Real-time prediction method of remaining life of lithium batteries based on dynamic uncertainty modeling

Abstract

The present invention belongs to the field of lithium battery remaining life prediction, and discloses a real-time prediction method for the remaining life of a lithium battery based on dynamic uncertainty modeling. It uses multiple sensors to collaboratively collect data on the entire life cycle of the lithium battery, and combines a convolutional neural network to construct a dynamic health score, effectively capturing the non-stationary and nonlinear characteristics of the data. A non-stationary random process is further used to quantify the uncertainty of the health score, and the relevant parameters are synchronously adjusted through a joint optimization objective function to enhance the adaptability of the model to complex degradation patterns. Based on the Bayesian reasoning framework, the non-stationary random process parameters are dynamically updated, and the conjugate prior distribution of historical data and real-time observations is combined to achieve parameter adaptive adjustment and dynamic prediction of the remaining life of the lithium battery. This method not only makes up for the lack of physical interpretability of the traditional data-driven method, but also overcomes the problem of insufficient accuracy of the single model-driven method due to the complex battery degradation mechanism.

Description

Technical Field

[0001] The present invention belongs to the field of lithium battery remaining life prediction, and in particular relates to a real-time prediction method for lithium battery remaining life based on dynamic uncertainty modeling. Background Art

[0002] Lithium battery remaining life prediction, a key technology for ensuring the safe operation of new energy vehicle powertrains and grid-scale energy storage devices, relies on establishing precise mathematical models of multi-dimensional degradation pathways, including the phase transition kinetics of electrode active materials, electrolyte oxidation and decomposition mechanisms, and solid-liquid interface side reactions. By quantifying the interconnectedness of these degradation mechanisms, dynamic early warning of critical failure points, such as sudden capacity drop and internal resistance mutation, can be achieved, providing a scientific basis for system health management. Under dynamic operating conditions, the battery's internal electrochemical polarization and concentration polarization effects form a complex nonlinear coupling with external thermal, mechanical, and electrical multi-physics fields. This cross-scale interaction not only accelerates microscopic degradation processes such as SEI (Solid Electrolyte Interphase) film reconstruction and lithium dendrite growth, but also leads to significant non-stationary characteristics in the time-varying degradation behavior at the material level. In this complex environment, early aging signals are often obscured by strong noise interference, making it difficult for traditional threshold-based linear prediction methods to effectively identify underlying degradation trends.

[0003] Existing lithium battery remaining life prediction methodologies have evolved into three paradigms: physical mechanism models, data-driven models, and hybrid intelligent architectures, driven by the multi-dimensional requirements of mechanism explainability, adaptability to operating conditions, and engineering practicality. Data-driven methods, leveraging machine learning and deep learning algorithms to extract underlying aging patterns from massive amounts of historical data, have become the mainstream research direction in this field. Their technological evolution has primarily evolved along three dimensions: deep learning architectures, kernel methodologies, and hybrid modeling strategies. Deep learning employs recurrent neural networks such as LSTM (Long Short-Term Memory) networks and temporal convolutional networks to capture the temporal dependencies of capacity decay and construct an end-to-end multi-step rolling prediction framework. Kernel methodologies leverage the kernel technique of support vector regression and the probabilistic framework of Gaussian process regression to characterize decay trajectories through nonlinear mapping and quantitatively assess prediction uncertainty. Hybrid modeling strategies fuse relevance vector machines with three-parameter degradation models to construct multi-scale prediction frameworks, or combine ARIMA (Auto Regressive Integrated Moving Average) time series analysis with hidden Markov models to address non-stationary issues such as capacity regeneration. Compared with physical models that require precise descriptions of electrochemical reactions, data-driven methods effectively circumvent the difficulties of complex mechanism modeling. However, their predictive performance is highly dependent on the completeness of feature engineering, the scale and quality of training data, and there are still inherent limitations in model interpretability and uncertainty quantification.

