A method, device and electronic equipment for predicting the remaining service life of a lithium ion battery

Abstract

This invention discloses a method, apparatus, and electronic device for predicting the remaining lifespan of a lithium-ion battery. The method involves acquiring the battery capacity sequence and basic feature set of the lithium-ion battery to be predicted within a preset time period. The battery capacity sequence and basic feature set are input into a remaining lifespan prediction model to obtain a battery capacity prediction set. The model includes a local denoising and temporal reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local denoising and temporal reconstruction module performs denoising, smoothing, and temporal reconstruction processing to obtain multivariate temporal features. The global inversion decoupling module obtains hidden layer state features. The global linear attention feature extraction module obtains deep representation features based on a global linear attention mechanism. The prediction module obtains the battery capacity prediction set. Based on the battery capacity prediction set and a preset battery lifespan termination threshold, a relatively accurate remaining lifespan of the lithium-ion battery to be predicted is obtained.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery monitoring technology, and in particular to a method, apparatus, and electronic device for predicting the remaining service life of lithium-ion batteries. Background Technology

[0002] Lithium-ion batteries, with their high energy density and excellent cycle performance, have become core components of new energy vehicles and large-scale energy storage systems. Accurately predicting the remaining lifespan of lithium-ion batteries is crucial to ensuring the safe and stable operation of these power vehicles and energy storage systems throughout their entire lifecycle.

[0003] In existing technologies, data-driven remaining lifespan prediction models do not require in-depth analysis of the extremely complex electrochemical degradation processes within the battery. They can directly extract features from battery operating data to construct capacity degradation mapping relationships, and have become the mainstream technology direction in this field. Specifically, by introducing a lithium-ion battery remaining lifespan prediction model based on the Transformer architecture, multi-dimensional health features are extracted from the historical operating trajectories such as charge and discharge voltage and current to conduct time-series modeling, thereby enabling the prediction of the remaining lifespan of lithium-ion batteries.

[0004] However, for remaining battery life prediction models incorporating the Transformer architecture, this architecture, inherently designed to solve long sequence modeling problems, directly adopts the processing paradigm of natural language processing. It simply concatenates multidimensional features at the input layer, disrupting the continuity of individual physical variables over time. This results in the prediction model's inability to analyze the underlying physicochemical coupling mechanism between battery capacity degradation and various health characteristics at a global level, leading to poor generalization ability when dealing with complex and variable battery aging mechanisms. Furthermore, real-world battery data not only contains significant high-frequency measurement noise but also frequently exhibits local capacity regeneration phenomena, easily causing abrupt interference in the prediction process of lithium-ion battery remaining battery life. In summary, existing technologies cannot obtain highly accurate prediction results for the remaining battery life of lithium-ion batteries. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, and electronic device for predicting the remaining lifespan of lithium-ion batteries, in order to solve the problem that existing technologies cannot obtain highly accurate prediction results for the remaining lifespan of lithium-ion batteries.

[0006] To achieve the above objectives, in a first aspect, embodiments of the present invention provide a method for predicting the remaining lifespan of a lithium-ion battery, comprising: Obtain the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence; wherein, the preset time period includes: multiple preset time windows, and the basic feature set includes: multiple basic feature subsets corresponding to the multiple preset time windows, and the basic feature subsets include: charging current integral area feature, charging voltage integral area feature, constant current charging time feature, discharge usable energy equivalent area feature, constant current discharging time feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature; The battery capacity sequence and the basic feature set are input into the trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted. The remaining lifespan prediction model includes: a local noise reduction and temporal reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local noise reduction and temporal reconstruction module is used to perform noise reduction, smoothing, and temporal reconstruction processing to obtain multivariate temporal features. The global inversion decoupling module is used to obtain hidden layer state features. The global linear attention feature extraction module is used to obtain deep representation features based on the global linear attention mechanism. The prediction module is used to obtain the battery capacity prediction set. The remaining lifespan of the lithium-ion battery to be predicted is determined based on the battery capacity prediction set and the preset battery life end threshold.

[0007] In one embodiment, before inputting the battery capacity sequence and the basic feature set into the trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted, the method further includes: Obtain a training set, which includes: training battery capacity sequences corresponding to multiple preset time windows, a training basic feature subset corresponding to each training battery capacity sequence, and a battery capacity label sequence for the next charge-discharge cycle corresponding to each preset time window. The training set is input into the initial remaining useful life prediction model for training. The weight parameters of the model are adjusted according to the preset loss function until the model converges, and the trained remaining useful life prediction model is obtained.

[0008] In one embodiment, before inputting the battery capacity sequence and the basic feature set into the trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted, the method further includes: The basic feature set is subjected to noise processing based on Gaussian noise.

