A lithium battery life prediction method based on decomposition migration multi-dimensional Gaussian process
By decomposing the multidimensional Gaussian process of migration, the lithium battery capacity decay curve is decomposed into the main trend and local fluctuation parts, which realizes accurate lifetime prediction under newly deployed batteries, reduces computational complexity, and is suitable for engineering applications of large-scale battery clusters.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ANHUI UNIV
- Filing Date
- 2026-03-31
- Publication Date
- 2026-06-12
AI Technical Summary
Existing methods for predicting the remaining lifespan of lithium-ion batteries are difficult to predict accurately in data-sparse scenarios. Overall migration is prone to negative migration and has high computational complexity, making it difficult to adapt to cold start scenarios with newly deployed batteries or changes in battery type, as well as the engineering practice of large-scale battery clusters.
The decomposition-transfer multidimensional Gaussian process method is adopted to decompose the lithium battery capacity decay curve into a main trend part, a local fluctuation part, and observation noise. The main trend part is transferred by sharing the historical battery, the local fluctuation part is fitted by a convolution process, and a joint covariance matrix with sparse structure is constructed to reduce computational complexity.
It significantly improves the accuracy of remaining life prediction, is suitable for cold start scenarios of newly deployed batteries, reduces computational complexity, and adapts to the engineering prediction needs of large-scale battery clusters.
Smart Images
Figure CN121955752B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of new energy lithium battery management technology, specifically involving a lithium battery lifetime prediction method based on a decomposition-migration multidimensional Gaussian process. Background Technology
[0002] Lithium-ion batteries, as a core component of green energy systems, are widely used in electric vehicles, grid energy storage, and other fields. The performance of lithium-ion batteries inevitably degrades with increasing cycle life, manifesting as capacity reduction and increased internal resistance. Failure to predict the remaining lifespan of lithium-ion batteries in a timely manner and replace degraded batteries can easily lead to system performance degradation or even catastrophic failures such as thermal runaway. Therefore, accurately predicting the remaining lifespan of lithium-ion batteries is crucial for ensuring safe system operation and reducing maintenance costs. Existing methods for predicting the remaining lifespan of lithium-ion batteries are mainly divided into filtering methods and decomposition methods. Filtering methods treat local fluctuations in the capacity decay process as high-frequency noise and extract the global decay trend using algorithms such as Kalman filtering and particle filtering. Decomposition methods attempt to decompose the capacity decay curve into different frequency components and predict lifespan by analyzing the variation patterns of each component. Furthermore, transfer learning methods offer a new approach to solving the problem of data scarcity by transferring rich historical battery knowledge to predict the remaining lifespan of in-service batteries with sparse data.
[0003] However, the capacity decay of lithium-ion batteries exhibits non-monotonic characteristics, resulting from the superposition of a global aging trend dominated by irreversible electrochemical degradation and local random fluctuations such as capacity regeneration caused by resting periods. Furthermore, these local fluctuations possess significant unit-specific characteristics. Existing filtering and decomposition methods rely on extensive historical observation data of the batteries to be predicted, making them unsuitable for cold-start scenarios involving newly deployed batteries or new battery types. Current transfer learning frameworks often employ a holistic transfer strategy, treating global trends and local fluctuations as a whole for knowledge transfer. However, the unit-specific nature of local fluctuations means that holistic transfer introduces random noise, such as capacity regeneration from source domain batteries, into the target domain model, leading to negative transfer problems and significantly increasing the remaining lifespan prediction error. Simultaneously, the covariance matrix of traditional multidimensional Gaussian process models is a dense matrix, resulting in high computational complexity for operations such as matrix inversion, making it difficult to adapt to practical engineering prediction scenarios involving large-scale battery clusters. Summary of the Invention
[0004] The purpose of this invention is to provide a lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process, in order to solve the technical problems of existing lithium battery remaining lifetime prediction methods, such as difficulty in accurate prediction in data-sparse scenarios, the tendency of overall transfer to induce negative transfer, and high computational complexity.
