process is as follows: By integrating historical forecast deviations and uncertainty information, the forecast results are calibrated to improve forecast accuracy. The specific implementation is as follows: Record historical forecast deviations and build a deviation correction model: ; in, Indicates the health factor status , operating conditions and time Prediction bias under Indicates the actual observed end time of battery life; Indicates the historically predicted end of battery life.

[0076] Estimate the bias of the current forecast based on the characteristics and uncertainty of the current state: ; in, Indicates the current time point estimates of forecast bias; Indicates the current time point The health factor state vector of Indicates the current time point Operating conditions; a measure of uncertainty in the prediction representing the current health state and point in time; Indicates that the parameter is The deviation estimation function of A set of parameters representing the bias estimation function.

[0077] Calibrate the prediction results: ; in, Indicates the remaining battery life prediction result after calibration; represents the original predicted remaining battery life; represents the estimated forecast bias.

[0078] In an exemplary embodiment of the present invention, after obtaining the predicted end-of-life time point of the battery pack according to the target differential equation model, the method further includes: Analyze the key factors affecting battery pack life and generate optimization strategies.

[0079] Specifically, the outputs of the previous steps are integrated to generate a complete result of the battery pack remaining life prediction. The specific output includes: Projected median remaining life estimate: , which represents the most likely time interval from the current moment to the end of the battery's life; Confidence intervals for life expectancy predictions: ; in, represents the 95% confidence interval of the predicted lifespan; Indicates the lower limit of the confidence interval; Indicates the upper limit of the confidence interval.

[0080] Prediction of the evolution trajectory of health factors: ; in, Represents the health factor state vector at time point t; Indicates the current time point; Represents the predicted end-of-life time; this trajectory describes the change in the battery's internal health status from the current moment to the predicted end-of-life time.

[0081] How forecast uncertainty changes over time: ; in, represents the forecast uncertainty measure at time point t; Indicates the current time point; Represents the predicted end-of-life time; this function describes how the uncertainty of the prediction results changes as the prediction time span increases.

[0082] Uncertainty typically includes epistemic uncertainty (uncertainty in the model itself) and data uncertainty (uncertainty caused by observation noise).

[0083] These results are presented in the form of numerical tables and graphical visualizations, making it easy to intuitively understand the health status and degradation trends of the battery.

[0084] The specific process of analyzing the key factors affecting battery pack life is as follows: Calculate the sensitivity of each external input to the evolution of the health factor: ; in, Indicates input parameters Health factors sensitivity; Indicates the health factors; Indicates the External input parameters; Represents the neural differential equation model describing the A function of the evolution of health factors; Represents the partial derivative of the rate of change of the health factor with respect to the input parameter.

[0085] Calculate the sensitivity of each health factor to the end of life: ; in, Indicates health factor sensitivity to end-of-battery life; represents the battery life end function predicted based on the health factor state; Indicates the end of life The partial derivative of the health factor.

[0086] Comprehensive analysis yields the overall impact of each external input on lifespan: ; in, Represents external input parameters Overall impact on battery life; Indicates the total number of health factors; It represents the sum of the effects of all health factors. This formula uses the chain rule to multiply the effects of external inputs on health factors by the effects of health factors on lifespan and then add them together to obtain the comprehensive impact of external inputs on lifespan.

[0087] Based on impact analysis, identify the top 3-5 factors that have the greatest impact on lifespan and generate corresponding usage optimization recommendations.

[0088] The specific output includes: Remaining life prediction report: including predicted life value, confidence interval, health status assessment, etc. Ranking of key influencing factors: List the factors that have the greatest impact on battery life and their degree of influence; Usage optimization suggestions: Based on the analysis of influencing factors, provide specific operation suggestions for extending battery life; Warning information: When the predicted lifespan is lower than the set threshold, a warning prompt will be issued.

[0089] Application examples of this implementation: This implementation has been verified and deployed in multiple battery application scenarios. The following uses two typical scenarios as examples to explain its implementation process and effects in detail.

[0090] Scenario 1: Electric vehicle battery management system; A new energy vehicle manufacturer has equipped its electric vehicle fleet with a battery pack life prediction system based on this embodiment. The system monitors and analyzes the health and lifespan of three different types of ternary lithium-ion battery packs (Type A, Type B, and Type C) in the fleet, each with different capacity and usage characteristics. The vehicles operate in diverse geographic locations (cold northern regions, warm southern regions, and temperate central regions) and with varying usage patterns (urban commuting, long-distance travel, and mixed use), providing an ideal environment for testing the system's adaptability under diverse conditions.

[0091] Scenario 2: Battery management of large energy storage power stations; A provincial power grid company implemented the prediction method described in this implementation plan at its newly built 100MWh energy storage power station, monitoring and predicting the health and lifespan of a large-scale battery array consisting of 5,000 modules. This energy storage station is primarily used for peak load regulation and renewable energy integration. The batteries operate in complex and variable conditions, including deep charge and discharge, frequent partial charge and discharge, and long-term floating charge. This poses significant challenges to the accuracy and adaptability of the prediction system.