[0009] In one embodiment, inputting the battery capacity sequence and the basic feature set into a trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted includes: The battery capacity sequence and the basic feature set are input into the local noise reduction and time-series reconstruction module. The local noise reduction and time-series reconstruction module performs noise reduction, smoothing and time-series reconstruction processing to obtain the multivariate time-series features. The multivariate temporal features are input into the global inversion decoupling module, and global inversion decoupling processing is performed through the global inversion decoupling module to obtain the hidden layer state features; The hidden layer state features are input into the global linear attention feature extraction module, and the global linear attention feature extraction module performs global linear attention feature extraction processing to obtain the deep representation features; The deep representation features are input into the prediction module, and the prediction module obtains the battery capacity prediction set.

[0010] In one embodiment, the local denoising and temporal reconstruction module includes: a denoising autoencoder, a gated loop unit, and a fusion unit; the step of inputting the battery capacity sequence and the basic feature set into the local denoising and temporal reconstruction module, and performing denoising smoothing and temporal reconstruction processing through the local denoising and temporal reconstruction module to obtain the multivariate temporal features includes: The basic feature set is input into the denoising autoencoder, and the denoising autoencoder performs denoising and smoothing processing on the basic feature set to obtain the underlying physical features. The underlying physical features are input into the gated loop unit, and the gated loop unit performs temporal reconstruction processing on the underlying physical features to obtain the gated hidden state features. The underlying physical features, the gated hidden state features, and the battery capacity sequence are input into the fusion unit. The fusion unit then performs splicing and fusion processing on the underlying physical features, the gated hidden state features, and the battery capacity sequence to obtain the multivariate time-series features.

[0011] In one embodiment, the global inversion decoupling module includes a feature transpose unit and a linear projection unit. The step of inputting the multivariate temporal features into the global inversion decoupling module and performing global inversion decoupling processing through the global inversion decoupling module to obtain the hidden layer state features includes: The multivariate time series features are input into the feature transpose unit, and the multivariate time series features are transposed by the feature transpose unit to obtain the transposed multivariate time series features. The transposed multivariate temporal features are input into the linear projection unit, and linear projection processing is performed through the linear projection unit to obtain the hidden layer state features.

[0012] In one embodiment, the global linear attention feature extraction module includes: a multi-head global linear attention feature extraction unit and a feedforward neural network residual connection unit. The step of inputting the hidden layer state features into the global linear attention feature extraction module and performing global linear attention feature extraction processing through the global linear attention feature extraction module to obtain the deep representation features includes: The hidden layer state features are input into the multi-head global linear attention feature extraction unit, and the multi-head global linear attention feature extraction unit performs global linear attention feature extraction processing on the hidden layer state features to obtain global linear attention features. The global linear attention features are input into the residual connection unit of the feedforward neural network to obtain the deep representation features.

[0013] In one embodiment, the prediction module includes: a shared linear prediction head, an inverse transpose and inverse normalization unit, and a multidimensional slicing unit. The step of inputting the deep representation features into the prediction module and obtaining the battery capacity prediction set through the prediction module includes: The deep representation features are input into a shared linear prediction head, and the shared linear prediction head is used to perform prediction processing on the deep representation features to obtain intermediate prediction features of battery capacity. The intermediate battery capacity prediction features are input into the reverse transpose and reverse normalization unit. The reverse transpose and reverse normalization unit performs reverse transpose and reverse normalization processing on the intermediate battery capacity prediction features to obtain the initial battery capacity prediction set. The initial battery capacity prediction set is input into the multidimensional slicing unit, and the multidimensional slicing unit performs multidimensional slicing operations on the initial battery capacity prediction set to obtain the battery capacity prediction set.

[0014] Secondly, embodiments of the present invention provide a lithium-ion battery remaining lifespan prediction device, comprising: The data acquisition module is used to acquire the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence; wherein, the preset time period includes: multiple preset time windows, and the basic feature set includes: multiple basic feature subsets corresponding to the multiple preset time windows, and the basic feature subsets include: charging current integral area feature, charging voltage integral area feature, constant current charging time feature, discharge usable energy equivalent area feature, constant current discharging time feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature; A battery capacity prediction set acquisition module is used to input the battery capacity sequence and the basic feature set into a trained remaining life prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted. The remaining life prediction model includes: a local denoising and temporal reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local denoising and temporal reconstruction module is used to perform denoising smoothing and temporal reconstruction processing to obtain multivariate temporal features. The global inversion decoupling module is used to obtain hidden layer state features. The global linear attention feature extraction module is used to obtain deep representation features based on a global linear attention mechanism. The prediction module is used to obtain the battery capacity prediction set. The remaining lifespan determination module is used to determine the remaining lifespan of the lithium-ion battery to be predicted based on the battery capacity prediction set and the preset battery lifespan termination threshold.