[0005] The present invention achieves the above objectives through the following technical solutions:
[0006] This invention proposes a lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process, the method comprising:
[0007] Obtain the cycle count and corresponding capacity observation data of historical and in-service batteries, and the corresponding battery capacity decay curves;
[0008] The historical battery capacity decay curve is input into a pre-built decomposition and migration multidimensional Gaussian process model;
[0009] The model decomposes the capacity decay curve of the in-service battery into a main trend part, a local fluctuation part, and observation noise. By sharing the latent Gaussian process corresponding to the main trend part in the historical battery capacity decay curve, the main trend part of the in-service battery is obtained through transfer, and the local fluctuation part of the in-service battery is fitted by a convolution process.
[0010] Based on the main trend component, the local fluctuation component, and the observation noise, a probabilistic prediction of the future capacity of in-service batteries is made.
[0011] Based on the capacity probability prediction results, when the capacity value decays to the preset end-of-life threshold, the remaining lifespan prediction information is generated in combination with the current cycle count. The prediction information includes the predicted mean of the remaining lifespan and its probability distribution.
[0012] As a preferred method, the steps of pre-constructing a decomposition and transfer multidimensional Gaussian process model include:
[0013] Establish a battery cluster consisting of N battery cells, and denote the first N-1 battery cells as historical batteries and the Nth battery cell as the in-service battery;
[0014] For the nth battery, its cycle count time series is defined as follows: , Let represent the cycle number of the nth battery in the mth cycle, and denote its corresponding capacity observation sequence as . ,in This represents the amount of capacity observation data for the battery. This represents the observed capacity of the nth battery during the mth cycle;
[0015] The capacity observation value of the nth battery in the mth cycle. Represented as ,in Let n be the potential degradation function of the nth battery. Follows a mean of 0 and a variance of The normal distribution represents independent and identically distributed Gaussian measurement noise.
[0016] Preferably, the model decomposes the capacity decay curve into a main trend component, a local fluctuation component, and observation noise, including:
[0017] Potentially degenerate function Decomposed into ,in The main trend component is transferable. This refers to the non-migratable local fluctuation component;
[0018] Observation noise correspondence This term is used to characterize random errors in the capacity measurement process.
[0019] Preferably, the main trend portion of the in-service battery is obtained by transferring the main trend portion corresponding to the main trend portion of the historical battery capacity degradation curve through a potential Gaussian process, including:
[0020] The main trend of historical batteries is defined as follows: ,in ;
[0021] The main trend of in-service batteries is defined as follows: ;
[0022] in, Obeying a Gaussian process For all batteries, a potential Gaussian process is shared. Let be the covariance function of the Gaussian process. , Both are cell-specific scaling factors, used to adjust the intensity of the aging trend of the nth historical cell and the in-service cell N, respectively.
[0023] Preferably, the covariance of the main trend component satisfies:
[0024]
[0025] Where n and n' are battery serial numbers, and N is the serial number of the battery in service. A scaling factor specific to the cell. Let be the covariance function of the Gaussian process.
[0026] Preferably, a convolution process is used to fit the local fluctuations of the in-service battery, including:
[0027] The local fluctuations are modeled using a convolution process, represented as follows: Perform fitting, where It is the specific smoothing kernel corresponding to the nth battery. It is a Gaussian white noise process with unit variance;
[0028] The local fluctuations of each battery are independent of each other, and their covariance satisfies ;
[0029] in, The Kronecker delta function is used to achieve complete isolation of local fluctuation knowledge between different cells. It takes the value 1 when n=n' and 0 otherwise.