[0092] Implementation process example: Electric vehicle battery management system realizes: In the electric vehicle application scenario, the specific implementation process of this embodiment is as follows: Data Collection and Preprocessing: The onboard battery management system collects battery monitoring data, including voltage, current, and temperature, every 10 seconds and uploads the data to the cloud platform every 10 minutes. The data preprocessing module performs quality checks, noise reduction, and standardization on the raw data to generate a standardized data matrix.

[0093] Electrochemical impedance spectroscopy reconstruction and health factor extraction: The system selects typical charge and discharge cycles daily and applies an impedance spectrum reconstruction algorithm to generate impedance spectra for different battery pack models. Table 1 shows the electrochemical impedance spectroscopy reconstruction accuracy and key health factors.

[0094] Table 1: Electrochemical impedance spectroscopy reconstruction accuracy and main health factors

[0095] Dynamic Evolution Modeling and Prediction: The system builds and trains customized neural differential equation models based on the characteristics of each vehicle's battery pack. Different electrochemical mechanism constraints are applied to vehicles in cold northern regions, warm southern regions, and temperate central regions to adapt to battery degradation characteristics under different environmental conditions.

[0096] Uncertainty-aware active learning: The system uses a Bayesian neural network approach to assess prediction uncertainty and dynamically adjusts the electrochemical constraint strength based on the uncertainty level. When the battery experiences abnormal operating conditions (such as extreme temperatures or rapid charging), the constraint strength is automatically adjusted to balance the model's physical plausibility and adaptability.

[0097] Battery management of large energy storage power stations: In the application scenario of a large energy storage power station, the specific implementation process of this embodiment is as follows: Tiered data collection strategy: Considering the massive amount of data from 5,000 battery modules, the system adopts a tiered sampling strategy: 20% of the battery modules receive high-frequency sampling (once per minute), 60% receive regular sampling (once every 10 minutes), and the remaining 20% ​​receive low-frequency sampling (once per hour). The data collection frequency is dynamically adjusted based on the battery's health and operating conditions.

[0098] Clustered Impedance Spectrum Reconstruction: Battery modules within the power station are divided into 12 clusters based on their characteristics and operating conditions. Representative samples are selected from each cluster for impedance spectrum reconstruction, and the reconstructed results are then applied to the entire cluster, significantly reducing computational complexity. During data processing, the system identifies nine primary modes of battery degradation, including high-temperature degradation, over-discharge, and rapid charging damage.

[0099] Multi-scale dynamic health factor monitoring: The system analyzes extracted health factors across three timescales: short-term (hours to days), medium-term (weeks to months), and long-term (quarters to years). This multi-scale health factor analysis successfully identifies changes in battery pack degradation rates across different seasons and load conditions.

[0100] Joint Optimization Strategy Generation: Based on the prediction results, the system generates an optimized charging and discharging strategy for the battery pack and works in conjunction with the grid dispatch system. Table 2 shows a comparison of battery pack efficiency before and after optimization.

[0101] Table 2: Comparison of optimized use of battery packs in energy storage power stations;

[0102] Technical effect verification: The key technical effects of this implementation have been verified through actual deployment and long-term operation in the two application scenarios mentioned above: Effect of improving prediction accuracy: The battery life prediction method of this embodiment significantly improves prediction accuracy compared to traditional methods. Table 3 shows the prediction accuracy comparison results under different scenarios and battery types.

[0103] Table 3: Comparison of the accuracy of different methods in remaining life prediction (average relative error in prediction %);

[0104] It is particularly noteworthy that in the early stages of the battery (SOH>90%) and under abnormal operating conditions (such as extreme temperatures and fast charging), the prediction advantage of this embodiment is more significant, with the average prediction error reduced by more than 60%.

[0105] Adaptability improvement effect: The uncertainty-aware active learning mechanism of this embodiment significantly improves the system's adaptability to new battery types and changing operating conditions. Table 4 shows the system's adaptability to unseen operating conditions.

[0106] Table 4: Comparison of the adaptability of different methods to new working conditions;

[0107] The percentages in the table represent the prediction accuracy under the new operating conditions. The results show that this implementation can quickly adapt and provide accurate predictions with little or no new operating condition data, a feature that is of great value in practical applications.

[0108] Comparison of battery pack life prediction performance: Table 5 compares the comprehensive performance of this embodiment with other current mainstream battery life prediction methods in terms of specific indicators.

[0109] Table 5: Comprehensive performance comparison of battery pack life prediction methods;

[0110] Generalization of prediction performance across different battery types: This embodiment has been verified on a variety of battery chemistry systems, and Table 6 shows its generalization capability.