[0015] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in the first aspect.

[0016] The technical solution provided by the embodiments of the present invention has the following advantages compared with the prior art: The present invention provides a method for predicting the remaining service life of a lithium-ion battery. This method obtains the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence. The preset time period includes multiple preset time windows, and the basic feature set includes multiple basic feature subsets corresponding to the multiple preset time windows. The basic feature subsets include: charging current integral area feature, charging voltage integral area feature, constant current charging time feature, discharge usable energy equivalent area feature, constant current discharging time feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature. The battery capacity sequence and basic feature set are input into a trained remaining lifespan prediction model to obtain a battery capacity prediction set for the lithium-ion battery to be predicted. The remaining lifespan prediction model includes: a local denoising and time-series reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local denoising and time-series reconstruction module performs denoising, smoothing, and time-series reconstruction to obtain multivariate time-series features, achieving high-frequency denoising of the basic feature set and extracting sufficient low-frequency time-series features, perfectly decoupling high-frequency denoising and low-frequency time-series feature extraction. The global inversion decoupling module obtains hidden layer state features, fully extracting dynamic hidden layer state features relative to the global time series. The global linear attention feature extraction module obtains deep representation features based on a global linear attention mechanism, reconstructing the deep co-evolution law between multiple features under a global perspective, reducing computational power consumption to a linear level, and improving the model's computational efficiency. The prediction module obtains the battery capacity prediction set. Finally, based on the battery capacity prediction set and a preset battery life termination threshold, the remaining lifespan of the lithium-ion battery to be predicted with high accuracy is obtained. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating a method for predicting the remaining lifespan of a lithium-ion battery according to an embodiment of the present invention. Figure 2 A schematic diagram of a remaining useful life prediction model provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a lithium-ion battery remaining life prediction device provided in an embodiment of the present invention. Detailed Implementation

[0018] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0019] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0020] In this invention, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between the associated objects, indicating that three relationships can exist.

[0021] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a method for predicting the remaining lifespan of a lithium-ion battery according to an embodiment of the present invention, specifically including the following steps: S10: Obtain the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence.

[0022] The preset duration includes multiple preset time windows T, each consisting of multiple charge / discharge cycles of lithium-ion batteries. The basic feature set includes multiple subsets of basic features corresponding to the preset time windows. Each subset of basic features includes: charging current integral area feature, charging voltage integral area feature, constant current charging duration feature, equivalent area feature of usable discharge energy, constant current discharging duration feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature.

[0023] It should be noted that the integral area characteristics of charging current, charging voltage, constant current charging duration, equivalent area characteristics of usable discharge energy, constant current discharging duration, peak voltage characteristics of incremental capacity, peak height characteristics of incremental capacity, and integral area characteristics of the main peak of incremental capacity are obtained by feature extraction and normalization processing based on battery data collected during the charge-discharge cycle of the lithium-ion battery for each preset time window. The battery capacity sequence mentioned above is also determined based on battery data collected during the charge-discharge cycle of the lithium-ion battery.

[0024] Specifically, for the lithium-ion battery to be predicted, its battery capacity sequence within a preset time period and the basic feature set corresponding to the battery capacity sequence are obtained.

[0025] S11: Input the battery capacity sequence and the basic feature set into the trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted.

[0026] Specifically, after obtaining the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence, the battery capacity sequence and the basic feature set are input into the trained remaining lifespan prediction model. The remaining lifespan prediction model is then used for prediction processing to obtain the battery capacity prediction set of the lithium-ion battery to be predicted.

[0027] Optionally, based on the above embodiments, in some embodiments of the present invention, the method further includes the following step before performing S11: S20: Obtain the training set.

[0028] The training set includes: training battery capacity sequences corresponding to multiple preset time windows, a subset of training basic features corresponding to each training battery capacity sequence, and a battery capacity label sequence for the next charge-discharge cycle corresponding to each preset time window. For a preset time window that includes multiple charge-discharge cycles of lithium batteries, the battery capacity label sequence is the actual battery capacity sequence for the next charge-discharge cycle of multiple lithium batteries. For example, assuming that a preset time window consists of N charge-discharge cycles of lithium batteries, the battery capacity label sequence is the actual battery capacity sequence corresponding to the (N+1)th charge-discharge cycle of the lithium battery. However, this invention is not limited to this and is not specifically restricted. Those skilled in the art can set it according to the actual situation.

[0029] S21: Input the training set into the initial remaining useful life prediction model for training, adjust the model's weight parameters according to the preset loss function until the model converges, and obtain the trained remaining useful life prediction model.

[0030] The preset loss function can be cross-entropy loss, but it is not limited to this. This invention is not specifically limited, and those skilled in the art can set it according to the actual situation.