[0030] Preferably, the model includes a hyperparameter optimization step before probabilistically predicting the future capacity of in-service batteries:
[0031] Construct all observation data of the battery cluster The joint covariance matrix K is formed by the superposition of the covariances of the main trend, local fluctuations, and observation noise. For any two observations... and The covariance satisfies:
[0032]
[0033] The matrix form of the joint covariance matrix is as follows: ,in Dense covariance matrix with dominant trend, The block diagonal covariance matrix is the result of local fluctuations. To determine the diagonal covariance matrix of the observation noise, It is the identity matrix. To observe the noise level;
[0034] Determine the set of model hyperparameters Hyperparameter estimation and optimization are achieved by maximizing the log-marginal likelihood function, which satisfies the following:
[0035]
[0036] in, This represents the total amount of observation data for the battery cluster. For the joint covariance matrix The determinant, For the joint covariance matrix The inverse matrix;
[0037] By utilizing the sparse structure of the joint covariance matrix, matrix inversion operations can be performed using block inversion or sparse solvers, thereby reducing the computational complexity of hyperparameter optimization.
[0038] The sparse structure is formed by the superposition of dense Ktrend, block-diagonal Klocal, and diagonal Knoise.
[0039] Preferably, based on the capacity probability prediction results, when the capacity value decays to a preset end-of-life threshold, prediction information of the remaining lifespan is generated in conjunction with the current cycle count, including:
[0040] For a new input cycle number x of the in-service battery N, based on the Gaussian process prediction principle, the predicted capacity distribution at that cycle number is obtained. ;
[0041] The predicted distribution follows a Gaussian distribution. ), where the predicted mean , For test points The covariance vector between all training observations, For test points Its own total covariance;
[0042] Based on the capacity probability prediction results, the full life cycle capacity degradation trajectory of in-service battery N is extrapolated and simulated until the capacity value drops to a preset battery life termination threshold. ;
[0043] Based on the extrapolated capacity decay trajectory, the number of cycles required for the battery to reach the end-of-life threshold is determined. Combined with the number of cycles already completed, the predicted average remaining lifespan is calculated.
[0044] Based on the Gaussian distribution of capacity prediction, the probability distribution of remaining useful life is obtained through propagation, generating remaining useful life prediction information that includes the prediction mean and probability distribution.
[0045] The battery life termination threshold is determined based on the nominal capacity of the lithium battery, and is the capacity value corresponding to a nominal capacity decay of 30% or more.
[0046] This invention proposes an application of the lithium battery remaining life prediction method as described above. The method is applicable to lithium-ion batteries with lithium nickel cobalt aluminum oxide and lithium cobalt oxide cathode materials, and can predict the remaining life of the battery in the early, middle and late stages of the entire life cycle when the observed data accounts for 30%, 50% and 70% respectively.
[0047] The beneficial effects of this invention are as follows:
[0048] 1. This invention decomposes the capacity decay curve into a transferable main trend part and a non-transferable local fluctuation part, and only performs transfer learning on the main trend, completely isolating the unit-specific knowledge of local fluctuations, fundamentally avoiding negative transfer caused by random information such as capacity regeneration, and significantly improving the accuracy of remaining service life prediction.
[0049] 2. This invention achieves knowledge transfer based on the global aging trend of historical batteries, without relying on a large amount of observation data of in-service batteries, and is suitable for cold start scenarios of newly deployed batteries.
[0050] 3. The joint covariance matrix constructed in this invention integrates the dense main trend matrix, the block-diagonal local fluctuation matrix, and the diagonal noise matrix to form a sparse structure. The matrix operation is completed by block inversion, which greatly reduces the computational complexity of the traditional multidimensional Gaussian process from cubic level, making it suitable for engineering prediction scenarios of large-scale battery clusters. Attached Figure Description
[0051] Figure 1 This is a flowchart illustrating a lithium battery remaining life prediction method based on a decomposition-transfer multidimensional Gaussian process proposed in this invention.
[0052] Figure 2 This is a schematic diagram of the overall structure of the lithium battery remaining lifetime prediction method based on the decomposition and migration multidimensional Gaussian process proposed in this invention.
[0053] Figure 3 Capacity decay trajectory plots of four lithium nickel cobalt aluminum oxide lithium-ion batteries (B0005, B0006, B0007, and B0018) in the NASA Battery Aging ARC-FY08Q4 dataset.
[0054] Figure 4 This is a prediction result of the capacity degradation trajectory of the in-service battery B0018 in the early stage (capacity data of the first 30% of the lifespan has been observed) and a distribution map of the 95% confidence interval.
