Lithium battery residual life prediction method and related device

Abstract

The invention discloses a lithium battery residual life prediction method and a related device, and relates to the technical field of lithium battery residual life prediction. Calling an intrinsic mode function data sequence prediction model and a residual data sequence prediction model to respectively process an intrinsic mode function data sequence and a residual data sequence obtained by decomposing the original battery capacity degradation data of the to-be-predicted lithium battery; the intrinsic mode function data prediction sequence and the residual data prediction sequence are obtained and fused to obtain the battery capacity degradation prediction sequence, and the original battery capacity degradation data are decomposed to improve the prediction efficiency and the prediction precision. And calling a lithium battery residual life prediction model based on a Wiener process to correct the fused battery capacity degradation prediction sequence to obtain a final battery residual life prediction result with further improved prediction precision. According to the method, the data driving model is combined with the physical model based on the Wiener process to predict the residual life of the lithium battery, so that the RUL prediction precision of the lithium battery is greatly improved.

Description

Technical Field

[0001] This application relates to the field of lithium battery remaining life prediction technology, and in particular to a lithium battery remaining life prediction method and related apparatus. Background Technology

[0002] With the continuous improvement of electronic system integration, lithium batteries are widely used in electric vehicles, energy storage systems, portable electronic devices, and other fields. Their lifespan degradation directly affects the safety, economy, and sustainability of these devices. As a core energy storage component, the remaining useful life (RUL) prediction of lithium batteries has become a key technology for ensuring system reliability. Accurate RUL prediction can provide core data support for online monitoring and health management, optimize production processes and material design, and reduce operation and maintenance costs through condition-based maintenance strategies.

[0003] Existing lithium battery RUL prediction technologies are mainly divided into two categories: physical model-driven and data-driven. Physical models are generally based on electrochemical mechanisms. While they can reflect the internal reaction processes of lithium batteries, they require a large amount of prior knowledge and involve complex parameter identification, making them difficult to adapt to dynamic operating conditions and environmental changes. For example, actual conditions such as temperature fluctuations and differences in charge / discharge rates can significantly affect model accuracy, leading to accumulated prediction errors and reduced prediction precision. Data-driven models employ traditional machine learning methods (such as support vector machines and random forests) that rely on manual feature extraction. They have limited ability to fuse multi-source heterogeneous data and are prone to overfitting when the sample size is insufficient, thus reducing the model's prediction accuracy.

[0004] In summary, the existing lithium battery RUL prediction technology has low prediction accuracy and cannot meet the demand for higher prediction accuracy. Summary of the Invention

[0005] In view of the above problems, this application provides a lithium battery life prediction method and related apparatus to improve the accuracy of lithium battery remaining life prediction. The specific solution is as follows:

[0006] The first aspect of this application provides a method for predicting the remaining lifespan of a lithium battery, including:

[0007] The original battery capacity degradation sequence of the lithium battery to be predicted is obtained and decomposed to obtain the intrinsic mode function data sequence and the residual data sequence. The original battery capacity degradation sequence contains the maximum discharge capacity data of the lithium battery to be predicted in each charge-discharge cycle.

[0008] The intrinsic mode function (IMF) data sequence prediction model is called to process the IMF data sequence to obtain the IMF data prediction sequence; the pre-trained residual data sequence prediction model is called to process the residual data sequence to obtain the residual data prediction sequence.

[0009] The predicted values ​​of the intrinsic mode function data prediction sequence and the residual data prediction sequence belonging to the same cycle number are added together to obtain the fused battery capacity degradation prediction sequence.

[0010] The fused battery capacity degradation prediction sequence is processed by calling the lithium battery remaining life prediction model based on the Wiener process to obtain the corrected battery remaining life prediction results.

[0011] In one possible implementation, a pre-trained intrinsic mode function (IMF) data sequence prediction model is invoked to process the IMF data sequence, resulting in an IMF data prediction sequence, including:

[0012] The intrinsic mode function data sequence is input into the temporal convolutional network layer of the intrinsic mode function data sequence prediction model to obtain multi-scale local temporal features;

[0013] Calculate the global dependency weights between local temporal features, and enhance the local temporal features based on the global dependency weights to obtain higher-order temporal features;

[0014] High-order temporal features are input into a fully connected layer to obtain the intrinsic mode function data prediction sequence.

[0015] In one possible implementation, a pre-trained residual data sequence prediction model is invoked to process the residual data sequence, resulting in a residual data prediction sequence, including:

[0016] The Transformer network in the residual data sequence prediction model is used to process the residual data sequence into a residual data input vector, which is an input vector after embedding location information.

[0017] Determine the similarity between a single residual data input vector and each other residual data input vector other than the single residual data input vector, obtain the similarity set corresponding to each of the different single residual data input vectors, and calculate the new residual data input vector for each of the different single residual data input vectors based on the similarity set corresponding to each of the different single residual data input vectors.

[0018] After performing a nonlinear transformation on the new residual data input vector with different single residual data input vectors, it is input into the linear layer for feature mapping to obtain the residual data prediction sequence.

[0019] In one possible implementation, a Wiener process-based lithium-ion battery remaining life prediction model is invoked to process the fused battery capacity degradation prediction sequence, resulting in a corrected battery remaining life prediction result, including:

[0020] The fused battery capacity degradation prediction sequence is processed by maximum likelihood estimation to obtain the model update parameters of the lithium battery remaining life prediction model. The model update parameters include at least the drift parameter and the diffusion coefficient.

[0021] The drift parameters and diffusion coefficient of the lithium battery remaining life prediction model are updated using the Bayesian algorithm to obtain a new lithium battery remaining life prediction model.

[0022] The uncertainty of the fused battery capacity degradation prediction sequence is quantified by calling the new lithium battery remaining life prediction model to obtain the corrected battery remaining life prediction results.

[0023] One possible implementation also includes:

[0024] Error evaluation parameters are determined based on the fused battery capacity degradation prediction sequence. The error evaluation parameters include at least one of the following: relative error, root mean square error, and mean absolute error.

[0025] The model evaluation results of the intrinsic mode function data sequence prediction model and the model assessment results of the residual data sequence prediction model are obtained based on the error evaluation parameters.