[0031] Specifically, a training set for training the remaining lifespan prediction model is obtained. This training set includes training battery capacity sequences corresponding to multiple preset time windows, a subset of training basic features corresponding to each training battery capacity sequence, and a battery capacity label sequence for the next charge-discharge cycle corresponding to each preset time window. After obtaining the training set for training the remaining lifespan prediction model, the training set is input into the initial remaining lifespan prediction model for training. The weight parameters of the model are adjusted according to the preset loss function until the model converges, and the trained remaining lifespan prediction model is obtained.

[0032] Optionally, based on the above embodiments, in some embodiments of the present invention, before performing S11, the method further includes: adding noise to the basic feature set according to Gaussian noise.

[0033] For example, for the basic feature set To the basic feature set Actively inject Gaussian noise This enables the addition of noise to the basic feature set.

[0034] Optionally, based on the above embodiments, refer to Figure 2 As shown, the remaining lifespan prediction model includes: a local denoising and temporal reconstruction module 10, a global inversion decoupling module 11, a global linear attention feature extraction module 12, and a prediction module 13. The local denoising and temporal reconstruction module 10 is used to perform denoising smoothing and temporal reconstruction processing to obtain multivariate temporal features. The global inversion decoupling module 11 is used to obtain hidden layer state features. The global linear attention feature extraction module 12 is used to obtain deep representation features based on a global linear attention mechanism. The prediction module 13 is used to obtain the battery capacity prediction set. Based on this, in some embodiments of the present invention, one implementation of S11 can be: S111: Input the battery capacity sequence and basic feature set into the local denoising and time-series reconstruction module. The local denoising and time-series reconstruction module performs denoising smoothing and time-series reconstruction processing to obtain multivariate time-series features.

[0035] Specifically, after obtaining the battery capacity sequence and the basic feature set, the battery capacity sequence and the basic feature set are input into the local denoising and temporal reconstruction module. The local denoising and temporal reconstruction module performs denoising and smoothing processing and temporal reconstruction processing to obtain multivariate temporal features.

[0036] Optionally, based on the above embodiments, continue to refer to... Figure 2As shown, the local denoising and timing reconstruction module 10 includes: a denoising autoencoder 101, a gated loop unit 102, and a fusion unit 103. Based on this, in some embodiments of the present invention, one implementation of S111 can be: S1111: Input the basic feature set into the denoising autoencoder, and perform denoising and smoothing processing on the basic feature set to obtain the underlying physical features.

[0037] Denoising autoencoders (DAEs) are a type of autoencoder that takes corrupted data (inputting Gaussian noise) as its input (i.e., a set of basic input features) and outputs predicted features from the original, uncorrupted basic data. This method forces the model to learn the essential features of the basic feature set through a noise injection mechanism. In an unsupervised state, it obtains smooth and robust underlying physical features through nonlinear approximation and physical manifold reconstruction using a low-dimensional hidden feature space. This avoids the drawback of traditional denoising autoencoders that simply copy the input data.

[0038] S1112: Input the underlying physical features into the gated loop unit, and perform temporal reconstruction processing on the underlying physical features through the gated loop unit to obtain the gated hidden state features.

[0039] In particular, the static recovery process of lithium-ion batteries during degradation can trigger extremely complex nonlinear fluctuations. Based on this, a dynamic valve mechanism can be constructed by the update gate and reset gate included in the gated loop unit. This mechanism can highly sensitively capture and store the short-term, local capacity degradation oscillations in the underlying physical characteristics, thereby obtaining the gated hidden state characteristics. This will fundamentally and effectively solve the above problems.

[0040] Specifically, the basic feature set is input into a denoising autoencoder, which performs denoising and smoothing on the basic feature set to obtain the underlying physical features. After obtaining the underlying physical features, the underlying physical features are input into a gated recurrent unit, which performs temporal reconstruction processing on the underlying physical features to obtain the gated hidden state features.

[0041] S1113: Input the underlying physical features, gated hidden state features and battery capacity sequence into the fusion unit. The fusion unit splices and fuses the underlying physical features, gated hidden state features and battery capacity sequence to obtain multivariate time series features.

[0042] Specifically, after obtaining the gated hidden state features, the underlying physical features, gated hidden state features, and battery capacity sequence are spliced ​​and fused by the fusion unit to obtain multivariate time-series features. .

[0043] It should be noted that for multivariate time series features The feature dimension is the sum of the dimensions of the underlying physical features, the gated hidden state features, and the battery capacity sequence.

[0044] S112: Input the multivariate temporal features into the global inversion decoupling module, and perform global inversion decoupling processing through the global inversion decoupling module to obtain the hidden layer state features.

[0045] Specifically, after obtaining the multivariate time series features, the multivariate time series features are input into the global inversion decoupling module. The global inversion decoupling module performs global inversion decoupling processing to obtain the hidden layer state features.