[0004] Furthermore, advances in sensor and computing technologies have provided data-driven approaches with a vast amount of labeled degradation data, driving the application of deep neural networks. While these approaches can circumvent complex electrochemical modeling through autonomous learning and accurately predict capacity inflection points in specific scenarios, their "black box" nature renders the decision-making basis opaque. In practical applications, differences in battery models and operating conditions make it difficult to migrate model parameters across scenarios, while full lifecycle labeling data requires months of experimental collection, a high cost that severely restricts industrial-scale application. Summary of the Invention

[0005] The purpose of the present invention is to provide a real-time prediction method for the remaining life of lithium batteries based on dynamic uncertainty modeling, which combines the advantages of data-driven in multi-source information processing with the ability of probabilistic models in uncertainty quantification. It not only makes up for the lack of physical interpretability of traditional data-driven methods, but also overcomes the problem of insufficient accuracy of single model-driven methods due to the complex battery degradation mechanism.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] A method for real-time prediction of the remaining life of a lithium battery based on dynamic uncertainty modeling, comprising:

[0008] Obtain the full life cycle data of lithium batteries collected by multiple sensors, and combine it with discrete Fourier transform to extract the degradation feature vector corresponding to each charge and discharge cycle in the full life cycle of the lithium battery;

[0009] Through convolutional neural network, the dynamic health score of lithium battery is output according to the degradation feature vector;

[0010] Based on the dynamic health score, a linear Wiener process model is used to construct a degradation process model of the remaining life of the lithium battery, and the remaining life prediction value of the lithium battery is calculated based on the obtained degradation process model;

[0011] Calculate the loss function that includes the remaining life prediction error term and the maximum likelihood estimation constraint term of the linear Wiener process, and update the parameters of the convolutional neural network and degradation process model according to the loss function until the training is completed;

[0012] The real-time dynamic health score of the target lithium battery is obtained through the updated convolutional neural network, and the drift coefficient of the degradation process model is updated in real time using Bayesian theory. The latest degradation process model is used to output the real-time remaining life prediction value of the target lithium battery.

[0013] Several optional methods are also provided below, but they are not intended to be additional limitations on the above-mentioned overall solution. They are merely further supplements or optimizations. Under the premise that there are no technical or logical contradictions, each optional method can be combined separately for the above-mentioned overall solution, or multiple optional methods can be combined.

[0014] Preferably, the extraction of the degradation feature vector corresponding to each charge and discharge cycle in the entire life cycle of the lithium battery in combination with discrete Fourier transform includes:

[0015] For each charge and discharge cycle, the charging current sequence is discrete Fourier transformed to calculate the corresponding total spectrum energy; the discharge voltage sequence is discrete Fourier transformed to calculate the corresponding total spectrum energy;

[0016] And calculate the average resistance of the charge and discharge cycle, the time of the constant voltage charging phase, the time of the constant current discharging phase, the average slope of the current decaying from A to B during the constant voltage charging phase, and the average slope of the voltage decaying from C to D during the constant current discharging phase, where A is greater than B and C is greater than D;

[0017] The total spectral energy corresponding to the charging current sequence, the total spectral energy corresponding to the discharging voltage sequence, the average resistance of the charge and discharge cycle, the time of the constant voltage charging stage, the time of the constant current discharging stage, the average slope of the current decaying from A to B in the constant voltage charging stage, and the average slope of the voltage decaying from C to D in the constant current discharging stage are normalized respectively, and all normalized features are combined as the degenerate feature vector.

[0018] Preferably, the linear Wiener process model is used to construct a degradation process model of the remaining life of the lithium battery based on the dynamic health score, including:

[0019] A linear Wiener process is used to describe the degradation path of the dynamic health score of lithium batteries throughout their life cycle, where the drift coefficient follows a normal distribution and the increment of the dynamic health score follows a normal distribution.

[0020] The first failure time is defined based on the degradation path of the dynamic health score that conforms to the linear Wiener process;

[0021] Construct the probability density function of the first failure time;

[0022] The integral process of the probability density function of the first failure time is used as the remaining life prediction function to complete the construction of the degradation process model of the remaining life of the lithium battery.

[0023] Preferably, the remaining life prediction error term includes:

[0024] The mean square error is calculated based on the predicted value of the remaining life corresponding to each charge and discharge cycle in the entire life cycle of the lithium battery and the actual value of the remaining life, and the mean square error is used as the remaining life prediction error term.

[0025] Preferably, the maximum likelihood estimation constraint term of the linear Wiener process includes:

[0026] Calculate the dynamic health score increment between two adjacent charge and discharge cycles in the entire life cycle of the lithium battery;

[0027] Construct the log-joint likelihood function of all dynamic health score increments;

[0028] The logarithmic joint likelihood function after removing the constant is used as the maximum likelihood estimation constraint term of the linear Wiener process.