[0026] In one possible implementation, the original battery capacity degradation sequence of the lithium battery to be predicted is obtained and decomposed to obtain an intrinsic mode function data sequence and a residual data sequence, including:

[0027] Obtain the original battery capacity degradation sequence of the lithium battery to be predicted;

[0028] The original battery capacity degradation sequence is subjected to fully integrated empirical decomposition and adaptive noise processing to obtain multiple intrinsic mode function data sequences and a residual data sequence.

[0029] A second aspect of this application provides a lithium battery remaining life prediction device, comprising:

[0030] The decomposition unit is used to obtain and decompose the original battery capacity degradation sequence of the lithium battery to be predicted, and obtain the intrinsic mode function data sequence and residual data sequence. The original battery capacity degradation sequence contains the maximum discharge capacity data of the lithium battery to be predicted in each charge and discharge cycle.

[0031] The prediction unit is used to call a pre-trained intrinsic mode function data sequence prediction model to process the intrinsic mode function data sequence and obtain the intrinsic mode function data prediction sequence; and to call a pre-trained residual data sequence prediction model to process the residual data sequence and obtain the residual data prediction sequence.

[0032] The fusion unit is used to add the predicted values ​​of the intrinsic mode function data prediction sequence and the residual data prediction sequence that belong to the same cycle number to obtain the fused battery capacity degradation prediction sequence.

[0033] The correction unit is used to call the lithium battery remaining life prediction model based on the Wiener process to process the fused battery capacity degradation prediction sequence and obtain the corrected battery remaining life prediction result.

[0034] A third aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the lithium battery remaining life prediction method described in the first aspect or any implementation thereof.

[0035] A fourth aspect of this application provides a lithium battery remaining life prediction device, comprising at least one processor and a memory connected to the processor, wherein:

[0036] Memory is used to store computer programs;

[0037] The processor is used to execute a computer program to enable the lithium battery remaining life prediction device to implement the lithium battery remaining life prediction method of the first aspect or any implementation thereof.

[0038] The fifth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the lithium battery remaining life prediction method described in the first aspect or any implementation thereof.

[0039] By employing the aforementioned technical solution, the lithium battery remaining life prediction method and related apparatus provided in this application utilize two data-driven models—an intrinsic mode function (IMF) data sequence prediction model and a residual data sequence prediction model—to process the IMF data sequence and residual data sequence obtained from the decomposition of the original battery capacity degradation data of the lithium battery to be predicted. These two data-driven models are then fused to obtain the battery capacity degradation prediction sequence. Decomposing the original battery capacity degradation data improves prediction efficiency and accuracy. Finally, a Wiener process-based lithium battery remaining life prediction model is used to correct the fused battery capacity degradation prediction sequence, resulting in a final battery remaining life prediction result with further improved accuracy. This application employs a data-driven model combined with a Wiener process-based physical model to predict the remaining life of lithium batteries, significantly improving the accuracy of lithium battery RUL prediction. Attached Figure Description

[0040] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0041] Figure 1 A flowchart illustrating the lithium battery remaining life prediction method provided in this application;

[0042] Figure 2 Box plot of the distribution of the number of iterations for filtering the data sequence of each intrinsic mode function of the fully adaptive noise empirical decomposition provided in this application;

[0043] Figure 3 Example figure showing the decomposition results of the original battery capacity degradation sequence of the lithium battery to be predicted provided in this application;

[0044] Figure 4 Example figure of the fused battery capacity degradation prediction sequence provided in this application;

[0045] Figure 5 A flowchart illustrating a lithium battery remaining life prediction method provided in this application;

[0046] Figure 6 Example diagram of the final remaining lifetime prediction results provided for this application;

[0047] Figure 7 A schematic diagram of a lithium battery remaining life prediction device provided in this application;

[0048] Figure 8 This is a schematic diagram of the structure of the lithium battery remaining life prediction device provided in this application. Detailed Implementation

[0049] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0050] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0051] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0052] Existing physical model-driven and data-driven lithium battery remaining life prediction technologies both suffer from low prediction accuracy. To address this issue, this application provides a lithium battery remaining life prediction method and related apparatus.

[0053] Optional, see Figure 1 This application provides a flowchart illustrating a method for predicting the remaining lifespan of lithium batteries.

[0054] like Figure 1 As shown, the method for predicting the remaining life of a lithium battery includes the following steps:

[0055] Step 101: Obtain and decompose the original battery capacity degradation sequence of the lithium battery to be predicted to obtain the intrinsic mode function data sequence and residual data sequence.

[0056] It should be noted that the original capacity degradation sequence of a lithium battery includes the maximum discharge capacity data for each charge-discharge cycle. The maximum discharge capacity data for each charge-discharge cycle represents the maximum amount of electricity that the lithium battery can stably release during repeated charge-discharge cycles. The original capacity degradation sequence of a lithium battery is essentially a nonlinear dynamic evolution process under the combined influence of the internal physicochemical processes of the lithium battery and external usage conditions. This sequence includes the following data characteristics:

[0057] The sequence exhibits several characteristics: continuity (recorded sequentially in chronological order), variable length (the lifespan of lithium batteries varies greatly, and the sequence length is determined by actual usage conditions), and multi-dimensional temporal sequence (in addition to capacity data, the sequence includes synchronous temporal data such as voltage curves, temperature, and SOC (State of Charge). Therefore, the original capacity degradation sequence of a lithium battery is a continuous time series of variable length.

[0058] For example, the initial capacity degradation sequence of a lithium battery can be represented as:

[0059] , ;

[0060] in, Let m be the maximum discharge capacity value of the i-th charge-discharge cycle, and m be the number of cycles.

[0061] The decomposition of the original battery capacity degradation sequence of lithium batteries includes, but is not limited to, the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) algorithm. This algorithm decomposes the original battery capacity degradation sequence into pure components of different frequencies by adaptively adding noise and performing multiple decomposition averaging. Specifically, it decomposes it into multiple intrinsic mode function (IMF) data sequences with different frequency and amplitude characteristics, and a residual data sequence. The residual data sequence is obtained by subtracting the multiple IMF data sequences from the original battery capacity degradation sequence and is used to characterize the error between the original battery capacity degradation sequence and the IMF data sequences.