[0055] Figure 5 This is a prediction result of the capacity degradation trajectory of the in-service battery B0018 in the mid-term stage (capacity data of the first 50% of the lifespan has been observed) and a distribution map of the 95% confidence interval.
[0056] Figure 6 This is a prediction result of the capacity degradation trajectory of the in-service battery B0018 in the later stage (capacity data of the first 70% of the lifespan has been observed) and a distribution map of the 95% confidence interval. Detailed Implementation
[0057] The following description provides specific application scenarios and requirements for this specification, intended to enable those skilled in the art to make and use the contents of this specification. Various partial modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but rather to the widest scope consistent with the claims.
[0058] The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not restrictive. For example, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. When used in this specification, the terms “comprising,” “including,” and / or “containing” mean that the associated integers, steps, operations, elements, and / or components are present, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components, and / or groups, or that other features, integers, steps, operations, elements, components, and / or groups may be added to the system / method.
[0059] Considering the following description, these and other features of this specification, as well as the operation and function of the related components of the structure, and the economy of assembly and manufacture of the parts, can be significantly improved. All of these form part of this specification with reference to the accompanying drawings. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.
[0060] The flowcharts used in this specification illustrate operations implemented according to some embodiments of this specification. It should be clearly understood that the operations in the flowcharts may not be implemented in a sequential order. Instead, the operations may be implemented in reverse order or simultaneously. Furthermore, one or more additional operations may be added to the flowcharts. One or more operations may be removed from the flowcharts.
[0061] Example 1
[0062] This embodiment uses the Battery Aging dataset published by NASA's Center for Failure Prediction of Excellence as the experimental subject to describe in detail the lithium battery remaining life prediction method based on decomposition-transfer multidimensional Gaussian process proposed in this invention. This dataset contains cycle aging test data for four lithium nickel cobalt aluminum oxide (Li-ion) batteries (B0005, B0006, B0007, and B0018), with a nominal capacity of 2 Ah and an experimental temperature of 24°C. The batteries were charged using a constant current / constant voltage mode and discharged using a constant current mode. The battery life was considered to end when the capacity decayed to 1.4 Ah, i.e., a 30% decrease in nominal capacity.
[0063] In this embodiment, three batteries, B0005, B0006, and B0007, are selected as historical batteries to form the source domain dataset; battery B0018 is selected as the in-service battery to form the target domain dataset. The actual cycle life of battery B0018 is 97 cycles. In this embodiment, the capacity observation data of the first 30%, 50%, and 70% of its capacity are used as known inputs to simulate the prediction scenario of the remaining service life of lithium batteries at different aging stages in the early, middle, and late stages.
[0064] See Figures 1-2 The present invention proposes a lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process, which specifically includes the following steps:
[0065] Step 1: Obtain the cycle count and corresponding capacity observation data of historical and in-service batteries, and construct battery capacity decay curves respectively.
[0066] First, extract the cycle test data for each battery from the dataset. For the nth battery, define its cycle count time series as follows: , Let represent the cycle number of the nth battery in the mth cycle, and denote its corresponding capacity observation sequence as . ,in This represents the amount of capacity observation data for the battery. This represents the observed capacity of the nth battery during the mth cycle.
[0067] Plot the capacity decay curves of historical batteries B0005, B0006, and B0007, as well as the capacity decay curve of in-service battery B0018, with the cycle number as the x-axis and the observed capacity value as the y-axis. For example... Figure 3 As shown, this is a capacity decay trajectory diagram of each battery in the NASA dataset. It can be observed from the diagram that the battery capacity shows an overall downward trend as the number of cycles increases, but there is a localized capacity regeneration phenomenon during the decline. This is due to the redistribution of internal chemicals during the battery's resting period.
[0068] Step 2: Input the historical battery capacity decay curve and the in-service battery capacity decay curve into the pre-built decomposition migration multidimensional Gaussian process model.