[0029] Preferably, the constructing of the logarithmic joint likelihood function of all dynamic health score increments includes:

[0030]

[0031] in, represents the log-joint likelihood function, is the total number of charge and discharge cycles in the entire life cycle of the lithium battery, For the The dynamic health score increment corresponding to the charge and discharge cycle, is the mean value of the drift coefficient, is the time increment between two adjacent charge and discharge cycles, is the diffusion coefficient.

[0032] Preferably, the loss function is a weighted sum of a remaining life prediction error term and a maximum likelihood estimation constraint term of a linear Wiener process;

[0033] The gradient descent method is used to minimize the loss function to update the parameters of the convolutional neural network and the degradation process model. After the training is completed, the optimal parameter set is output. The optimal parameter set includes the optimal weight parameter of the convolutional neural network, the optimal drift coefficient of the degradation process model, the optimal diffusion coefficient of the degradation process model, and the optimal failure threshold of the degradation process model. The drift coefficient obeys the normal distribution, that is, the optimal mean value of the drift coefficient and the optimal standard deviation of the drift coefficient are obtained.

[0034] Preferably, obtaining the real-time dynamic health score of the target lithium battery through the updated convolutional neural network includes:

[0035] Acquire multi-sensor data of the target lithium battery during its current charge and discharge cycle;

[0036] Combined with discrete Fourier transform, the degradation feature vector of multi-sensor data is extracted;

[0037] Through the updated convolutional neural network, the real-time dynamic health score of the target lithium battery is output according to the feature vector.

[0038] Preferably, the method of updating the drift coefficient of the degradation process model in real time using Bayesian theory includes:

[0039] The drift coefficient of the degradation process model is updated in real time using Bayesian theory. The updated drift coefficient obeys the posterior distribution. The posterior distribution is calculated based on the optimal parameters of the degradation process model obtained after training. The calculation formula is as follows:

[0040]

[0041] in, is the standard deviation of the drift coefficient after Bayesian theory update, is the mean of the drift coefficient after Bayesian theory update, is the optimal diffusion coefficient of the degradation process model, is the standard deviation of the optimal drift coefficient of the degradation process model, is the current charge and discharge cycle number of the target lithium battery, is the time increment between two adjacent charge and discharge cycles, is the mean value of the optimal drift coefficient of the degradation process model, Target lithium battery The dynamic health score increment corresponding to the charge and discharge cycle.

[0042] The present invention provides a real-time prediction method for the remaining life of lithium batteries based on dynamic uncertainty modeling. By designing a global objective function of multi-dimensional feature fusion and uncertainty collaborative optimization, it realizes the efficient joint optimization of data-driven convolutional neural network deep health score extraction and model-driven linear Wiener process degradation modeling. Different from the traditional life prediction method that relies on a single sensor signal or a fixed parameter model, the method of the present invention starts from the deep mining of full life cycle data, uses convolutional neural networks to adaptively fuse battery degradation characteristics, and combines the Bayesian framework to dynamically update the parameters of non-stationary random processes to accurately characterize the randomness and time-varying characteristics of battery performance degradation. The method of the present invention can realize real-time dynamic prediction of remaining life, and has the advantages of high reliability, strong interpretability and low deployment cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of a method for real-time prediction of the remaining life of a lithium battery based on dynamic uncertainty modeling according to the present invention;

[0044] Figure 2 This is a parameter variation diagram of the linear Wiener process of the present invention;

[0045] Figure 3 The dynamic health score and remaining life degradation trajectory diagram of the convolutional neural network and degradation process model trained based on the CS2_35 lithium battery dataset of the present invention;

[0046] Figure 4 The dynamic health score and remaining life degradation trajectory diagram of the convolutional neural network and degradation process model tested based on the CS2_35 lithium battery dataset of the present invention;

[0047] Figure 5 This is a diagram showing the application of the real-time prediction method for the remaining life of a lithium battery based on dynamic uncertainty modeling in the present invention on the CS2_36 lithium battery dataset. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0050] like Figure 1 As shown, this embodiment proposes a real-time prediction method for the remaining life of a lithium battery based on dynamic uncertainty modeling, comprising the following steps:

[0051] Step 1: Obtain the full life cycle data of the lithium battery collected by multiple sensors, and combine it with discrete Fourier transform to extract the degradation feature vector corresponding to each charge and discharge cycle in the full life cycle of the lithium battery.