[0062] Optionally, the original battery capacity degradation sequence of the lithium battery to be predicted is obtained, and the sequence is subjected to fully integrated empirical decomposition and adaptive noise processing to obtain multiple intrinsic mode function data sequences and a residual data sequence.

[0063] For example, the process of performing fully integrated empirical decomposition and adaptive noise processing on the original battery capacity degradation sequence of the lithium battery to be predicted can be as follows:

[0064] 1> The EMD (Empirical Mode Decomposition) algorithm is used to decompose the original battery capacity degradation sequence to obtain IMFs and residual terms. The formula can be expressed as follows:

[0065] ;

[0066] in, This can be the original battery capacity degradation sequence, or it can be the maximum discharge capacity data to be decomposed. For the k-th IMF component in the decomposed IMFs sequence, For the residual term, N is the total number of IMF components in the decomposition.

[0067] 2> Adaptive noise is added to the decomposed IMFs sequence, which can be expressed by the following formula:

[0068] ;

[0069] Where n represents adaptive noise, The IMFs sequence after introducing noise.

[0070] 3> Use the EMD algorithm to decompose the residual terms This process yields a new IMFs sequence and a residual data sequence. Adaptive noise is then added to the new IMFs sequence. This step is repeated until the residual term converges, resulting in the final IMFs sequence and residual data sequence. Therefore, multiple IMFs sequences and one residual data sequence are obtained.

[0071] For example, see Figure 2 The box plot provided in this application is an example of the distribution of the number of iterations for filtering the data sequence of each intrinsic mode function in the fully adaptive noise empirical decomposition.

[0072] like Figure 2 As shown, the horizontal axis represents the IMF components obtained from the decomposition: IMF1 to IMF7, and the vertical axis represents the number of IMF screening iterations during the decomposition process, ranging from 0 to 35 times.

[0073] Depend on Figure 2 It can be seen that the number of iterations of most IMFs is concentrated between 8 and 15, with the median of IMF3 to IMF6 being close to 10 to 12, showing a relatively stable distribution.

[0074] In summary, after obtaining the original battery capacity degradation sequence of the lithium battery in this step, the sequence is decomposed to obtain multiple intrinsic mode function data sequences and a residual data sequence. This process mainly transforms continuous time series data of indefinite length into fixed-length structural segments, resulting in battery capacity data that reflects the global characteristics of battery capacity. This adapts to the processing requirements of subsequent data-driven models, facilitating the model's accurate capture of local details and overall trends in the battery degradation process, and providing a rich and accurate data foundation for subsequent feature extraction and analysis.

[0075] For example, see Figure 3 Example figure of the decomposition results of the original battery capacity degradation sequence of the lithium battery to be predicted provided in this application.

[0076] like Figure 3 As shown, the lithium battery capacity degradation signal sequence is decomposed into 6 IMF components and a residual data sequence.

[0077] These six IMF components have different frequencies and amplitude characteristics. Further analysis of the degradation mechanisms of each IMF component at different frequencies can improve the accuracy of lithium battery remaining life prediction.

[0078] Step 102: Call the pre-trained intrinsic mode function data sequence prediction model to process the intrinsic mode function data sequence and obtain the intrinsic mode function data prediction sequence; call the pre-trained residual data sequence prediction model to process the residual data sequence and obtain the residual data prediction sequence.

[0079] It should be noted that the pre-trained intrinsic mode function (IMF) data sequence prediction model can specifically be an IMF data sequence prediction model based on an improved temporal convolutional network: the SA-TCN (Self-Attention based Temporal Convolutional Network) IMF data sequence prediction model. The training data for this model consists of IMF data sequences. These IMF data sequences are divided into training and test sets. An SA-TCN network architecture for predicting IMF data sequences is constructed. After training the model on the training set, the test set is used to validate and obtain the predicted IMF sequences.

[0080] Next, we will introduce the construction process of the intrinsic mode function data sequence prediction model:

[0081] Suppose the given input sequence is The predicted output is The relationship between the predicted output and the input sequence is as follows:

[0082] ;

[0083] From this formula, we can see that Only with the input sequence at time i and earlier Related to future input Irrelevant, the intrinsic mode function (IMF) data sequence prediction model aims to establish a mapping relationship between the input sequence and the output sequence, with the goal of making the actual output... and The error loss between them is minimized.

[0084] Therefore, dilated causal convolution is introduced into the TCN network. By exponentially expanding the dilation coefficient, the receptive range of the convolution kernel is increased layer by layer, effectively avoiding linear growth in network depth or convolution kernel size, efficiently controlling computational complexity, and maintaining a balance between model prediction performance and training speed. The formula for calculating dilated convolution can be:

[0085] ;

[0086] in, For sequence The convolution result of the i-th element; d is the filter, also known as the convolution kernel; d is the scaling factor, and when d=1 it is a standard causal convolution.

[0087] To alleviate the gradient vanishing and planning problems in deep network training, this model employs the concept of residual networks to simplify the training process. Assuming the input of the residual block is x and the output is o, the formula obtained through linear transformation and mapping via an activation function can be as follows:

[0088] ;

[0089] in, For activation function, Let x be the relational function. Since the residuals are never zero in practice, the stacked layers in a deep learning network can always learn new features, meaning that the learning performance of deep networks will not degrade.

[0090] Furthermore, a self-attention mechanism (SA mechanism) is introduced into the designed and constructed TCN model to assist its operation, thereby enhancing its ability to handle nonlinear features in time series prediction and improving its performance. In TCN, a pre-activated residual connection scheme is adopted, placing both Batch Normalization (BN) and the activation function before the extended causal convolution operation. Before the first extended convolution, Leaky ReLU is used as the activation function, which sacrifices the hard zero sparsity of the gradient, resulting in higher robustness during parameter optimization. The activation function formula is as follows:

[0091] ;

[0092] Embedding the Self-Attention (SA) mechanism into the TCN model allows for more flexible focus on key and effective features across different time series segments and adaptive learning, which is particularly important for analyzing complex lithium-ion battery degradation data. By effectively learning the relationships between different time steps within the sequence, the SA mechanism can more accurately capture changes in the lithium-ion battery degradation trajectory over different time periods. The SA mechanism uses Softmax dot product self-attention, as shown in the following formula:

[0093] ;

[0094] in, Let K be the query matrix and V be the key matrix. The feature dimension of each key is used for weight scaling and is normalized to the [0, 1] interval by Softmax.