[0069] Before inputting the model, a decomposition and transfer multidimensional Gaussian process model needs to be pre-constructed. The construction process includes: establishing a battery cluster consisting of N battery cells, designating the first N-1 battery cells as historical cells, and the Nth battery cell as the in-service cell. In this embodiment, N=4, the historical cells are B0005, B0006, and B0007, and the in-service cell is B0018.
[0070] For the nth battery, its capacity observation in the mth cycle is... Represented as ,in Let n be the potential degradation function of the nth battery. Follows a mean of 0 and a variance of The normal distribution represents independent and identically distributed Gaussian measurement noise. By separating the observation noise from the potential degradation function, the true degradation pattern of the battery can be captured more accurately, avoiding the interference of measurement errors on the prediction results.
[0071] Step 3: The model decomposes the in-service battery capacity decay curve into the main trend part, the local fluctuation part, and the observation noise.
[0072] Based on the mathematical representation in step 2, the model will determine the potential degenerate function. Decomposed into ,in The main trend component that can be transferred represents the irreversible chemical aging process shared by all batteries in the battery cluster, which is dominated by irreversible electrochemical degradation mechanisms such as solid electrolyte interface growth and loss of active materials. This is a non-migratable local fluctuation component, characterizing cell-specific fluctuations such as battery capacity regeneration caused by random resting periods and environmental noise.
[0073] By decomposing the capacity decay curve into these three components, differentiated processing strategies can be adopted for different parts, laying the foundation for subsequent selective migration.
[0074] Step 4: By using the potential Gaussian process corresponding to the main trend part of the shared historical battery capacity decay curve, the main trend part of the in-service battery is obtained.
[0075] To achieve knowledge sharing on aging between batteries while taking into account the differences in degradation rates among different batteries, this invention employs a shared latent Gaussian process combined with a cell-specific scaling factor for subject trend transfer. Specifically:
[0076] The main trend of historical batteries is defined as follows: ,in ;
[0077] The main trend of in-service batteries is defined as follows: ;
[0078] in, Obeying a Gaussian process For all batteries, a potential Gaussian process is shared. Let be the covariance function of the Gaussian process. , Both are cell-specific scaling factors, used to adjust the intensity of the aging trend of the nth historical cell and the in-service cell N, respectively.
[0079] The covariance of the main trend component satisfies:
[0080]
[0081] By employing the aforementioned weighted combination method, this invention achieves both the effective transfer of historical battery aging knowledge and the flexibility of preserving the individual characteristics of in-service batteries. Compared to traditional overall transfer methods, this invention only transfers common aging patterns, avoiding the introduction of individual specificities from historical batteries into in-service batteries, thereby effectively preventing negative transfer.
[0082] Step 5: Use a convolution process to fit the local fluctuations of the in-service battery.
[0083] For cell-specific local fluctuations, to prevent their propagation between cells and resulting in negative migration, this invention employs a convolution process for modeling and ensures that the local fluctuations of each cell are independent. Specifically: Perform fitting, where It is a specific smoothing kernel corresponding to the nth cell, used to control the smoothing degree of local fluctuations. It is a Gaussian white noise process with unit variance, used to introduce randomness. Through convolution operation, the Gaussian white noise is smoothed by a kernel filter to generate a local fluctuation process with a correlated structure.
[0084] The local fluctuations of each battery are independent of each other, and their covariances satisfy:
[0085] ;
[0086] in, Let be the Kronecker delta function, used to achieve complete isolation of local fluctuation knowledge between different batteries. It takes a value of 1 when n=n', and 0 otherwise. This definition ensures complete isolation of local fluctuation knowledge between different batteries. Even if historical batteries exhibit strong capacity regeneration phenomena, these local fluctuation characteristics will not be transferred to the in-service battery model, fundamentally avoiding the negative transfer problem caused by random information such as capacity regeneration.
[0087] Step 6: Before the model makes probabilistic predictions about the future capacity of in-service batteries, hyperparameter optimization is performed.
