Method and device for predicting service life of energy storage battery

Abstract

The invention relates to the technical field of energy storage battery life prediction, and discloses an energy storage battery life prediction method and device, and the method comprises the steps: obtaining a multi-dimensional time sequence feature vector of an energy storage battery, extracting a local space-time feature based on the multi-dimensional time sequence feature vector of the energy storage battery, obtaining a feature vector fusing time sequence position information, and carrying out the prediction of the life of the energy storage battery. Performing time sequence dependency relation extraction on the feature vector fused with the time sequence position information to obtain a multi-dimensional context feature sequence, performing multi-time scale information fusion on the multi-dimensional context feature sequence to obtain a multi-source joint feature vector, and predicting the service life of the energy storage battery based on the multi-source joint feature vector. And obtaining an energy storage battery life prediction result. The method fully considers the space-time correlation between the characteristics of the energy storage battery, and improves the accuracy of the prediction result of the service life of the energy storage battery.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage battery life prediction, and particularly relates to an energy storage battery life prediction method and device. BACKGROUND

[0002] With large-scale deployment of energy storage systems in photovoltaic, wind power, electric transportation and power frequency modulation scenarios, accurate prediction of the life status of electrochemical batteries as core energy units puts higher requirements on the reliability, safety and economy of system operation.

[0003] However, the related energy storage battery life prediction method does not consider the spatio-temporal correlation between energy storage battery characteristics, resulting in low accuracy of energy storage battery life prediction results. SUMMARY

[0004] Therefore, the present application provides an energy storage battery life prediction method and device to solve the problem of low accuracy of energy storage battery life prediction results caused by the related energy storage battery life prediction method not considering the spatio-temporal correlation between energy storage battery characteristics.

[0005] In a first aspect, the present application provides an energy storage battery life prediction method, which comprises:

[0006] obtaining a multi-dimensional time series feature vector of an energy storage battery, extracting local spatio-temporal features based on the multi-dimensional time series feature vector of the energy storage battery, and obtaining a feature vector fused with time sequence position information;

[0007] extracting time sequence dependency relationship of the feature vector fused with time sequence position information, and obtaining a multi-dimensional context feature sequence;

[0008] performing multi-time scale information fusion on the multi-dimensional context feature sequence, and obtaining a multi-source joint feature vector;

[0009] predicting the life of the energy storage battery based on the multi-source joint feature vector, and obtaining an energy storage battery life prediction result.

[0010] The energy storage battery life prediction method provided in the embodiment realizes extraction of local space-time features by obtaining a multi-dimensional time sequence feature vector of the energy storage battery, extracting local space-time features based on the multi-dimensional time sequence feature vector of the energy storage battery, and obtaining a feature vector fusing time sequence position information, enhances the pertinence and robustness of feature expression, realizes extraction of deep time sequence dependency by extracting time sequence dependency of the feature vector fusing time sequence position information, obtains a multi-dimensional context feature sequence, enhances the perception ability of the energy storage battery life prediction result to the space-time correlation of features, realizes sufficient integration of short-term fluctuation features and long-term trend changes by performing multi-time scale information fusion on the multi-dimensional context feature sequence, and obtains a multi-source joint feature vector, realizes accurate prediction of the energy storage battery life by predicting the energy storage battery life based on the multi-source joint feature vector, fully considers the space-time correlation between energy storage battery features, and improves the accuracy of the energy storage battery life prediction result.

[0011] In an optional implementation, the local space-time features are extracted based on the multi-dimensional time sequence feature vector of the energy storage battery, and the feature vector fusing time sequence position information is obtained, including:

[0012] The multi-dimensional time sequence feature vector of the energy storage battery is expanded into a multi-dimensional time sequence feature map of the energy storage battery.

[0013] The multi-dimensional time sequence feature map of the energy storage battery is convoluted to obtain a multi-granularity local space-time feature map.

[0014] The multi-granularity local space-time feature map is positionally encoded to obtain the feature vector fusing time sequence position information.