[0095] In summary, the intrinsic mode function data sequence prediction model is mainly based on TCN modeling, using a network structure that combines residual blocks and DCC, and introducing the SA mechanism, thereby effectively improving the feature learning ability and robustness of the TCN model.

[0096] Next, we will introduce the training process of the intrinsic mode function data sequence prediction model:

[0097] The decomposed IMFs data sequences are divided into training and testing sets. The training set includes the number of charging cycles and the IMFs data. The number of charging cycles is used as the training sample. This process is the training process for the intrinsic mode function data sequence prediction model, which specifically includes:

[0098] Take the first 30% of the sample values ​​from the k decomposed data sequences IMF1 to IMF2. As a training set, k IMF data sequence prediction models TCN-1 to TCN-k are trained, where... This represents the nth data point in the IMF data sequence.

[0099] The IMF data sequence of the test set Substituting the values ​​into the corresponding TCN prediction model, we obtain the predicted new sequence of IMF data. .

[0100] The intrinsic mode function data sequence prediction model mainly consists of three layers: a temporal convolutional network layer, a self-attention mechanism layer, and a fully connected layer.

[0101] Optionally, the intrinsic mode function (IMF) data sequence is input into the temporal convolutional network layer of the IMF data sequence prediction model to obtain multi-scale local temporal features. The global dependency weights between the local temporal features are calculated, and the local temporal features are enhanced based on the global dependency weights to obtain higher-order temporal features. The higher-order temporal features are then input into the fully connected layer to obtain the IMF data prediction sequence.

[0102] Specifically, the temporal convolutional network layer expands its receptive field by using dilated causal convolution to capture long-term dependencies in the time series and obtain multi-scale local temporal features. The self-attention mechanism layer dynamically enhances the influence of important features by calculating the correlation weights between each time step in the sequence. Specifically, the input sequence undergoes a query-key-value matrix transformation, the attention weight matrix is ​​calculated, and after Softmax normalization, a weighted sum is obtained to obtain the enhanced feature representation: high-order temporal features. Finally, the high-order temporal features are input into the fully connected layer. Specifically, the weight matrix of the fully connected layer maps the high-order temporal features to a low-dimensional latent space and outputs the intrinsic mode function data prediction sequence.

[0103] It should be noted that calling the pre-trained residual data sequence prediction model can specifically be a Transformer-based residual data sequence prediction model. The training data for this model is the residual data sequence. The residual data sequence is divided into a training set and a test set, and a Transformer network architecture for predicting the residual data sequence is constructed. After training the model on the training set, the residual data sequence is verified and obtained using the test set.

[0104] Next, we will introduce the training and prediction process of the residual data sequence prediction model:

[0105] The residual data sequence is divided into a training set and a test set. The samples in the training set include the number of charging cycles and the residual data. The number of charging cycles is used as the training sample. This process is the training process of the residual data sequence prediction model, which specifically includes:

[0106] Take the first 30% of the sample values ​​from the decomposed residual data sequence. As a training set, in which This represents the nth data point in the residual data sequence. The training set is input into the Transformer network, which uses data embedding layers and linear layers to process the input training set, thereby training a residual data sequence prediction model and obtaining the residual data prediction sequence.

[0107] Substituting the residual data sequence from the test set into the residual data sequence prediction model yields the predicted new residual data sequence. .

[0108] In the residual data sequence prediction model, the data embedding layer can embed positional information into the input vector by encoding positional information, enabling the network to capture the order of the input sequence. Specifically, the data embedding layer can embed positional information into the input vector using the following formula:

[0109] ;

[0110] ;

[0111] Where PE represents the position encoding matrix; t represents the position of the position encoding; and k is a number from 0 to 1. Integer values ​​between -1 and s represent the dimension index of the encoded vector; s represents the dimension of each vector in the Transformer network.

[0112] Furthermore, multi-head self-attention is used to calculate the similarity between each input vector and all other input vectors, and the similarity is then weighted and summed to obtain a new representation for each input vector. The multi-head self-attention mechanism allows the network to simultaneously focus on multiple key regions, ignoring the distance between sequences. Its core process involves calculating the attention weights using the query vector matrix Q and the key vector matrix K, and then applying these weights to the value vector matrix V to obtain the overall weights and output. Specifically, the data embedding layer can use formulas to calculate the weight values ​​and output information for each input vector:

[0113] ;

[0114] ;

[0115] In this context, Q stands for Query, which is a vector matrix; K stands for Key, which is a key vector matrix; Q and K are used to calculate the weights of attention; and V stands for Value, which is a value vector matrix.

[0116] Q (Query), K (Key), and V (Value) are three vectors obtained from the input residual data sequence through three linear mapping layers, with dimensions of respectively. , and ; This represents the linear mapping parameters of the output.

[0117] Here, Concat represents the concatenation function, which concatenates all the output results of the head along the last dimension to form the final output; Let Q, K, and V represent the weight matrices corresponding to the Q, K, and V vectors of the i-th head, respectively.

[0118] ;

[0119] Where h is the number of self-attention heads, usually set to 8, i.e., h=8; and Then Softmax represents the activation function. It is the transpose of K. Used for normalization. Normalization helps stabilize gradients during training, thus avoiding the vanishing gradient problem.

[0120] Furthermore, a feedforward network is used to perform a nonlinear transformation on the vector after self-attention processing. This feedforward neural network can consist of two fully connected layers and a ReLU activation function. Specifically, the data embedding layer can perform the nonlinear transformation using the following formula:

[0121] ;

[0122] Where x represents vector data of dimension s from the input data; , These are bias parameters; , Let be the parameter matrix of the linear mapping.