[0052] Step 1.1: Use multiple sensors to collect data on the entire life cycle of the lithium battery, output the current data of the constant voltage charging stage and the voltage data of the constant current discharge stage in each charge and discharge cycle, and perform discrete Fourier transform on each of them to calculate the corresponding total spectrum energy and other characteristics.

[0053] This embodiment uses a voltage sensor, a current sensor, and an internal resistance sensor to collect voltage, current, and internal resistance data over the entire life cycle of the lithium battery. The charging process uses a constant voltage mode, where the charging current decays nonlinearly as the battery capacity increases, terminating when it reaches the current threshold. The discharging process uses a constant current mode, where the terminal voltage decreases monotonically as the battery capacity decreases, terminating when it reaches the voltage threshold, thus completing the charge and discharge cycle.

[0054] Taking the current sensor as an example, the charging current sequence of a charge and discharge cycle is defined as , represents the sampling time, Indicates the number of charge and discharge cycles. This embodiment collects the charging current sequence of a charge and discharge cycle process. Perform discrete Fourier transform and calculate the total spectrum energy of the charging process.

[0055] After discrete Fourier transform, the output frequency sequence Expressed as:

[0056]

[0057] in, represents a discrete frequency index, Indicates the sequence length, that is, the total sampling time, is the imaginary part.

[0058] The frequency sequence of the output Calculate the total spectral energy:

[0059]

[0060] in, Indicates in Charging current sequence in the second charge and discharge cycle The total spectral energy, represents the L1 norm.

[0061] Then we get the The total spectrum energy of the charging current sequence in the first charge and discharge cycle can also be obtained by the above steps. The total spectral energy of the discharge voltage sequence in a charge and discharge cycle.

[0062] The total spectral energy of the charging current sequence and the total spectral energy of the discharging voltage sequence are used as part of the features. Other features include the average resistance of the charge and discharge cycle, the time of the constant voltage charging phase, the time of the constant current discharging phase, the average slope of the current decaying from A to B in the constant voltage charging phase, and the average slope of the voltage decaying from C to D in the constant current discharging phase, where A is greater than B and C is greater than D. The calculation formulas for the charge and discharge cycles are:

[0063]

[0064]

[0065]

[0066]

[0067]

[0068] in, Indicates the The average resistance of the charge and discharge cycle, Indicates the current sequence (the sequence collected during the entire charging and discharging process) Sample internal resistance value, Indicates the total number of sampling points in the current sequence, express The time of the constant voltage charging phase of the charge and discharge cycle, Indicates the time of constant voltage charging stage, express The time of the constant current discharge phase of the charge and discharge cycle, Indicates the time of constant current discharge stage, express The average slope of the current decaying from A (e.g. 0.6A) to B (e.g. 0.2A) during the constant voltage charging phase of the charge and discharge cycle, The inverse function of the linear interpolation function of the current sequence in the constant voltage charging stage is represented by: Indicates The average slope of the voltage decay from C to D during the constant current discharge phase of the charge and discharge cycle, The inverse function of the linear interpolation function of the voltage series during the constant current discharge phase.

[0069] Step 1.2: Normalize the total spectrum energy and other features to obtain normalized features.

[0070] Total spectrum energy and other statistical characteristics, for any of them, are expressed as Take the expression as an example, and transform it to the same range through normalization processing to reduce the data dimension difference and noise interference, that is:

[0071]

[0072] in, Indicates the Features, Indicates the charge and discharge cycle data. and Corresponding to The maximum and minimum values of is the normalized feature obtained after normalization, and The vector composed of all normalized features in the charge and discharge cycle is used to integrate the degraded feature vector of time-frequency domain characteristics. To express.

[0073] Step 2: Use a convolutional neural network to output the dynamic health score of the lithium battery based on the feature vector. Compared with traditional time-domain features that can only statically describe the degradation process of lithium batteries through statistics such as mean and variance, the dynamic health score overcomes the limitations of fixed statistical dimensions in characterizing complex degradation patterns by introducing nonlinear dynamic modeling methods.

[0074] This embodiment constructs a dynamic health score through a convolutional neural network , the formula is as follows:

[0075]

[0076] in, represents a convolutional neural network, Represents the weights of the convolutional neural network.

[0077] Step 3: Based on the dynamic health score, a linear Wiener process model is used to construct a degradation process model of the remaining life of the lithium battery, and the remaining life prediction value of the lithium battery is calculated based on the obtained degradation process model.