[0015] The energy storage battery life prediction method provided in the embodiment realizes extraction of local space-time features by obtaining a multi-dimensional time sequence feature vector of the energy storage battery, extracting local space-time features based on the multi-dimensional time sequence feature vector of the energy storage battery, and obtaining a feature vector fusing time sequence position information, enhances the pertinence and robustness of feature expression, realizes extraction of deep time sequence dependency by extracting time sequence dependency of the feature vector fusing time sequence position information, obtains a multi-dimensional context feature sequence, enhances the perception ability of the energy storage battery life prediction result to the space-time correlation of features, realizes sufficient integration of short-term fluctuation features and long-term trend changes by performing multi-time scale information fusion on the multi-dimensional context feature sequence, and obtains a multi-source joint feature vector, realizes accurate prediction of the energy storage battery life by predicting the energy storage battery life based on the multi-source joint feature vector, fully considers the space-time correlation between energy storage battery features, and improves the accuracy of the energy storage battery life prediction result.

[0016] In an optional implementation, the local space-time features are extracted based on the multi-dimensional time sequence feature vector of the energy storage battery, and the feature vector fusing time sequence position information is obtained, including:

[0017] Obtain the time steps corresponding to the multi-granularity local spatiotemporal feature maps, and map the time steps corresponding to the multi-granularity local spatiotemporal feature maps into position encoding vectors;

[0018] The location encoding vector is concatenated element-wise with the multi-granularity local spatiotemporal feature map to obtain a feature vector that integrates temporal location information.

[0019] The energy storage battery lifetime prediction method provided in this embodiment obtains the time steps corresponding to multi-granularity local spatiotemporal feature maps, maps the time steps corresponding to the multi-granularity local spatiotemporal feature maps to location encoding vectors, and transforms the time steps corresponding to the multi-granularity features into quantized location encoding vectors. This provides a computable time reference for time series information fusion, lays the foundation for associating the temporal context of features of different granularities, and obtains a feature vector that fuses temporal and location information by concatenating the location encoding vector with the multi-granularity local spatiotemporal feature maps element by element. This allows the features to retain both local spatiotemporal patterns and explicit temporal and location information, enhancing the ability to model time sequence dependencies.

[0020] In one optional implementation, temporal dependencies are extracted from the feature vectors fused with temporal location information to obtain a multidimensional contextual feature sequence, including:

[0021] Attention enhancement is applied to the feature vector that integrates temporal location information to obtain the attention-enhanced feature vector;

[0022] Bidirectional state prediction is performed based on the attention-enhanced feature vectors to obtain the hidden states at each time step;

[0023] Calculate the contextual attention weights based on the hidden states at each time step;

[0024] The hidden states at each time step are weighted and summed based on the context attention weights to obtain the context feature vector;

[0025] The hidden states and context feature vectors at each time step are fused to obtain a multidimensional context feature sequence.

[0026] The energy storage battery life prediction method provided in the embodiment suppresses noise interference, improves the discriminability of features, performs bidirectional state prediction based on the attention-enhanced feature vectors, obtains the hidden states of each time step, captures the correlation between each time step and the previous and subsequent time steps, enriches the time sequence context information of the hidden states, calculates the context attention weight based on the hidden states of each time step, quantifies the contribution of each time step to the overall context, allocates a higher weight to the key time sequence information, performs weighted summation on the hidden states of each time step based on the context attention weight, obtains the context feature vector, the weighted summation aggregates the global key time sequence information, forms a compact feature representation containing the overall context, and fuses the hidden states of each time step and the context feature vector to obtain a multi-dimensional context feature sequence, so that each time step feature contains both local time sequence details and global context information, and the context correlation of the sequence is enhanced.

[0027] In an optional implementation, the multi-dimensional context feature sequence is subjected to multi-time scale information fusion to obtain a multi-source joint feature vector, including:

[0028] The multi-dimensional context feature sequence is subjected to multi-window division according to time scales to obtain energy storage battery life feature vectors of multiple time scales;

[0029] Contribution weight is calculated based on the energy storage battery life feature vectors of multiple time scales;

[0030] The multi-dimensional context feature sequence is subjected to weighted summation based on the contribution weight to obtain the multi-source joint feature vector.