[0111] Table 6: Comparison of prediction accuracy for different battery chemistries (relative error %)

[0112] Table 7: Correspondence between health factor characterization effects and degradation mechanisms;

[0113] Through the above application examples and effect verification, it can be seen that this implementation method has achieved significant technological progress in two key aspects: prediction accuracy and adaptability, fully demonstrating its technological innovation value and practicality in the field of battery pack life prediction.

[0114] At least one embodiment of the present invention discloses a device for predicting battery life, the device comprising: An acquisition unit is used to acquire battery pack monitoring data and perform preprocessing to obtain a standardized monitoring data matrix; wherein the standardized monitoring data matrix includes a standard voltage data sequence and a standard current data sequence; A generating unit, configured to generate a reconstructed electrochemical impedance spectrum data matrix based on a standard voltage data sequence, a standard current data sequence, and a preset impedance spectrum reconstruction algorithm; wherein the reconstructed electrochemical impedance spectrum matrix is ​​obtained by reconstructing and weighted fusion according to each preset stage of the battery pack charge and discharge process; An extraction unit is configured to extract a plurality of health factors from a standardized monitoring data matrix and a reconstructed electrochemical impedance spectroscopy data matrix based on a nonlinear manifold learning method; wherein each health factor has a corresponding physicochemical degradation mechanism; and each health factor corresponds to at least one key battery pack aging parameter; The prediction unit is used to obtain the predicted end-of-life time point of the battery pack based on the target differential equation model; the target differential equation model is obtained by incorporating constraints, scaling, adjusting the constraint strength, and controlling the balance between known and unknown areas into the initial neural differential equation model; the initial neural differential equation is used to describe the evolution of health factors over time.

[0115] In an exemplary embodiment of the present invention, an electronic device capable of implementing the above method is also provided.

[0116] Those skilled in the art will appreciate that various aspects of the present invention may be implemented as systems, methods, or program products. Therefore, various aspects of the present invention may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, microcode, etc.), or in a combination of hardware and software, collectively referred to herein as "circuits," "modules," or "systems."

[0117] The electronic device according to this embodiment of the present invention is merely an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0118] The electronic device is implemented as a general-purpose computing device. Components of the electronic device may include, but are not limited to, the at least one processor, the at least one memory, and a bus connecting different system components (including the memory and the processor).

[0119] The storage stores program codes, which can be executed by the processor, so that the processor executes the steps according to various exemplary embodiments of the present invention described in the above “Exemplary Method” section of this specification.

[0120] The memory may include readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory, and may further include read only memory (ROM).

[0121] The storage may also include a program / utility having a set (at least one) of program modules, such program modules including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0122] The bus may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures.

[0123] The electronic device may also communicate with one or more external devices (e.g., a keyboard, pointing device, Bluetooth device, etc.), one or more devices that enable a user to interact with the electronic device, and / or any device that enables the electronic device to communicate with one or more other computing devices (e.g., a router, modem, etc.). This communication may occur via an input / output (I / O) interface. Furthermore, the electronic device may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter. As shown in the figure, the network adapter communicates with other modules of the electronic device via a bus. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0124] From the above description of the embodiments, it will be readily apparent to those skilled in the art that the exemplary embodiments described herein can be implemented via software or via a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present invention can be embodied in the form of a software product, which can be stored on a non-volatile storage medium (such as a CD-ROM, USB flash drive, or mobile hard drive) or on a network and includes instructions for causing a computing device (such as a personal computer, server, terminal device, or network device) to execute the methods according to the embodiments of the present invention.

[0125] In an exemplary embodiment of the present invention, a computer-readable storage medium is further provided, on which is stored a program product capable of implementing the methods described above in this specification. In some possible implementations, various aspects of the present invention may also be implemented in the form of a program product, which includes program code. When the program product is executed on a terminal device, the program code is used to cause the terminal device to perform the steps according to various exemplary embodiments of the present invention described in the "Exemplary Methods" section above.

[0126] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0127] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0128] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0129] Program code for performing the operations of the present invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as C or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0130] Furthermore, the above-described figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes illustrated in the above-described figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0131] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0132] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for predicting battery life, characterized in that: The method comprises: Collecting battery pack monitoring data and preprocessing it to obtain a standardized monitoring data matrix; wherein the standardized monitoring data matrix includes a standard voltage data sequence and a standard current data sequence; Generate a reconstructed electrochemical impedance spectrum data matrix based on the standard voltage data sequence, the standard current data sequence, and a preset impedance spectrum reconstruction algorithm; wherein the reconstructed electrochemical impedance spectrum matrix is ​​reconstructed and weightedly fused according to each preset stage of the battery pack charge and discharge process; Using a nonlinear manifold learning approach, several health factors are extracted from a standardized monitoring data matrix and a reconstructed electrochemical impedance spectroscopy data matrix; each health factor has a corresponding physicochemical degradation mechanism; and each health factor corresponds to at least one key battery pack aging parameter. According to the target differential equation model, the predicted end-of-life time point of the battery pack is obtained; wherein, the target differential equation model is obtained by incorporating constraints, expanding the scale, adjusting the constraint strength, and controlling the balance between known and unknown areas into the initial neural differential equation model; the initial neural differential equation is used to describe the evolution dynamics of the health factor over time.