[0078] During initial iterative training, the dynamic health score output by the convolutional neural network is a non-stationary random process. To balance model accuracy and computational efficiency, this embodiment simplifies the complex non-stationary degradation process into a linear Wiener Process Model (LWPM) process through parameter constraints. Specifically, a maximum likelihood estimation constraint term for the linear Wiener process is added to the calculation of the loss function. Through iterative training, the transformation from a non-stationary random process to a linear Wiener process is achieved, thereby constructing a degradation process model for the remaining life of a lithium battery. The goal of this embodiment is to converge towards a linear Wiener process. Therefore, the degradation process model is constructed based on the assumption that the dynamic health score conforms to a linear Wiener process. During iterative training, the actual dynamic health score is driven to continuously converge towards the linear Wiener process. After the iterative training is completed, a degradation process model that conforms to the linear Wiener process is ultimately obtained.

[0079] Step 3.1: Use the linear Wiener process to describe the degradation path of the dynamic health score of the lithium battery throughout its life cycle, including:

[0080]

[0081] in, Indicates the current Dynamic health score of charge and discharge cycles, represents the initial dynamic health score, Represents the drift coefficient and satisfies the normal distribution , characterizing individual differences, is the mean value of the drift coefficient, is the standard deviation of the drift coefficient, represents the diffusion coefficient, is a standard Brownian motion. The increment of the dynamic health score is independent and identically distributed, which obeys the normal distribution, that is:

[0082]

[0083] in, represents the increment of the dynamic health score, and the increment is a random variable. A possible value representing the increment of the dynamic health score, represents the time increment between two adjacent charge and discharge cycles, is the probability density function. According to the degradation path of the dynamic health score that conforms to the linear Wiener process, the first failure time is defined as:

[0084]

[0085] in, Indicates the The remaining life value of the lithium battery after the first charge and discharge cycle is a random variable. is the supremum, Indicates the A possible value of the remaining life of a lithium battery after 10 charge and discharge cycles, Indicates from Charge and discharge cycles to Degradation path of the charge-discharge cycle, represents the failure threshold, It represents the dynamic health score of the detection from the beginning to the current charge and discharge cycle. The degradation path is a Wiener process, and the probability density function of the first failure time is expressed as:

[0086]

[0087] choose The mathematical expectation of The point estimate gives the remaining life prediction function as follows:

[0088]

[0089] in, Indicates the The predicted value of the remaining life of a lithium battery after 10 charge and discharge cycles.

[0090] Step 3.2, by estimating The standard deviation is measured The uncertainty of , is calculated using the following formula:

[0091]

[0092] in, Indicates the increment of the dynamic health score between two adjacent charge and discharge cycles, Indicates the current Dynamic health score of charge and discharge cycles, It is the total number of charge and discharge cycles in the entire life cycle of the lithium battery.

[0093] This step uses the linear Wiener process model to construct a degradation process model for the dynamic health score of the lithium battery throughout its life cycle, preparing for the subsequent construction of the optimization objective function. After obtaining the dynamic health score of the current charge and discharge cycle, it is substituted into the degradation process model to obtain the remaining life prediction value of the lithium battery.

[0094] Step 4: Calculate the loss function including the remaining life prediction error term and the maximum likelihood estimation constraint term of the linear Wiener process, and update the parameters of the convolutional neural network and the degradation process model according to the loss function until the training is completed.

[0095] This embodiment minimizes the loss function and iteratively adjusts the parameter weights in the dynamic convolutional neural network and the drift coefficient, diffusion coefficient, and failure threshold in the linear Wiener process to balance data-driven features with physical statistical characteristics and avoid overfitting or underfitting caused by single optimization. Specifically, the following steps are performed:

[0096] Step 4.1: Calculate the mean square error between the predicted remaining lifespan and the actual remaining lifespan for each charge and discharge cycle in the entire life cycle of the lithium battery, and use the mean square error as the remaining lifespan prediction error term:

[0097]

[0098] in, The mean square error is approximately, Represents the true value of the remaining life of the lithium battery. The corresponding remaining life prediction value can be obtained by combining the dynamic health score output by the convolutional neural network with step 3.1.

[0099] Step 4.2: Introduce the maximum likelihood estimation constraint of the linear Wiener process into the loss function, which is the logarithmic joint likelihood function of all dynamic health score increments, expressed as:

[0100]

[0101] in, is the log-joint likelihood function.