[0088] To achieve accurate capacity prediction, it is necessary to optimize the set of hyperparameters in the model. theta First, construct all observation data for the battery cluster. The joint covariance matrix K is formed by the superposition of the covariances of the main trend, local fluctuations, and observation noise. For any two observations... and The covariance satisfies:
[0089]
[0090] The matrix form of the joint covariance matrix is as follows: ,in Dense covariance matrix with dominant trend, The block diagonal covariance matrix is the result of local fluctuations. To determine the diagonal covariance matrix of the observation noise, It is the identity matrix. To observe the noise level; determine the set of model hyperparameters. Hyperparameter estimation and optimization are achieved by maximizing the log-marginal likelihood function, which satisfies the following:
[0091]
[0092] in, This represents the total amount of observation data for the battery cluster. For the joint covariance matrix The determinant, For the joint covariance matrix The inverse matrix;
[0093] By utilizing the sparse structure of the joint covariance matrix, matrix inversion is performed using block inversion or a sparse solver, reducing the computational complexity of hyperparameter optimization. The sparse structure is formed by superimposing dense Ktrend, block-diagonal Klocal, and diagonal Knoise.
[0094] Step 7: Based on the main trend component, the local fluctuation component, and the observation noise, make a probabilistic prediction of the future capacity of the in-service batteries.
[0095] After hyperparameter optimization, the future capacity of the in-service battery N can be probabilistically predicted. For a new input cycle number x of the in-service battery N, based on the Gaussian process prediction principle, the predicted capacity distribution at that cycle number is obtained. The predicted distribution follows a Gaussian distribution. ), where the predicted mean , For test points The covariance vector between all training observations, For test points The total covariance of the battery itself; based on the capacity probability prediction results, extrapolation simulation is performed on the full life cycle capacity decay trajectory of the in-service battery N until the capacity value drops to the preset battery life termination threshold. Based on the extrapolated capacity decay trajectory, the number of cycles required for the battery to reach the end-of-life threshold is determined. Combined with the number of cycles already completed, the predicted mean of the remaining lifespan is calculated. Based on the Gaussian distribution of the capacity prediction, the probability distribution of the remaining lifespan is obtained, generating remaining lifespan prediction information containing the predicted mean and probability distribution. The battery lifespan end-of-life threshold is determined based on the nominal capacity of the lithium battery, which is the capacity value corresponding to a nominal capacity decay of 30% or more.
[0096] Step 8: Based on the capacity probability prediction results, when the capacity value decays to the preset end-of-life threshold, generate prediction information of the remaining lifespan by combining the current number of cycles.
[0097] Based on the capacity probability prediction results obtained in step 7, the full life cycle capacity degradation trajectory of the in-service battery N is extrapolated and simulated until the capacity value drops to a preset battery life termination threshold. In this embodiment, the battery life termination threshold is 1.4Ah, corresponding to a 30% degradation of the nominal capacity of 2Ah.
[0098] Based on the extrapolated capacity decay trajectory, the number of cycles required for the battery to reach its end-of-life threshold is determined. Combined with the current number of completed cycles, the predicted mean of the remaining lifespan is calculated. Simultaneously, based on the Gaussian distribution of the capacity prediction, an uncertainty propagation process is used to obtain the probability distribution of the remaining lifespan, generating a remaining lifespan prediction that includes both the predicted mean and the probability distribution.
[0099] This embodiment uses observed capacity data of the B0018 battery at 30%, 50%, and 70% capacity as known inputs to simulate the prediction effect at different aging stages. The prediction results show that with 30% data input, the predicted total lifetime is 105 cycles, with an error of 8 cycles compared to the actual total lifetime of 97 cycles; with 50% data input, the predicted total lifetime is 104 cycles, with an error of 7 cycles; and with 70% data input, the predicted total lifetime is 101 cycles, with an error of 4 cycles. Simultaneously, the model can output a 95% confidence interval for the predicted trajectory, such as... Figures 4-6 As shown, this is a prediction of the capacity degradation trajectory of lithium battery B0018 in the early, middle and late stages, and the distribution of its 95% confidence interval.
[0100] The prediction results show that even in the early aging stage with only 30% of the observation data, the method of this invention can still achieve relatively accurate predictions of remaining useful life, verifying its applicability in cold start scenarios. As the amount of observation data increases, the prediction accuracy gradually improves, demonstrating the model's excellent learning ability.