[0123] Finally, based on the nonlinear transformation results obtained from the data embedding layer, the linear layer of the Transformer maps them to obtain the prediction results of the residual data sequence, which indicates the remaining lifespan of the battery.

[0124] Optionally, the Transformer network in the residual data sequence prediction model is used to process the residual data sequence into residual data input vectors. The residual data input vectors are input vectors after embedding position information. The similarity between a single residual data input vector and each other residual data input vector is determined, resulting in a similarity set corresponding to each single residual data input vector. Based on the similarity sets corresponding to each single residual data input vector, new residual data input vectors for each single residual data input vector are calculated. After nonlinear transformation, the new residual data input vectors are input into a linear layer for feature mapping to obtain the residual data prediction sequence.

[0125] In summary, this step employs a pre-trained intrinsic mode function (IMF) data sequence prediction model and a pre-trained residual data sequence prediction model to make predictions based on the IMF data sequence and residual data sequence, respectively, resulting in IMF data prediction sequences and residual data prediction sequences. This method of decomposing and using separate prediction models can improve the prediction accuracy of the remaining lifespan of lithium batteries.

[0126] Step 103: Add the predicted values ​​of the intrinsic mode function data prediction sequence and the residual data prediction sequence that belong to the same cycle number to obtain the fused battery capacity degradation prediction sequence.

[0127] Optionally, the predicted values ​​belonging to the same cycle number in the intrinsic mode function data prediction sequence and the residual data prediction sequence are added together to obtain the fused battery capacity degradation prediction sequence.

[0128] For example, suppose the intrinsic mode function data prediction sequence is... Let the residual data prediction sequence be... The corresponding cycle counts in the two sequences are added together to generate a fused battery capacity degradation prediction sequence, which can also be called the maximum discharge capacity data sequence of lithium-ion power batteries. Finally, based on the maximum discharge capacity data sequence of lithium-ion power batteries, the future capacity aging trajectory of lithium-ion power batteries is plotted to obtain the prediction results of the remaining life of lithium-ion power batteries.

[0129] After obtaining the fused battery capacity degradation prediction sequence, the process also includes an evaluation step for the two data-driven models mentioned above: the pre-trained intrinsic mode function data sequence prediction model and the pre-trained residual data sequence prediction model.

[0130] Optionally, error evaluation parameters are determined based on the fused battery capacity degradation prediction sequence. The error evaluation parameters include at least one of relative error, root mean square error, and mean absolute error. Based on the error evaluation parameters, the model evaluation results of the intrinsic mode function data sequence prediction model and the model evaluation results of the residual data sequence prediction model are obtained.

[0131] For example, error evaluation parameters are determined based on the fused battery capacity degradation prediction sequence, and then error analysis is performed on the intrinsic mode function data sequence prediction model and the residual data sequence prediction model based on the error evaluation parameters.

[0132] The formula for calculating relative error is as follows:

[0133] ;

[0134] The formula for calculating the root mean square error is as follows:

[0135] ;

[0136] The formula for calculating the mean absolute error is as follows:

[0137] .

[0138] The purpose of using error evaluation parameters to evaluate the intrinsic mode function data sequence prediction model and the residual data sequence prediction model is mainly to verify the accuracy of these two data-driven models.

[0139] For example, see Figure 4 Example diagram of the fused battery capacity degradation prediction sequence provided in this application.

[0140] This figure shows the evaluation of the data-driven model's intrinsic mode function data sequence prediction model and the residual data sequence prediction model. The evaluation figure includes a line graph and an error histogram.

[0141] The line graph compares the trend of predicted battery capacity with the number of cycles. As the number of cycles increases from 0 to 180, the capacity decreases from 1.8Ah to about 1.3Ah, and the trends are basically consistent.

[0142] The error histogram shows the distribution of prediction errors, with the horizontal axis representing the error value and the vertical axis representing the frequency. The errors are mostly concentrated around 0, with a few larger errors distributed within the range of ±0.05Ah.

[0143] Figure 4The paper also shows the values ​​of the model's key evaluation metrics, including RMSE (root mean square error) of 0.015661 and MAE (mean absolute error) of 0.012421, both of which are relatively small, indicating that the final fused battery capacity degradation prediction sequence has high prediction accuracy and good model prediction performance.

[0144] R 2 The coefficient of determination is 0.95924, which is close to 1, indicating that the model has an excellent fit.

[0145] In summary, the intrinsic mode function data sequence prediction model and residual data sequence prediction model provided in this application accurately predict the decay trend of lithium battery capacity with the number of cycles, with concentrated error distribution and low overall error level, and good prediction performance.

[0146] Step 104: Call the lithium battery remaining life prediction model based on Wiener process to process the fused battery capacity degradation prediction sequence and obtain the corrected battery remaining life prediction results.

[0147] It should be noted that, considering the battery degradation pattern and the uncertainty of degradation, this application adopts the method of establishing a lithium battery remaining life prediction model based on the Wiener process to predict the battery remaining life, and the output can be the RUL probability density distribution.

[0148] Optionally, the fused battery capacity degradation prediction sequence is subjected to maximum likelihood estimation to obtain the model update parameters of the lithium battery remaining life prediction model. The model update parameters include at least the drift parameter and the diffusion coefficient. The drift parameter and the diffusion coefficient of the lithium battery remaining life prediction model are updated using a Bayesian algorithm to obtain a new lithium battery remaining life prediction model. The new lithium battery remaining life prediction model is then used to perform uncertainty quantification on the fused battery capacity degradation prediction sequence to obtain the corrected battery remaining life prediction result.

[0149] Specifically, the fused battery capacity degradation prediction sequence is used as the observation value of the physical model of the lithium battery remaining life prediction model based on the Wiener process. The parameters of the physical model are estimated by the maximum likelihood estimation method, and the parameters of the physical model are dynamically updated by the Bayesian algorithm to obtain a new lithium battery remaining life prediction model. The fused battery capacity degradation prediction sequence is then processed by the new lithium battery remaining life prediction model to obtain the corrected battery remaining life prediction result.