[0102] The logarithmic joint likelihood function after removing the constant is used as the maximum likelihood estimation constraint term of the linear Wiener process, which is expressed as:

[0103]

[0104] in, Represents the linear Wiener process constraint. The mean square error constraint and the maximum likelihood estimation constraint of the linear Wiener process are weighted and fused into the total loss function, that is:

[0105]

[0106] in is the total loss function, is the weight hyperparameter.

[0107] Step 4.3: Use the gradient descent method to minimize the loss function and solve for the optimal parameter set. , Represents the optimal parameter weights in the convolutional neural network, represents the optimal mean value of the drift coefficient, represents the standard deviation of the optimal drift coefficient, represents the optimal diffusion coefficient, represents the optimal failure threshold.

[0108] This example uses the gradient descent method to minimize the loss function and obtain the optimal parameter set ,Finally, the feedback between dynamic health scoring and modeling is realized, ,thus forming a closed-loop feedback mechanism between dynamic health scoring ,construction and linear Wiener process model.

[0109] Step 5: Obtain the real-time dynamic health score of the target lithium battery through the updated convolutional neural network, and use the Bayesian theory to update the drift coefficient of the degradation process model in real time, and use the latest degradation process model to output the remaining life prediction value of the target lithium battery.

[0110] This embodiment uses the obtained optimal parameter set to perform Bayesian real-time updates on the drift parameters corresponding to the target lithium battery dataset. Real-time prediction of the remaining life of the lithium battery is achieved by calculating the expected value of the first failure time, thereby achieving dynamic integration from offline joint optimization to online probabilistic reasoning. The steps include:

[0111] Step 5.1: For the target lithium battery sample, perform a discrete Fourier transform on the data obtained by multiple sensors during the current charge and discharge cycle of the target lithium battery. The discrete Fourier transform method is the same as before.

[0112] Step 5.2: Calculate the corresponding total spectrum energy and other features, perform feature normalization, and obtain the degraded feature vector.

[0113] Step 5.3: Use the optimal convolutional neural network parameter weights , outputs a real-time dynamic health score based on the degradation feature vector of the target lithium battery.

[0114] Step 5.4: Use Bayesian theory to update the drift parameter using the following formula:

[0115]

[0116] in, Indicates the target lithium battery dynamic health score from 0 to Observation data of charge and discharge cycles, Indicates If the event is known, The probability of an event occurring, Represents the prior of the drift parameter, integrating the observed prior distribution of the historical dynamic health score and the real-time dynamic health score. The updated drift parameter obeys the posterior distribution, and the posterior distribution of the drift parameter can be calculated , the formula is as follows:

[0117]

[0118] in, is the standard deviation of the drift coefficient after Bayesian theory update, is the mean of the drift coefficient after Bayesian theory update, is the diffusion coefficient in the optimal parameter set, is the standard deviation of the drift coefficient in the optimal parameter set, is the current charge and discharge cycle number of the target lithium battery, is the time increment between two adjacent charge and discharge cycles, is the drift coefficient in the optimal parameter set, For the target lithium battery from The dynamic health score increment corresponding to the charge and discharge cycle.

[0119] Step 5.5: Substitute the optimal diffusion coefficient and failure threshold obtained after training, as well as the drift coefficient after the Bayesian theory update, into the degradation process model to obtain the latest degradation process model; then calculate the dynamic health score of the target lithium battery in the current charge and discharge cycle. Substitute the latest degradation process model to output the real-time remaining life prediction value of the target lithium battery in the current charge and discharge cycle. The remaining life prediction function in the real-time prediction uses the following formula:

[0120]

[0121] in, Indicates the real-time remaining life prediction value of the target lithium battery in the current charge and discharge cycle.

[0122] This embodiment establishes a dynamic update mechanism for linear Wiener process parameters based on a Bayesian inference framework in real-time prediction, enabling online real-time prediction and uncertainty quantification of the remaining life of lithium batteries. It should be noted that steps 1-4 are the training process, which typically requires multiple iterations until the training end condition is reached. Step 5 is the application process, which is executed a corresponding number of times based on actual needs. The training process can be executed separately, the application process can be executed separately, or the training process and application process can be executed simultaneously.