[0046] Optionally, based on the above embodiments, continue to refer to... Figure 2 As shown, the global inversion decoupling module 20 includes a feature transpose unit 201 and a linear projection unit 202. Based on this, in some embodiments of the present invention, one implementation of S112 can be: S1121: Input the multivariate time series features into the feature transpose unit, and transpose the multivariate time series features through the feature transpose unit to obtain the transposed multivariate time series features.

[0047] Specifically, for multivariate time series features Multivariate time series features The input is fed into the feature transpose unit, which transposes the multivariate time series features to obtain the transposed multivariate time series features. .

[0048] S1122: Input the transposed multivariate temporal features into the linear projection unit, perform linear projection processing through the linear projection unit, and obtain the hidden layer state features.

[0049] Specifically, in obtaining the transposed multivariate time series features Next, the transposed multivariate time series features will be... The input is fed into a linear projection unit, where linear projection is performed to obtain the hidden layer state features. .

[0050] Optionally, based on the above embodiments, in some embodiments of the present invention, the hidden layer state characteristics may be defined by the following expression: ; in, and This represents the extraction weights and bias matrices for transposed multivariate time series features.

[0051] S113: Input the hidden layer state features into the global linear attention feature extraction module, and perform global linear attention feature extraction processing through the global linear attention feature extraction module to obtain deep representation features.

[0052] Specifically, after obtaining the hidden layer state features, these features are input into the global linear attention feature extraction module. The module then performs global linear attention feature extraction on the hidden layer state features to obtain deep representation features. .

[0053] Optionally, based on the above embodiments, continue to refer to... Figure 2 As shown, the global linear attention feature extraction module 30 includes: a multi-head global linear attention feature extraction unit 301 and a feedforward neural network residual connection unit 302. Based on this, in some embodiments of the present invention, one implementation of S113 can be: S1131: Input the hidden layer state features into the multi-head global linear attention feature extraction unit, and perform global linear attention feature extraction processing on the hidden layer state features through the multi-head global linear attention feature extraction unit to obtain global linear attention features.

[0054] Specifically, the hidden layer state features are input into the multi-head global linear attention feature extraction unit. In the global linear attention feature extraction unit, the features are extracted based on the linear projection matrix. Generate the target query matrix corresponding to the hidden layer state features. Key matrix Sum matrix Furthermore, a non-negative kernel feature mapping function is introduced, and global linear attention features are extracted through the GLA core equation to obtain global linear attention features.

[0055] Optionally, based on the above embodiments, in some embodiments of the present invention, the GLA core equation may be defined by the following expression: ; in, This represents element-wise division, used to normalize global linear attention weights; It is a matrix of all 1s.

[0056] It should be noted that by utilizing the introduction of a non-negative kernel feature mapping function... Global linear attention feature extraction is performed on the GLA core equation, since the computation is prioritized. It can transform the massive interaction of hidden layer state features into feature matrix multiplication of extremely small dimensions, reducing the time and space complexity of the original global linear attention computation from quadratic to quadratic. Shrink to an extremely lightweight linear This improves the computational efficiency of the model.

[0057] S1132: Input the global linear attention features into the residual connection unit of the feedforward neural network to obtain deep representation features.

[0058] Specifically, after obtaining the global linear attention features, the global linear attention features are input into the residual connection unit of the feedforward neural network. The residual connection unit of the feedforward neural network performs residual processing to obtain deep representation features.

[0059] S114: Input the deep representation features into the prediction module, and obtain the battery capacity prediction set through the prediction module.

[0060] Specifically, after obtaining the deep representation features, the deep representation features are input into the prediction module, and the prediction module obtains the battery capacity prediction set.

[0061] Optionally, based on the above embodiments, continue to refer to... Figure 2 As shown, the prediction module 40 includes: a shared linear prediction head 401, an inverse transpose and inverse normalization unit 402, and a multidimensional slicing unit 403. Based on this, in some embodiments of the present invention, one implementation of S114 can be: S1141: Input the deep representation features into the shared linear prediction head, and use the shared linear prediction head to perform prediction processing on the deep representation features to obtain intermediate prediction features of battery capacity.

[0062] Specifically, after obtaining the deep representation features, the deep representation features are input into the shared linear prediction head, and the shared linear prediction head is used to perform prediction processing on the deep representation features to obtain the intermediate prediction features of battery capacity.

[0063] Optionally, based on the above embodiments, in some embodiments of the present invention, one implementation of S1141 may be: By sharing a linear prediction head, the deep representation matrix Directly mapped to the expected remaining useful life prediction step size The intermediate prediction features of battery capacity are obtained. The intermediate prediction features of battery capacity can be defined by the following expression: ; in, To share the weight matrix of the linear prediction head, The bias matrix for a shared linear prediction head.