[0150] For example, suppose the observed values ​​of the physical model for predicting the remaining life of a lithium battery at time t are:

[0151] ;

[0152] in, and They are independent of each other; The observation values ​​at the starting point; The drift coefficient, It is used to characterize the heterogeneity of individual devices of the same type; The diffusion coefficient is denoted as . For It is a nonlinear function with parameters, used to characterize the linear features of lithium battery capacity degradation.

[0153] definition Substituting into the above equation, we get:

[0154] ;

[0155] Because the degradation processes of lithium batteries are independent of each other, in the observed values The following is about building Maximum likelihood estimation function:

[0156] ;

[0157] Find the expression about the above. Taking the first-order partial derivative of and setting it equal to 0, we get... The maximum likelihood estimate is:

[0158] ;

[0159] The initial model parameters are estimated offline using the maximum likelihood estimation method, yielding the profile likelihood function of the maximum likelihood estimate:

[0160] ;

[0161] Find the expression in the above formula. The partial derivatives are used to obtain the maximum likelihood estimate, which is then substituted into... The maximum likelihood estimation is obtained .

[0162] Then, the drift parameters of the nonlinear Wiener process are updated using the Bayesian algorithm. ,make express Historical state detection information of lithium battery degradation at all times, Time drift coefficient The posterior estimate is:

[0163] ;

[0164] in, for Time drift coefficient Posterior distribution, ; For a known sequence of observations and drift coefficient under conditions The distribution function.

[0165] According to the properties of the Wiener process, Following a normal distribution, substituting into the above equation, the drift coefficient can be derived. exist Parameter estimates at time:

[0166] ;

[0167] exist Remaining lifespan at any given moment for: ;

[0168] Given the drift coefficient Under these conditions, remaining life The conditional probability density function is:

[0169] ;

[0170] Finally, the remaining lifetime is calculated using the law of total probability. The probability density function:

[0171] ;

[0172] It should be noted that this application uses some lithium battery accelerated aging data measured at 24°C from the NASA database for example demonstration, and the selected NASA database lithium battery number is B00005 to verify the model of the invention.

[0173] This application constructs a comparative experimental framework to evaluate the performance of the proposed method, selecting single TCN and Transformer as comparison algorithms. The experiments use the NASA database B00005 lithium battery maximum discharge capacity dataset. The experimental results are shown in Table 1. The proposed model achieves an RMSE of 0.015661, significantly lower than the prediction results of single TCN (0.023078) and Transformer (0.023078). The proposed method achieves an R² value of 0.96761, higher than the other two comparison algorithms. The comprehensive validation of multiple performance indicators demonstrates that the ensemble method proposed in this application exhibits excellent performance and stability in the task of predicting the remaining life of lithium batteries.

[0174] Table 1

[0175]

[0176] Furthermore, the average error of the remaining lifetime prediction results in this application was calculated.

[0177] Table 2

[0178]

[0179] Clearly, as shown in Table 2, the remaining lifetime prediction accuracy of the model in this application is far higher than that of a single model. Therefore, the battery remaining lifetime prediction method using a combination of numerical and analog models in this application meets the requirement for higher prediction accuracy.

[0180] In summary, the lithium battery remaining life prediction method provided in this application utilizes two data-driven models—an intrinsic mode function (IMF) data sequence prediction model and a residual data sequence prediction model—to process the IMF data sequence and residual data sequence obtained from the decomposition of the original battery capacity degradation data of the lithium battery to be predicted. These two data-driven models are then fused to obtain the battery capacity degradation prediction sequence. Decomposing the original battery capacity degradation data improves prediction efficiency and accuracy. Finally, a Wiener process-based lithium battery remaining life prediction model is used to correct the fused battery capacity degradation prediction sequence, resulting in a final battery remaining life prediction result with further improved accuracy. This application employs a data-driven model combined with a Wiener process-based physical model to predict the remaining life of lithium batteries, significantly improving the accuracy of lithium battery RUL prediction.

[0181] For example, see Figure 5 A flowchart illustrating a method for predicting the remaining lifespan of lithium batteries provided in this application.

[0182] First, the original battery capacity degradation sequence is obtained. The original battery capacity degradation sequence is decomposed using CEEMDAN to obtain the iMFs data sequence and the residual data sequence. Then, the iMFs data sequence and the residual data sequence are further processed by sliding window segmentation to establish the input / output mapping relationship.

[0183] Then, the SA-TCN battery capacity degradation model based on IMFs data is used to process the IMFs data sequence to obtain the IMFs prediction sequence. The Transformer battery degradation model based on residual data is then used to process the residual data sequence to output the residual data prediction sequence. The IMFs prediction sequence and the residual data prediction sequence are fused to obtain the remaining life prediction result of the lithium battery. Considering the battery degradation law and prediction uncertainty, a nonlinear Wiener process lithium battery remaining life prediction model based on an exponential function is established. The maximum likelihood estimation method is used to estimate the model parameters offline at the initial time. A Bayesian algorithm is used to update the model parameters of the Wiener process lithium battery remaining life prediction model online, specifically the drift coefficient parameters, to achieve iterative updates. Finally, the probability density function of the remaining life of the lithium battery is output.

[0184] For example, see Figure 6 Example diagram of the final remaining lifetime prediction results provided in this application.

[0185] The X-axis represents the starting time point for predicting the remaining life of the lithium battery, with a value range of 108 to 168. The Y-axis represents the predicted remaining usable life of the battery, in units of time or cycle count. The Z-axis represents the probability density of different remaining lifespans, with a higher peak value indicating a greater likelihood of that remaining lifespan.

[0186] Among the multiple colored curves, each curve corresponds to a prediction starting point, showing the probability distribution of remaining lifespan from that starting point. The earlier the starting point, that is, the larger the X-axis value, the higher and more concentrated the probability distribution peak, such as the orange curve, indicating that the early prediction has a higher confidence in the remaining lifespan; the later the starting point, that is, the smaller the X-axis value, the more dispersed the distribution, such as the blue and green curves, reflecting the increased uncertainty of later predictions.

[0187] This figure illustrates the quantified probability of remaining battery life at different prediction starting points.