[0031] The energy storage battery life prediction method provided in the embodiment subjects the multi-dimensional context feature sequence to multi-window division according to time scales to obtain energy storage battery life feature vectors of multiple time scales, the multi-window division covers life features of different time scales, captures multi-dimensional life-related patterns of short-term fluctuations and long-term trends, calculates contribution weight based on the energy storage battery life feature vectors of multiple time scales, quantifies the importance of features of different time scales to life evaluation, highlights key scale information, and subjects the multi-dimensional context feature sequence to weighted summation based on the contribution weight to obtain a multi-source joint feature vector, the weighted summation integrates multi-scale key features to form a comprehensive life feature representation that takes into account the importance of each time scale, and improves the integrity and discriminability of the multi-source joint feature vector.

[0032] In an optional implementation, the energy storage battery life is predicted based on the multi-source joint feature vector to obtain an energy storage battery life prediction result, including:

[0033] Attention weight is calculated based on the multi-source joint feature vector.

[0034] The multi-source joint feature vector is weighted and summed based on the attention weight to obtain a storage battery context feature vector;

[0035] The storage battery context feature vector is mapped to a target regression output space to obtain a storage battery life prediction result.

[0036] The storage battery life prediction method provided in the embodiment focuses on the core information in the multi-source features that is strongly related to the life, enhances the pertinence and relevance of the features, integrates the key multi-source information to form a more compact and discriminative context feature, improves the representation ability of the features, realizes the accurate conversion from the features to the life prediction value, and ensures that the output result meets the quantitative task requirements of the life prediction.

[0037] In a second aspect, the present application provides a storage battery life prediction device, which comprises:

[0038] A first extraction module is configured to obtain a storage battery multi-dimensional time sequence feature vector, extract a local space-time feature based on the storage battery multi-dimensional time sequence feature vector, and obtain a feature vector fused with time sequence position information;

[0039] A second extraction module is configured to extract a time sequence dependency relationship from the feature vector fused with the time sequence position information, and obtain a multi-dimensional context feature sequence;

[0040] A fusion module is configured to perform multi-time scale information fusion on the multi-dimensional context feature sequence, and obtain a multi-source joint feature vector;

[0041] A prediction module is configured to predict the life of the storage battery based on the multi-source joint feature vector, and obtain a storage battery life prediction result.

[0042] In a third aspect, the present application provides a computer device, which comprises a memory and a processor, the memory and the processor are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the storage battery life prediction method of the first aspect or any of the corresponding embodiments thereof.

[0043] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions, and the computer instructions are used to make the computer execute the storage battery life prediction method of the first aspect or any of the corresponding embodiments thereof.

[0044] In a fifth aspect, the present application provides a computer program product comprising computer instructions for causing a computer to perform the energy storage battery life prediction method of the first aspect above or any of its corresponding embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings required to be used in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings described below are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0046] Figure 1 is a flowchart of an energy storage battery life prediction method according to an embodiment of the present application (one);

[0047] Figure 2 is a flowchart of an energy storage battery life prediction method according to an embodiment of the present application (two);

[0048] Figure 3 is a flowchart of an energy storage battery life prediction method according to an embodiment of the present application (three);

[0049] Figure 4 is a flowchart of an energy storage battery life prediction method according to an embodiment of the present application (four);

[0050] Figure 5 is a flowchart of an energy storage battery life prediction method according to an embodiment of the present application (five);

[0051] Figure 6 is a structural block diagram of an energy storage battery life prediction device according to an embodiment of the present application;

[0052] Figure 7 is a hardware structure diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0054] Under the driving of energy structure transformation, energy storage systems play an increasingly important role in power systems, especially in new energy access, peak regulation and frequency modulation, and power grid stability guarantee. Among them, the performance degradation and life prediction of batteries, as the core of energy storage systems, directly affect the economy, safety and reliability of the system. Therefore, conducting high-precision battery life prediction research is not only one of the key technologies to realize the long-term stable operation of energy storage systems, but also has important significance for promoting green and low-carbon energy development.