[0123] The present invention uses multiple sensors to collaboratively collect data on the entire life cycle of lithium batteries, combines convolutional neural networks to construct dynamic health scores, and effectively captures the non-stationary and nonlinear characteristics of the data. Non-stationary random processes are further used to quantify the uncertainty of the health score, and related parameters are synchronously adjusted through joint optimization of the objective function to enhance the model's adaptability to complex degradation patterns. Based on the Bayesian reasoning framework, the parameters of the non-stationary random process are dynamically updated, and the conjugate prior distribution of historical data and real-time observations is combined to achieve adaptive parameter adjustment and dynamic prediction of the remaining life of the lithium battery. This method combines the advantages of data-driven in multi-source information processing with the ability of probabilistic models in uncertainty quantification. It not only makes up for the lack of physical interpretability of traditional data-driven methods, but also overcomes the problem of insufficient accuracy of single model-driven methods due to the complex battery degradation mechanism.

[0124] In a specific embodiment, this embodiment is verified by the University of Maryland lithium battery public data set, as follows:

[0125] The charging process was set to a constant voltage mode (CV, 4.2V). As the battery capacity increased, the charging current nonlinearly decayed to 0.2A, terminating the charge. The discharge process was set to a constant current mode (CC, 0.2A). As the battery capacity decreased, the terminal voltage monotonically decreased to 2.7V, terminating the discharge. This constituted a complete charge-discharge cycle. Seven charge-discharge cycle features were extracted from the full life cycle data, including the total spectral energy of the charge current sequence, the total spectral energy of the discharge voltage sequence, the average resistance of the charge-discharge cycle, the duration of the constant voltage charge phase, the duration of the constant current discharge phase, the average slope of the current decay from 0.6A to 0.2A during the constant voltage charge phase, and the average slope of the voltage decay from 3.8V to 2.8V during the constant current discharge phase. The CS2_35 lithium battery dataset was used to model the degradation process of the dynamic health score. The optimal convolutional neural network parameters and degradation process model parameters were obtained. Based on these parameters, Bayesian updating was used to achieve online prediction of remaining life for the CS2_36 lithium battery dataset. Figure 2 It can be seen that the linear Wiener process parameters show good convergence characteristics in the iterative process and eventually reach a stable state; Figure 3 and Figure 4 It can be seen from the above that the convolutional neural network constructed based on the CS2_35 lithium battery dataset can effectively characterize the battery degradation process, and the remaining life prediction results of the degradation process model are highly consistent with the actual degradation trajectory; Figure 5 As can be seen from the figure, the Bayesian online update strategy is effective when applied to the CS2_36 lithium battery dataset. Through the real-time parameter correction mechanism, the mean square error between the predicted results and the measured values is kept within the ideal range of 10.81 days. Overall, the proposed method achieves an online fitting accuracy of 99.48% for the remaining life of lithium batteries, demonstrating excellent remaining life prediction results.

[0126] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0127] The above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A real-time prediction method for the remaining life of a lithium battery based on dynamic uncertainty modeling, characterized in that: The real-time prediction method for the remaining life of a lithium battery based on dynamic uncertainty modeling includes: Obtain the full life cycle data of lithium batteries collected by multiple sensors, and combine it with discrete Fourier transform to extract the degradation feature vector corresponding to each charge and discharge cycle in the full life cycle of the lithium battery; Through convolutional neural network, the dynamic health score of lithium battery is output according to the degradation feature vector; Based on the dynamic health score, a linear Wiener process model is used to construct a degradation process model of the remaining life of the lithium battery, and the remaining life prediction value of the lithium battery is calculated based on the obtained degradation process model; Calculate the loss function that includes the remaining life prediction error term and the maximum likelihood estimation constraint term of the linear Wiener process, and update the parameters of the convolutional neural network and degradation process model according to the loss function until the training is completed; The updated convolutional neural network is used to obtain the real-time dynamic health score of the target lithium battery. The drift coefficient of the degradation process model is updated in real time using Bayesian theory, and the latest degradation process model is used to output the real-time remaining life prediction value of the target lithium battery. The method of constructing a degradation process model of the remaining life of a lithium battery based on a dynamic health score and using a linear Wiener process model includes: A linear Wiener process is used to describe the degradation path of the dynamic health score of lithium batteries throughout their life cycle, where the drift coefficient follows a normal distribution and the increment of the dynamic health score follows a normal distribution. The first failure time is defined based on the degradation path of the dynamic health score that conforms to the linear Wiener process; Construct the probability density function of the first failure time; The integral process of the probability density function of the first failure time is used as the remaining life prediction function to complete the construction of the degradation process model of the remaining life of the lithium battery; The maximum likelihood estimation constraint term of the linear Wiener process includes: Calculate the dynamic health score increment between two adjacent charge and discharge cycles in the entire life cycle of the lithium battery; Construct the log-joint likelihood function of all dynamic health score increments; The logarithmic joint likelihood function after removing the constant is used as the maximum likelihood estimation constraint term of the linear Wiener process; The loss function is a weighted sum of the remaining life prediction error term and the maximum likelihood estimation constraint term of the linear Wiener process; The gradient descent method is used to minimize the loss function to update the parameters of the convolutional neural network and the degradation process model. After the training is completed, the optimal parameter set is output. The optimal parameter set includes the optimal weight parameter of the convolutional neural network, the optimal drift coefficient of the degradation process model, the optimal diffusion coefficient of the degradation process model, and the optimal failure threshold of the degradation process model. The drift coefficient obeys the normal distribution, that is, the optimal mean value of the drift coefficient and the optimal standard deviation of the drift coefficient are obtained.