[0188] In summary, the lithium battery remaining life prediction method provided in this application proposes a numerical simulation-based remaining life prediction method, achieving a deep integration of deep learning and nonlinear stochastic degradation modeling. This provides strong support for predictive maintenance, task planning optimization, and resource allocation decisions. The main contributions are as follows:

[0189] This application transforms continuous time series data of varying lengths into fixed-length structural segments to adapt to model processing requirements and capture local dynamic features, thereby enabling online prediction of power battery life.

[0190] To address the difficulty in quantifying uncertainty in deep learning-based remaining lifetime prediction methods, a remaining lifetime prediction method based on a nonlinear Wiener process with an exponential function is proposed. The maximum likelihood estimation method is used to estimate the parameters of the battery physical model, and the Bayesian algorithm is used to dynamically update the model parameters, thereby achieving battery remaining lifetime prediction with digital-analytical linkage.

[0191] By obtaining the original capacity sequence of the battery, the original capacity sequence can be preprocessed according to the CEEMDAN algorithm to obtain battery capacity data that reflects the global characteristics of battery capacity, accurately capture the local details and overall trends in the battery degradation process, and provide a rich and accurate data foundation for subsequent feature extraction and analysis.

[0192] A deep learning method is proposed to predict the remaining lifespan of lithium-ion power batteries. Since the IMFs data sequence and residual data sequence obtained by decomposition each have their own distribution, the deep learning method can capture their respective features, thereby improving the accuracy of the prediction.

[0193] Committed to integrating the advantages of model-based and data-driven methods, the present invention combines the Wiener process and the SA-TCN-Transformer network in a model-data linkage method, allowing the two methods to correct each other during the prediction process.

[0194] A method for predicting the remaining lifespan of lithium-ion power batteries by combining the CEEMDAN algorithm with numerical simulation is proposed, which can capture the phenomenon of capacity regeneration during the charging and discharging process.

[0195] The above describes a method for predicting the remaining life of a lithium battery provided in this application. The following describes the apparatus for performing the above-described method for predicting the remaining life of a lithium battery.

[0196] Please see Figure 7 , Figure 7 This is a schematic diagram of a lithium battery remaining life prediction device provided in this application. Figure 7 As shown, the device includes:

[0197] The system comprises a decomposition unit 10, a prediction unit 20, a fusion unit 30, and a correction unit 40; wherein:

[0198] The decomposition unit 10 is used to acquire and decompose the original battery capacity degradation sequence of the lithium battery to be predicted, and obtain the intrinsic mode function data sequence and the residual data sequence. The original battery capacity degradation sequence contains the maximum discharge capacity data of the lithium battery to be predicted in each charge and discharge cycle.

[0199] Prediction unit 20 is used to call a pre-trained intrinsic mode function data sequence prediction model to process the intrinsic mode function data sequence and obtain the intrinsic mode function data prediction sequence; and to call a pre-trained residual data sequence prediction model to process the residual data sequence and obtain the residual data prediction sequence.

[0200] The fusion unit 30 is used to add the predicted values ​​of the intrinsic mode function data prediction sequence and the residual data prediction sequence that belong to the same cycle number to obtain the fused battery capacity degradation prediction sequence.

[0201] The correction unit 40 is used to call the lithium battery remaining life prediction model based on the Wiener process to process the fused battery capacity degradation prediction sequence and obtain the corrected battery remaining life prediction result.

[0202] In one embodiment, the prediction unit 20 is specifically used for:

[0203] The intrinsic mode function data sequence is input into the temporal convolutional network layer of the intrinsic mode function data sequence prediction model to obtain multi-scale local temporal features;

[0204] Calculate the global dependency weights between local temporal features, and enhance the local temporal features based on the global dependency weights to obtain higher-order temporal features;

[0205] High-order temporal features are input into a fully connected layer to obtain the intrinsic mode function data prediction sequence.

[0206] In one embodiment, the prediction unit 20 is specifically used for:

[0207] The Transformer network in the residual data sequence prediction model is used to process the residual data sequence into a residual data input vector, which is an input vector after embedding location information.

[0208] Determine the similarity between a single residual data input vector and each other residual data input vector other than the single residual data input vector, obtain the similarity set corresponding to each of the different single residual data input vectors, and calculate the new residual data input vector for each of the different single residual data input vectors based on the similarity set corresponding to each of the different single residual data input vectors.

[0209] After performing a nonlinear transformation on the new residual data input vector with different single residual data input vectors, it is input into the linear layer for feature mapping to obtain the residual data prediction sequence.

[0210] In one embodiment, the correction unit 40 is specifically used for:

[0211] The fused battery capacity degradation prediction sequence is processed by maximum likelihood estimation to obtain the model update parameters of the lithium battery remaining life prediction model. The model update parameters include at least the drift parameter and the diffusion coefficient.

[0212] The drift parameters and diffusion coefficient of the lithium battery remaining life prediction model are updated using the Bayesian algorithm to obtain a new lithium battery remaining life prediction model.

[0213] The uncertainty of the fused battery capacity degradation prediction sequence is quantified by calling the new lithium battery remaining life prediction model to obtain the corrected battery remaining life prediction results.

[0214] In one embodiment, the lithium battery remaining life prediction device further includes a model evaluation unit; the model evaluation unit is specifically used for:

[0215] Error evaluation parameters are determined based on the fused battery capacity degradation prediction sequence. The error evaluation parameters include at least one of the following: relative error, root mean square error, and mean absolute error.

[0216] The model evaluation results of the intrinsic mode function data sequence prediction model and the model assessment results of the residual data sequence prediction model are obtained based on the error evaluation parameters.

[0217] In one embodiment, the decomposition unit 10 is specifically used for:

[0218] Obtain the original battery capacity degradation sequence of the lithium battery to be predicted;

[0219] The original battery capacity degradation sequence is subjected to fully integrated empirical decomposition and adaptive noise processing to obtain multiple intrinsic mode function data sequences and a residual data sequence.

[0220] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the lithium battery remaining life prediction methods provided in this application.