[0055] Most related battery life prediction methods rely on statistical analysis, empirical modeling or deep learning models based on sequence learning, such as LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), and Transformer (a neural network architecture based on self-attention mechanism). Deep learning models have made significant achievements in battery life prediction, especially in capturing complex nonlinear relationships and long-term dependencies. Recurrent neural networks represented by LSTM and GRU have been widely used in life prediction tasks. They can effectively depict the degradation trend of battery performance over time through time series modeling. For example, by introducing capacity, internal resistance and other time series data to train the LSTM network, dynamic estimation of battery State of Health (SOH) and Remaining Useful Life (RUL) is achieved. In addition, the Transformer model has superior performance in modeling cross-step dependencies due to its powerful attention mechanism. Some studies have begun to introduce it into battery life prediction, especially in handling multivariate long sequence data, achieving better accuracy than traditional recurrent networks.

[0056] However, although the above deep learning models have obvious advantages in representing complex degradation behavior, they still have some limitations in practical applications. First, models such as LSTM and GRU are sensitive to input sequence length, which can easily cause gradient disappearance or memory loss in long sequence cases, limiting their ability to model the entire life cycle. Second, although Transformer has strong modeling ability, its high computational complexity and large number of parameters make it more demanding for edge deployment, especially in embedded energy storage devices. In addition, the above models usually use uniform processing methods for input features, ignoring the physical differences and non-equal contribution to life of different types of signals such as voltage, current and temperature, resulting in insufficient sensitivity of the model to key features, which affects the final prediction performance and interpretability.

[0057] Therefore, in recent years, the research trend gradually shifts from simple sequence modeling to deep network design that integrates structure optimization and attention mechanism enhancement, for example, by introducing a feature attention module to weight multi-source sensor data to highlight the role of key features, and some research based on graph neural network models the dynamic dependency structure between multiple variables to improve feature interaction modeling capability. The above methods all show that constructing a prediction model with spatiotemporal perception ability and feature selection mechanism is an effective way to improve the prediction accuracy and stability of battery life.

[0058] To solve the above technical problems, the embodiment of the present application provides a method for predicting the life of an energy storage battery. A spatiotemporal gating convolution feature extraction module encodes a multi-dimensional time series feature vector, extracts local spatial and temporal correlation features, and injects sequence position information through a position encoding mechanism to construct a feature vector that integrates time sequence position information. The fused features enhance the feature vector that integrates time sequence position information through a CBAM (Convolutional Block Attention Module) attention mechanism, and are further input into a bidirectional attention LSTM to extract deep time sequence dependency relationships and form a multi-dimensional context feature sequence. Based on an attention-guided multi-scale memory fusion module, short-term and long-term information interaction modeling is completed, a multi-source joint feature vector that takes into account global and local perception is constructed, and finally, the life prediction result of the energy storage battery is output through a feature fusion layer to realize life prediction and state regression of the target sequence, providing accurate basis for subsequent intelligent analysis.

[0059] By introducing a spatiotemporal gating convolution feature extraction module into the model structure, on the one hand, the time evolution process of battery life influencing factors can be modeled to extract context dependency relationships in the time dimension, and on the other hand, the spatial correlation between sensor features can be mined to enhance the selectivity and interaction ability between features. The spatiotemporal gating mechanism has the important function of dynamically adjusting information channels, can effectively filter redundant signals, and can strengthen the contribution of key features in degradation trend modeling, thereby achieving high-precision and strong-robustness prediction of battery life. At the same time, the CBAM network can dynamically select key features to better mine feature information on the time series and improve the accuracy of life prediction.