[0064] S1142: Input the intermediate battery capacity prediction features into the inverse transpose and inverse normalization unit, and perform inverse transpose and inverse normalization processing on the intermediate battery capacity prediction features through the inverse transpose and inverse normalization unit to obtain the initial battery capacity prediction set.

[0065] Optionally, based on the above embodiments, in some embodiments of the present invention, one implementation of S1142 may be: The intermediate battery capacity prediction features are inverted using a spatial topology transpose operation via inverse transpose and inverse normalization units. Combining Hadamard product and broadcast addition mechanisms, numerical recalibration with inverse normalization is performed to accurately restore the true numerical scale structure of the intermediate battery capacity prediction features, resulting in an initial battery capacity prediction set. : ; in, The standard deviation of the intermediate predictive features of battery capacity. For Hadama product, This represents the mean of the intermediate predicted features of battery capacity.

[0066] S1143: Input the initial battery capacity prediction set into the multidimensional slicing unit, and perform multidimensional slicing operation on the initial battery capacity prediction set through the multidimensional slicing unit to obtain the battery capacity prediction set.

[0067] Specifically, the obtained initial battery capacity prediction set is input into the multidimensional slicing unit, and the initial battery capacity prediction set is sliced ​​in a multidimensional way to obtain the battery capacity prediction set.

[0068] S12: Determine the remaining lifespan of the lithium-ion battery to be predicted based on the battery capacity prediction set and the preset battery lifespan termination threshold.

[0069] The preset battery life termination threshold is a value set to determine the remaining lifespan of the lithium-ion battery to be predicted. The preset battery life termination threshold can be, for example, 80% of the nominal capacity of the battery, but is not limited thereto. The present invention does not specifically limit this, and those skilled in the art can set it according to the actual situation.

[0070] Specifically, among the multiple battery capacity prediction values ​​included in the battery capacity prediction set, the battery capacity prediction value that first falls below the preset battery life end threshold is determined as a specific cycle point. The total number of cycles between the initial predicted battery capacity value and the initial predicted battery capacity value that first falls below the preset battery life end threshold is determined. Based on the preset total number of cycles and the total number of cycles, the remaining lifespan of the lithium-ion battery to be predicted is determined.

[0071] Thus, the lithium-ion battery remaining lifespan prediction method provided in this embodiment obtains the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence. The preset time period includes multiple preset time windows, and the basic feature set includes multiple basic feature subsets corresponding to the multiple preset time windows. The basic feature subsets include: charging current integral area feature, charging voltage integral area feature, constant current charging time feature, discharge usable energy equivalent area feature, constant current discharging time feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature. The battery capacity sequence and basic feature set are input into a trained remaining lifespan prediction model to obtain a battery capacity prediction set for the lithium-ion battery to be predicted. The remaining lifespan prediction model includes: a local denoising and time-series reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local denoising and time-series reconstruction module performs denoising, smoothing, and time-series reconstruction to obtain multivariate time-series features, achieving high-frequency denoising of the basic feature set and extracting sufficient low-frequency time-series features, perfectly decoupling high-frequency denoising and low-frequency time-series feature extraction. The global inversion decoupling module obtains hidden layer state features, fully extracting dynamic hidden layer state features relative to the global time series. The global linear attention feature extraction module obtains deep representation features based on a global linear attention mechanism, reconstructing the deep co-evolution law between multiple features under a global perspective, reducing computational power consumption to a linear level, and improving the model's computational efficiency. The prediction module obtains the battery capacity prediction set. Finally, based on the battery capacity prediction set and a preset battery life termination threshold, the remaining lifespan of the lithium-ion battery to be predicted with high accuracy is obtained.

[0072] In one embodiment, such as Figure 3 As shown, Figure 3 A schematic diagram of a lithium-ion battery remaining life prediction device provided in an embodiment of the present invention includes: a data acquisition module 10 to be predicted, a battery capacity prediction set acquisition module 11, and a remaining life determination module 12.

[0073] The data acquisition module 10 is used to acquire the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence. The preset time period includes multiple preset time windows, and the basic feature set includes multiple basic feature subsets corresponding to the multiple preset time windows. The basic feature subsets include: charging current integral area feature, charging voltage integral area feature, constant current charging time feature, discharge usable energy equivalent area feature, constant current discharging time feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature.

[0074] The battery capacity prediction set acquisition module 11 is used to input the battery capacity sequence and basic feature set into the trained remaining service life prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted. The remaining service life prediction model includes: a local noise reduction and temporal reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local noise reduction and temporal reconstruction module is used to perform noise reduction, smoothing, and temporal reconstruction processing to obtain multivariate temporal features. The global inversion decoupling module is used to obtain hidden layer state features. The global linear attention feature extraction module is used to obtain deep representation features based on the global linear attention mechanism. The prediction module is used to obtain the battery capacity prediction set.