[0221] This application also provides a lithium battery remaining life prediction device in its embodiments. (Reference) Figure 8 The diagram illustrates a suitable structure for implementing the lithium battery remaining life prediction device provided in this application. The lithium battery remaining life prediction device in this application may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 8 The lithium battery remaining life prediction device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0222] like Figure 8 As shown, the lithium battery remaining life prediction device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. When the lithium battery remaining life prediction device is powered on, the RAM 603 also stores various programs and data required for the operation of the lithium battery remaining life prediction device. The processing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0223] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 608 including, for example, memory card, hard disk, etc.; and communication devices 609. Communication device 609 allows the lithium battery remaining life prediction device to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8 A lithium battery remaining life prediction device with various devices is shown; however, it should be understood that implementation or possession of all the devices shown is not required. More or fewer devices may be implemented alternatively.

[0224] This application also provides a computer-readable storage medium carrying one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the lithium battery remaining life prediction methods provided in this application.

[0225] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0226] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0227] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0228] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method of predicting the remaining life of a lithium battery, characterized by, The method comprises the following steps: obtain the original battery capacity degradation sequence of the lithium battery to be predicted and decompose it to obtain the intrinsic mode function data sequence and the residual data sequence, wherein the original battery capacity degradation sequence contains the maximum discharge capacity data of each charge and discharge cycle of the lithium battery to be predicted; call the pre-trained intrinsic mode function data sequence prediction model to process the intrinsic mode function data sequence to obtain the intrinsic mode function data prediction sequence; call the pre-trained residual data sequence prediction model to process the residual data sequence to obtain the residual data prediction sequence; add the predicted values in the intrinsic mode function data prediction sequence and the residual data prediction sequence that belong to the same cycle number to obtain the fused battery capacity degradation prediction sequence; call the lithium battery remaining life prediction model based on the Wiener process to process the fused battery capacity degradation prediction sequence to obtain the corrected battery remaining life prediction result.

2. The lithium battery remaining life prediction method of claim 1, wherein, The method comprises the following steps: input the intrinsic mode function data sequence into the time convolution network layer of the intrinsic mode function data sequence prediction model to obtain multi-scale local time sequence features; calculate the global dependence weight between the local time sequence features, enhance the local time sequence features based on the global dependence weight to obtain high-order time sequence features; input the high-order time sequence features into the full connection layer to obtain the intrinsic mode function data prediction sequence.

3. The lithium battery remaining life prediction method of claim 1, wherein, The method comprises the following steps: use the Transformer network in the residual data sequence prediction model to process the residual data sequence into a residual data input vector, wherein the residual data input vector is an input vector after embedding position information; determine the similarity between a single residual data input vector and each residual data input vector other than the single residual data input vector to obtain a similarity set corresponding to each single residual data input vector, and calculate a new residual data input vector of each single residual data input vector based on the similarity set corresponding to each single residual data input vector; perform nonlinear conversion on the new residual data input vector of each single residual data input vector and input it into a linear layer for feature mapping to obtain the residual data prediction sequence.

4. The lithium battery remaining life prediction method of claim 1, wherein The method comprises the following steps: perform maximum likelihood estimation processing on the fused battery capacity degradation prediction sequence to obtain model update parameters of the lithium battery remaining life prediction model, wherein the model update parameters at least include a drift parameter and a diffusion coefficient; update the drift parameter and the diffusion coefficient of the lithium battery remaining life prediction model using the Bayesian algorithm to obtain a new lithium battery remaining life prediction model; The new lithium battery remaining life prediction model is called to perform uncertainty quantification processing on the fused battery capacity degradation prediction sequence, to obtain the corrected battery remaining life prediction result.

5. The lithium battery remaining life prediction method of claim 1, wherein, Also comprising: An error evaluation parameter is determined according to the fused battery capacity degradation prediction sequence, and the error evaluation parameter at least includes one of relative error, root mean square error and mean absolute error; Based on the error evaluation parameter, a model evaluation result of the intrinsic mode function data sequence prediction model and a model evaluation result of the residual data sequence prediction model are obtained.

6. The lithium battery remaining life prediction method of claim 1, wherein The original battery capacity degradation sequence of the lithium battery to be predicted is obtained and decomposed to obtain the intrinsic mode function data sequence and the residual data sequence, comprising: An original battery capacity degradation sequence of a lithium battery to be predicted is obtained; The original battery capacity degradation sequence is subjected to complete ensemble empirical decomposition and adaptive noise processing to obtain a plurality of intrinsic mode function data sequences and a residual data sequence.

7. A lithium battery remaining life prediction device, characterized by, Comprising: An acquisition unit is configured to obtain an original battery capacity degradation sequence of a lithium battery to be predicted and decompose the original battery capacity degradation sequence to obtain an intrinsic mode function data sequence and a residual data sequence, wherein the original battery capacity degradation sequence contains maximum discharge capacity data in each charge and discharge cycle of the lithium battery to be predicted; A prediction unit is configured to call a pre-trained intrinsic mode function data sequence prediction model to process the intrinsic mode function data sequence to obtain an intrinsic mode function data prediction sequence, and call a pre-trained residual data sequence prediction model to process the residual data sequence to obtain a residual data prediction sequence; A fusion unit is configured to add predicted values belonging to the same cycle number in the intrinsic mode function data prediction sequence and the residual data prediction sequence to obtain a fused battery capacity degradation prediction sequence; A correction unit is configured to call a lithium battery remaining life prediction model based on a Wiener process to process the fused battery capacity degradation prediction sequence to obtain a corrected battery remaining life prediction result.

8. A computer program product, characterised in that, The computer readable instructions, when executed on an electronic device, cause the electronic device to implement the lithium battery remaining life prediction method of any one of claims 1 to 6.

9. A lithium battery remaining life prediction device characterized by comprising: Comprising at least one processor and a memory connected to the processor, wherein: The memory is configured to store a computer program; The processor is configured to execute the computer program to enable the lithium battery remaining life prediction device to implement the lithium battery remaining life prediction method of any one of claims 1 to 6.

10. A computer storage medium, characterized in that, The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the lithium battery remaining life prediction method of any one of claims 1 to 6.

Images