[0060] The embodiment of the present application provides a kind of energy storage battery life prediction method, it needs to be indicated that, the energy storage battery life prediction method provided in the embodiment of the present application, its execution subject can be the device of energy storage battery life prediction, the device of energy storage battery life prediction can be realized by software, hardware or software and hardware combination way becomes part or all of electronic equipment, wherein the electronic equipment can be server or terminal, wherein the server in the embodiment of the application can be a server, can also be the server cluster of being made of multiple servers, the terminal in the embodiment of the application can be smart phone, personal computer, tablet computer, wearable device and smart robot and other intelligent hardware equipment.The method embodiment is described below, and the execution subject is electronic equipment as an example.

[0061] According to the embodiment of the present application, an energy storage battery life prediction method is provided, and it should be noted that the steps shown in the flowchart can be executed in a computer system, such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order.

[0062] In the present embodiment, an energy storage battery life prediction method is provided, which can be used in the above-mentioned electronic device, Figure 1 is a flowchart of the energy storage battery life prediction method according to the embodiment of the present application, as Figure 1 shown, the flow includes the following steps:

[0063] Step S101, obtain the multi-dimensional time series feature vector of the energy storage battery, extract the local space-time feature based on the multi-dimensional time series feature vector of the energy storage battery, and obtain the feature vector fused with time sequence position information.

[0064] Specifically, the input data of multiple sensors (such as voltage, current, temperature, internal resistance) of the energy storage battery (i.e. the multi-dimensional time series feature vector of the energy storage battery) often has significant time sequence characteristics and spatial heterogeneity, in order to fully exploit these space-time feature information, a spatio-temporal gated convolutional feature extraction module is used to realize the multi-dimensional time series feature vector of the energy storage battery, the spatio-temporal gated convolutional feature extraction module uses the strong representation ability of convolutional neural network in space modeling, combined with the gating mechanism to dynamically adjust the importance of different dimension information, enhance the pertinence and robustness of feature expression.

[0065] Step S102, time sequence dependent relationship extraction is performed on the feature vector fused with time sequence position information, and a multi-dimensional context feature sequence is obtained.

[0066] In step S103, multi-time scale information fusion is performed on the multi-dimensional context feature sequence to obtain a multi-source joint feature vector.

[0067] Specifically, in order to fully integrate the short-term fluctuation characteristics and long-term trend changes implied in the multi-dimensional sequence of the energy storage battery, a multi-scale memory fusion module based on attention guidance is designed on the basis of the multi-dimensional context feature sequence, which realizes the interactive modeling of global and local dependent information of the sequence (i.e., multi-time scale information fusion). Through the combination of multi-scale window division, time perception embedding and attention mechanism, information of different time granularities is fused into a unified multi-source joint feature vector, which provides more context depth support for the prediction task.

[0068] Further, compared with the sequence model which can only rely on a single time granularity for modeling, the multi-scale memory fusion module based on attention guidance explicitly introduces a multi-scale structure and dynamically fuses its contribution with the help of the attention mechanism, effectively improving the diversity and discriminability of feature expression, and better adapting to the degradation law and fluctuation behavior of the energy storage system under different time scales, thereby providing more reliable data support for downstream state assessment and life prediction.

[0069] In step S104, the life of the energy storage battery is predicted based on the multi-source joint feature vector to obtain an energy storage battery life prediction result.

[0070] Specifically, the life of the energy storage battery is predicted based on the multi-source joint feature vector, and the energy storage battery life prediction result is obtained through the feature fusion layer, which realizes the life prediction and state regression of the target sequence, and provides accurate basis for subsequent intelligent analysis.

[0071] The energy storage battery life prediction method provided in this embodiment realizes the extraction of local spatio-temporal features by obtaining the multi-dimensional time sequence feature vector of the energy storage battery, extracting local spatio-temporal features based on the multi-dimensional time sequence feature vector of the energy storage battery, and obtaining the feature vector fused with time sequence position information, enhances the pertinence and robustness of feature expression, realizes the extraction of deep spatio-temporal dependence by extracting the time sequence dependence relationship of the feature vector fused with time sequence position information, and enhances the perception ability of the energy storage battery life prediction result to the spatio-temporal correlation of features, realizes the full integration of short-term fluctuation characteristics and long-term trend changes by performing multi-time scale information fusion on the multi-dimensional context feature sequence, and obtains a multi-source joint feature vector, realizes the accurate prediction of the life of the energy storage battery by predicting the life of the energy storage battery based on the multi-source joint feature vector, fully considers the spatio-temporal correlation between the features of the energy storage battery, and improves the accuracy of the energy storage battery life prediction result.