[0101] Example 2
[0102] This embodiment illustrates the application scenarios of the method of the present invention. The lithium battery remaining life prediction method based on the decomposition and migration multidimensional Gaussian process proposed in this invention is applicable to lithium-ion batteries with various cathode materials, including but not limited to lithium nickel cobalt aluminum oxide, lithium cobalt oxide, and lithium iron phosphate. Since batteries with different material systems have different electrochemical characteristics and degradation laws, but all follow an irreversible aging trend superimposed with a capacity degradation mode of local random fluctuations, the method of the present invention has wide applicability.
[0103] Regarding the amount of observation data, the method of this invention can predict the remaining service life in the early, middle, and late stages of the entire life cycle, with observation data accounting for 30%, 50%, and 70% respectively. In the early stage, since only a small amount of observation data is needed to start the prediction, it is particularly suitable for the cold start scenario of newly deployed batteries. In the middle and late stages, as observation data accumulates, the prediction accuracy is further improved, which can provide continuous and reliable decision support for battery operation and maintenance management.
[0104] In practical engineering applications, the battery management system can collect battery cycle count and capacity data in real time and periodically execute the method of this invention to dynamically update the remaining service life prediction results. When the predicted remaining service life is lower than the safety threshold, the system can issue a maintenance warning in advance and arrange a battery replacement plan, thereby avoiding system downtime or safety accidents caused by sudden battery failure.
[0105] Furthermore, the method of this invention can also be applied to battery reuse scenarios. For retired batteries, their remaining lifespan can be predicted using short-term capacity observation data to assess their suitability for secondary applications such as energy storage, thereby maximizing the utilization of battery resources.
[0106] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.
[0107] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.
[0108] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.
Claims
1. A method for predicting the lifetime of lithium batteries based on a decomposition-transfer multidimensional Gaussian process, characterized in that the method... include: Obtain the cycle count and corresponding capacity observation data of historical and in-service batteries, and the corresponding battery capacity decay curves; The historical battery capacity decay curve is input into a pre-built decomposition and migration multidimensional Gaussian process model; The model decomposes the capacity decay curve of the in-service battery into a main trend part, a local fluctuation part, and observation noise. By sharing the latent Gaussian process corresponding to the main trend part in the historical battery capacity decay curve, the main trend part of the in-service battery is obtained through transfer, and the local fluctuation part of the in-service battery is fitted by a convolution process. Based on the main trend component, the local fluctuation component, and the observation noise, a probabilistic prediction of the future capacity of in-service batteries is made. Based on the capacity probability prediction results, when the capacity value decays to the preset end-of-life threshold, the remaining lifespan prediction information is generated in combination with the current cycle count. The prediction information includes the predicted mean of the remaining lifespan and its probability distribution.
2. The lithium battery lifetime prediction method based on decomposition-transfer multidimensional Gaussian process according to claim 1, characterized in that, The steps for pre-constructing a decomposition and transfer multidimensional Gaussian process model include: Establish a battery cluster consisting of N battery cells, and denote the first N-1 battery cells as historical batteries and the Nth battery cell as the in-service battery; For the nth battery, its cycle count time series is defined as follows: , Let represent the cycle number of the nth battery in the mth cycle, and denote its corresponding capacity observation sequence as . ,in This represents the amount of capacity observation data for the battery. This represents the observed capacity of the nth battery during the mth cycle; The capacity observation value of the nth battery in the mth cycle. Represented as ,in Let n be the potential degradation function of the nth battery. Follows a mean of 0 and a variance of The normal distribution represents independent and identically distributed Gaussian measurement noise.
3. A lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process according to claim 2, characterized in that, The model decomposes the capacity decay curve into a main trend component, a local fluctuation component, and observation noise, including: Potentially degenerate function Decomposed into ,in The main trend component is transferable. This refers to the non-migratable local fluctuation component; Observation noise correspondence This term is used to characterize random errors in the capacity measurement process.