[0075] The remaining lifespan determination module 12 is used to determine the remaining lifespan of the lithium-ion battery to be predicted based on the battery capacity prediction set and the preset battery lifespan termination threshold.

[0076] In the above embodiments, the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence are obtained by the data acquisition module to be predicted. The preset time period includes multiple preset time windows, and the basic feature set includes multiple basic feature subsets corresponding to the multiple preset time windows. The basic feature subsets include: charging current integral area feature, charging voltage integral area feature, constant current charging time feature, discharge usable energy equivalent area feature, constant current discharging time feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature. Furthermore, the battery capacity prediction set acquisition module inputs the battery capacity sequence and basic feature set into the trained remaining life prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted. The remaining life prediction model includes: a local noise reduction and temporal reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local noise reduction and temporal reconstruction module performs noise reduction, smoothing, and temporal reconstruction processing to obtain multivariate temporal features, achieving high-frequency noise reduction of the basic feature set and extraction of sufficient low-frequency temporal features, and perfectly decoupling high-frequency noise reduction and low-frequency temporal feature extraction. By acquiring hidden layer state features through a global inversion decoupling module, dynamic hidden layer state features relative to the global time series can be fully extracted. By acquiring deep representation features based on a global linear attention mechanism through a global linear attention feature extraction module, deep co-evolution laws among multiple features can be reconstructed under a global perspective, and the computational power consumption is reduced to a linear level, improving the computational efficiency of the model. A battery capacity prediction set is obtained through a prediction module. Finally, the remaining lifespan determination module obtains the remaining lifespan of the lithium-ion battery to be predicted with high accuracy based on the battery capacity prediction set and a preset battery lifespan termination threshold.

[0077] Specific limitations regarding the lithium-ion battery remaining lifespan prediction device can be found in the limitations of the lithium-ion battery remaining lifespan prediction method described above, and will not be repeated here. Each module in the aforementioned server can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the corresponding operations of each module.

[0078] This invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement the lithium-ion battery remaining lifespan prediction method provided in this invention. For example, when the processor executes the computer program, it can implement... Figure 1 The technical solutions of any of the method embodiments shown are similar in implementation principle and technical effect, and will not be described again here.

[0079] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static random access memory (SRAM) and dynamic random access memory (DRAM), etc.

[0080] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0081] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely exemplary descriptions of the invention as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A method for predicting the remaining service life of a lithium-ion battery, characterized in that, include: Obtain the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence; wherein, the preset time period includes: multiple preset time windows, and the basic feature set includes: multiple basic feature subsets corresponding to the multiple preset time windows, and the basic feature subsets include: charging current integral area feature, charging voltage integral area feature, constant current charging time feature, discharge usable energy equivalent area feature, constant current discharging time feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature; The battery capacity sequence and the basic feature set are input into the trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted. The remaining lifespan prediction model includes: a local noise reduction and temporal reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local noise reduction and temporal reconstruction module is used to perform noise reduction, smoothing, and temporal reconstruction processing to obtain multivariate temporal features. The global inversion decoupling module is used to obtain hidden layer state features. The global linear attention feature extraction module is used to obtain deep representation features based on the global linear attention mechanism. The prediction module is used to obtain the battery capacity prediction set. The remaining lifespan of the lithium-ion battery to be predicted is determined based on the battery capacity prediction set and the preset battery life end threshold.

2. The method of claim 1, wherein, Before inputting the battery capacity sequence and the basic feature set into the trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted, the method further includes: Obtain a training set, which includes: training battery capacity sequences corresponding to multiple preset time windows, a training basic feature subset corresponding to each training battery capacity sequence, and a battery capacity label sequence for the next charge-discharge cycle corresponding to each preset time window. The training set is input into the initial remaining useful life prediction model for training. The weight parameters of the model are adjusted according to the preset loss function until the model converges, and the trained remaining useful life prediction model is obtained.

3. The method of claim 2, wherein, Before inputting the battery capacity sequence and the basic feature set into the trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted, the method further includes: The basic feature set is subjected to noise processing based on Gaussian noise.

4. The method of claim 3, wherein, The step of inputting the battery capacity sequence and the basic feature set into the trained remaining lifespan prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted includes: The battery capacity sequence and the basic feature set are input into the local noise reduction and time-series reconstruction module. The local noise reduction and time-series reconstruction module performs noise reduction, smoothing and time-series reconstruction processing to obtain the multivariate time-series features. The multivariate temporal features are input into the global inversion decoupling module, and global inversion decoupling processing is performed through the global inversion decoupling module to obtain the hidden layer state features; The hidden layer state features are input into the global linear attention feature extraction module, and the global linear attention feature extraction module performs global linear attention feature extraction processing to obtain the deep representation features; The deep representation features are input into the prediction module, and the prediction module obtains the battery capacity prediction set.