2. The method for real-time prediction of the remaining life of a lithium battery based on dynamic uncertainty modeling according to claim 1, characterized in that: The method combines discrete Fourier transform to extract the degradation feature vector corresponding to each charge and discharge cycle in the entire life cycle of the lithium battery, including: For each charge and discharge cycle, the charging current sequence is discrete Fourier transformed to calculate the corresponding total spectrum energy; the discharge voltage sequence is discrete Fourier transformed to calculate the corresponding total spectrum energy; And calculate the average resistance of the charge and discharge cycle, the time of the constant voltage charging phase, the time of the constant current discharging phase, the average slope of the current decaying from A to B during the constant voltage charging phase, and the average slope of the voltage decaying from C to D during the constant current discharging phase, where A is greater than B and C is greater than D; The total spectral energy corresponding to the charging current sequence, the total spectral energy corresponding to the discharging voltage sequence, the average resistance of the charge and discharge cycle, the time of the constant voltage charging stage, the time of the constant current discharging stage, the average slope of the current decaying from A to B in the constant voltage charging stage, and the average slope of the voltage decaying from C to D in the constant current discharging stage are normalized respectively, and all normalized features are combined as the degenerate feature vector.

3. The method for real-time prediction of the remaining life of a lithium battery based on dynamic uncertainty modeling according to claim 1, characterized in that: The remaining life prediction error term includes: The mean square error is calculated based on the remaining life prediction value corresponding to each charge and discharge cycle in the entire life cycle of the lithium battery and the actual remaining life value, and the mean square error is used as the remaining life prediction error term.

4. The method for real-time prediction of the remaining life of a lithium battery based on dynamic uncertainty modeling according to claim 1, characterized in that: The logarithmic joint likelihood function of all dynamic health score increments is constructed, including: ； in, represents the log-joint likelihood function, is the total number of charge and discharge cycles in the entire life cycle of the lithium battery, For the The dynamic health score increment corresponding to the charge and discharge cycle, is the mean value of the drift coefficient, is the time increment between two adjacent charge and discharge cycles, is the diffusion coefficient.

5. The method for real-time prediction of the remaining life of a lithium battery based on dynamic uncertainty modeling according to claim 1, characterized in that: The real-time dynamic health score of the target lithium battery is obtained by the updated convolutional neural network, including: Acquire multi-sensor data of the target lithium battery during its current charge and discharge cycle; Combined with discrete Fourier transform, the degradation feature vector of multi-sensor data is extracted; Through the updated convolutional neural network, the real-time dynamic health score of the target lithium battery is output according to the feature vector.

6. The method for real-time prediction of remaining life of lithium batteries based on dynamic uncertainty modeling according to claim 1, characterized in that: The method of using Bayesian theory to update the drift coefficient of the degradation process model in real time includes: The drift coefficient of the degradation process model is updated in real time using Bayesian theory. The updated drift coefficient obeys the posterior distribution. The posterior distribution is calculated based on the optimal parameters of the degradation process model obtained after training. The calculation formula is as follows: ； in, is the standard deviation of the drift coefficient after Bayesian theory update, is the mean of the drift coefficient after Bayesian theory update, is the optimal diffusion coefficient of the degradation process model, is the standard deviation of the optimal drift coefficient of the degradation process model, is the current charge and discharge cycle number of the target lithium battery, is the time increment between two adjacent charge and discharge cycles, is the mean value of the optimal drift coefficient of the degradation process model, Target lithium battery The dynamic health score increment corresponding to the charge and discharge cycle.

Images