4. A lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process according to claim 3, characterized in that, By using a latent Gaussian process corresponding to the main trend portion of the historical battery capacity degradation curve, the main trend portion of the in-service battery is obtained, including: The main trend of historical batteries is defined as follows: ,in ; The main trend of in-service batteries is defined as follows: ; in, Obeying a Gaussian process For all batteries, a potential Gaussian process is shared. Let be the covariance function of the Gaussian process. , Both are cell-specific scaling factors, used to adjust the intensity of the aging trend of the nth historical cell and the in-service cell N, respectively.
5. A lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process according to claim 4, characterized in that, The covariance of the main trend component satisfies: ； Where n and n' are battery serial numbers, and N is the serial number of the battery in service. A scaling factor specific to the cell. Let be the covariance function of the Gaussian process.
6. A lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process according to claim 3, characterized in that, The convolution process is used to fit the local fluctuations of the in-service battery, including: The local fluctuations are modeled using a convolution process, represented as follows: ； Perform fitting, where It is the specific smoothing kernel corresponding to the nth battery. It is a Gaussian white noise process with unit variance; The local fluctuations of each battery are independent of each other, and their covariance satisfies ； in, The Kronecker delta function is used to achieve complete isolation of local fluctuation knowledge between different cells. It takes the value 1 when n=n' and 0 otherwise.
7. A lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process according to claim 3, characterized in that, Before the model makes probabilistic predictions about the future capacity of in-service batteries, it also includes a hyperparameter optimization step: Construct all observation data of the battery cluster The joint covariance matrix K is formed by the superposition of the covariances of the main trend, local fluctuations, and observation noise. For any two observations... and The covariance satisfies: ； The matrix form of the joint covariance matrix is as follows: ,in Dense covariance matrix with dominant trend, The block diagonal covariance matrix is the result of local fluctuations. To determine the diagonal covariance matrix of the observation noise, It is the identity matrix. To observe the noise level; Determine the set of model hyperparameters Hyperparameter estimation and optimization are achieved by maximizing the log-marginal likelihood function, which satisfies the following: ； in, This represents the total amount of observation data for the battery cluster. For the joint covariance matrix The determinant, For the joint covariance matrix The inverse matrix; By utilizing the sparse structure of the joint covariance matrix, matrix inversion operations can be performed using block inversion or sparse solvers, thereby reducing the computational complexity of hyperparameter optimization. The sparse structure is formed by the superposition of dense Ktrend, block-diagonal Klocal, and diagonal Knoise.
8. A lithium battery lifetime prediction method based on a decomposition-transfer multidimensional Gaussian process according to claim 1, characterized in that, Based on the capacity probability prediction results, when the capacity value decays to a preset end-of-life threshold, prediction information for the remaining lifespan is generated by combining the current cycle count, including: For a new input cycle number x of the in-service battery N, based on the Gaussian process prediction principle, the predicted capacity distribution at that cycle number is obtained. ; The predicted distribution follows a Gaussian distribution. ), where the predicted mean , For test points The covariance vector between all training observations; Based on the capacity probability prediction results, the full life cycle capacity degradation trajectory of in-service battery N is extrapolated and simulated until the capacity value drops to a preset battery life termination threshold. ; Based on the extrapolated capacity decay trajectory, the number of cycles required for the battery to reach the end-of-life threshold is determined. Combined with the number of cycles already completed, the predicted average remaining lifespan is calculated. Based on the Gaussian distribution of capacity prediction, the probability distribution of remaining useful life is obtained through propagation, generating remaining useful life prediction information that includes the prediction mean and probability distribution. The battery life termination threshold is determined based on the nominal capacity of the lithium battery, and is the capacity value corresponding to a nominal capacity decay of 30% or more.
9. An application of the lithium battery life prediction method according to any one of claims 1-8, characterized in that, The method is applicable to lithium-ion batteries with lithium nickel cobalt aluminum oxide and lithium cobalt oxide cathode materials, and can predict the remaining service life in the early, middle and late stages of the entire life cycle when the observed data accounts for 30%, 50% and 70% respectively.