5. The method of claim 4, wherein, The local denoising and temporal reconstruction module includes: a denoising autoencoder, a gated loop unit, and a fusion unit; the battery capacity sequence and the basic feature set are input into the local denoising and temporal reconstruction module, and denoising, smoothing, and temporal reconstruction are performed by the local denoising and temporal reconstruction module to obtain the multivariate temporal features, including: The basic feature set is input into the denoising autoencoder, and the denoising autoencoder performs denoising and smoothing processing on the basic feature set to obtain the underlying physical features. The underlying physical features are input into the gated loop unit, and the gated loop unit performs temporal reconstruction processing on the underlying physical features to obtain the gated hidden state features. The underlying physical features, the gated hidden state features, and the battery capacity sequence are input into the fusion unit. The fusion unit then performs splicing and fusion processing on the underlying physical features, the gated hidden state features, and the battery capacity sequence to obtain the multivariate time-series features.

6. The method of claim 5, wherein, The global inversion decoupling module includes a feature transpose unit and a linear projection unit. The process of inputting the multivariate temporal features into the global inversion decoupling module and performing global inversion decoupling processing to obtain the hidden layer state features includes: The multivariate time series features are input into the feature transpose unit, and the multivariate time series features are transposed by the feature transpose unit to obtain the transposed multivariate time series features. The transposed multivariate temporal features are input into the linear projection unit, and linear projection processing is performed through the linear projection unit to obtain the hidden layer state features.

7. The method of claim 6, wherein, The global linear attention feature extraction module includes a multi-head global linear attention feature extraction unit and a feedforward neural network residual connection unit. The hidden layer state features are input to the global linear attention feature extraction module, and global linear attention feature extraction processing is performed through the global linear attention feature extraction module to obtain the deep representation features, including: The hidden layer state features are input into the multi-head global linear attention feature extraction unit, and the multi-head global linear attention feature extraction unit performs global linear attention feature extraction processing on the hidden layer state features to obtain global linear attention features. The global linear attention features are input into the residual connection unit of the feedforward neural network to obtain the deep representation features.

8. The method of claim 7, wherein, The prediction module includes: a shared linear prediction head, an inverse transpose and inverse normalization unit, and a multidimensional slicing unit. The deep representation features are input into the prediction module, and the battery capacity prediction set is obtained through the prediction module, including: The deep representation features are input into a shared linear prediction head, and the shared linear prediction head is used to perform prediction processing on the deep representation features to obtain intermediate prediction features of battery capacity. The intermediate battery capacity prediction features are input into the reverse transpose and reverse normalization unit. The reverse transpose and reverse normalization unit performs reverse transpose and reverse normalization processing on the intermediate battery capacity prediction features to obtain the initial battery capacity prediction set. The initial battery capacity prediction set is input into the multidimensional slicing unit, and the multidimensional slicing unit performs multidimensional slicing operations on the initial battery capacity prediction set to obtain the battery capacity prediction set.

9. A lithium-ion battery remaining useful lifetime prediction apparatus, characterized by, include: The data acquisition module is used to acquire the battery capacity sequence of the lithium-ion battery to be predicted within a preset time period and the basic feature set corresponding to the battery capacity sequence; wherein, the preset time period includes: multiple preset time windows, and the basic feature set includes: multiple basic feature subsets corresponding to the multiple preset time windows, and the basic feature subsets include: charging current integral area feature, charging voltage integral area feature, constant current charging time feature, discharge usable energy equivalent area feature, constant current discharging time feature, incremental capacity peak voltage feature, incremental capacity peak height feature, and incremental capacity main peak integral area feature; A battery capacity prediction set acquisition module is used to input the battery capacity sequence and the basic feature set into a trained remaining life prediction model to obtain the battery capacity prediction set of the lithium-ion battery to be predicted. The remaining life prediction model includes: a local denoising and temporal reconstruction module, a global inversion decoupling module, a global linear attention feature extraction module, and a prediction module. The local denoising and temporal reconstruction module is used to perform denoising smoothing and temporal reconstruction processing to obtain multivariate temporal features. The global inversion decoupling module is used to obtain hidden layer state features. The global linear attention feature extraction module is used to obtain deep representation features based on a global linear attention mechanism. The prediction module is used to obtain the battery capacity prediction set. The remaining lifespan determination module is used to determine the remaining lifespan of the lithium-ion battery to be predicted based on the battery capacity prediction set and the preset battery lifespan termination threshold.

10. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the lithium-ion battery remaining life prediction method according to any one of claims 1 to 8.

Images