2. The method for predicting battery life according to claim 1, wherein: Incorporate constraints into the initial neural differential equation model, scale it up, adjust the constraint strength, and control the balance between known and unknown regions, including: Incorporate battery electrochemical mechanism constraints into the initial neural differential equation model; The initial neural differential equation model is scaled up according to the health factor set obtained by classification according to the time scale to obtain the key neural differential equation model; wherein the key neural differential equation is a multi-time scale coupled dynamic equation model; Predicting uncertainty values ​​of key neural differential equation models at different states; Adjust the strength of the constraints on the battery electrochemical mechanism in the key neural differential equation model based on the uncertainty value; According to the preset balance strategy, the balance of the key neural differential equation model in the known area and the unknown area is controlled.

3. The method for predicting battery life according to claim 1, wherein: Obtaining the predicted end-of-life time point of the battery pack according to the target differential equation model includes: According to the target neural differential equation model, the future evolution trajectory of each health factor is predicted; Based on the future evolution trajectory of each health factor, the predicted end-of-life time of the battery pack is determined.

4. The method for predicting battery life according to claim 2, wherein: The initial neural differential equation model is scaled up according to the health factor set obtained by classification according to the time scale to obtain a key neural differential equation model, including: Several health factors are classified according to time scales to obtain several health factor sets; wherein each health factor set has a corresponding time scale; the time scales include a first change scale, a second change scale, and a third change scale; wherein the change speed corresponding to the first change scale is greater than the change speed corresponding to the second change scale; the change speed corresponding to the second change scale is greater than the change speed corresponding to the third change scale; each time scale has a corresponding step size; the step size corresponding to the first change scale is smaller than the step size corresponding to the second change scale; the step size corresponding to the second change scale is smaller than the step size corresponding to the third change scale; The corresponding health factor time scale change rate is solved according to the step size corresponding to the time scale of each health factor to obtain the key neural differential equation model.

5. The method for predicting battery life according to claim 2, wherein: The uncertainty values ​​of the predicted key neural differential equation model in different states include: Quantify the uncertainty of each health factor included in the key neural differential equation model based on Bayesian neural network; Obtain the uncertainty value of each health factor under different conditions; The uncertainty values ​​of the key neural differential equation model in different states are obtained by weighted summing up the uncertainty values ​​of each health factor in different states and the preset uncertainty weights corresponding to each health factor.

6. The method for predicting battery life according to claim 2, characterized in that: After obtaining the predicted end-of-life time point of the battery pack according to the target differential equation model, the method further includes: Analyze the key factors affecting battery pack life and generate optimization strategies.

7. The method for predicting battery life according to claim 3, characterized in that: Based on the future evolution trajectory of each health factor, the predicted end-of-life time of the battery pack is determined, including: Determine the initial end-of-life time of the battery pack based on the future evolution trajectory of each health factor; The initial end-of-life time point is calibrated according to the historical prediction deviation and the uncertainty of each health factor to obtain the predicted end-of-life time point.

8. A device for predicting battery life, characterized in that: The device comprises: An acquisition unit is used to acquire battery pack monitoring data and perform preprocessing to obtain a standardized monitoring data matrix; wherein the standardized monitoring data matrix includes a standard voltage data sequence and a standard current data sequence; A generating unit, configured to generate a reconstructed electrochemical impedance spectrum data matrix based on a standard voltage data sequence, a standard current data sequence, and a preset impedance spectrum reconstruction algorithm; wherein the reconstructed electrochemical impedance spectrum matrix is ​​obtained by reconstructing and weighted fusion according to each preset stage of the battery pack charge and discharge process; An extraction unit is configured to extract a plurality of health factors from a standardized monitoring data matrix and a reconstructed electrochemical impedance spectroscopy data matrix based on a nonlinear manifold learning method; wherein each health factor has a corresponding physicochemical degradation mechanism; and each health factor corresponds to at least one key battery pack aging parameter; The prediction unit is used to obtain the predicted end-of-life time point of the battery pack based on the target differential equation model; the target differential equation model is obtained by incorporating constraints, scaling, adjusting the constraint strength, and controlling the balance between known and unknown areas into the initial neural differential equation model; the initial neural differential equation is used to describe the evolution of health factors over time.

9. A non-transitory computer-readable storage medium, characterized in that The storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the method according to any one of claims 1 to 7.

Images