[0072] The application provides a method for predicting the service life of an energy storage battery, which can be used for the electronic device, Figure 2 is a flowchart of the method for predicting the service life of the energy storage battery according to an embodiment of the application, as Figure 2 shown, the flowchart comprises the following steps:

[0073] In step S201, a multi-dimensional time sequence feature vector of the energy storage battery is obtained, and local space-time features are extracted based on the multi-dimensional time sequence feature vector of the energy storage battery to obtain a feature vector fused with time sequence position information.

[0074] Specifically, the multi-dimensional time sequence feature vector of the energy storage battery is encoded by a space-time gating convolution feature extraction module, features related to local space and time are extracted, and sequence position information is injected through a position encoding mechanism to construct a feature vector fused with time sequence position information. The feature vector fused with time sequence position information not only encodes the spatial correlation of the original features, but also explicitly embeds the time position information, thereby providing a more optimal input expression for subsequent attention mechanisms and time sequence modeling.

[0075] Specifically, step S201 comprises the following steps.

[0076] In step S2011, the multi-dimensional time sequence feature vector of the energy storage battery is expanded into a multi-dimensional time sequence feature map of the energy storage battery.

[0077] Specifically, the expression of the multi-dimensional time sequence feature vector X1 of the energy storage battery is as follows:

[0078]

[0079] wherein T is a time step, and C is a sensor feature dimension.

[0080] Further, to simultaneously capture the spatial dependence between features and the temporal dependence between sequences, the multi-dimensional time sequence feature vector of the energy storage battery is first expanded into a three-dimensional structure to obtain a multi-dimensional time sequence feature map of the energy storage battery. The expression of the multi-dimensional time sequence feature map X2 of the energy storage battery is as follows:

[0081]

[0082] wherein B is a batch size.

[0083] In step S2012, the multi-dimensional time sequence feature map of the energy storage battery is convolved to obtain a multi-granularity local space-time feature map.

[0084] Specifically, the input tensor is scanned by a plurality of stacked 1D or 2D convolution kernels W to extract local space-time features of different granularities, thereby obtaining a multi-granularity local space-time feature map. The expression of the multi-granularity local space-time feature map is as follows: The expression is:

[0085]

[0086] Where * denotes convolution operation, σ is a non-linear activation function, l represents the number of convolutional layers, and W (l) b is the weight of the convolutional kernel in layer l. (l) is the bias term of the l-th layer, and L is the total number of convolutional layers.

[0087] Step S2013: Position encoding is performed on the multi-granularity local spatiotemporal feature map to obtain a feature vector that integrates temporal position information.

[0088] Specifically, in order to further enhance the temporal awareness of feature representation, multi-granularity local spatiotemporal feature maps will undergo position encoding. The position encoding mechanism is mainly used to inject position information of each time point in the sequence, enabling the model to capture the "temporal order".

[0089] In some optional implementations, step S2013 above includes:

[0090] Step a1: Obtain the time steps corresponding to the multi-granularity local spatiotemporal feature maps, and map the time steps corresponding to the multi-granularity local spatiotemporal feature maps into position encoding vectors.

[0091] Specifically, a sine / cosine function encoding method is used to map time step t into a position encoding vector; where the position encoding vector PE t The expression is:

[0092]

[0093] Furthermore, the sinusoidal representation of the position encoding vector PE (t,2i) Cosine representation of the position encoding vector PE (t，2i+1) The expressions are as follows:

[0094]

[0095] Where d is the feature dimension and i is the dimension index.

[0096] Step a2 involves concatenating the location encoding vector with the multi-granularity local spatiotemporal feature map element by element to obtain a feature vector that fuses temporal location information.

[0097] Specifically, the feature vector F that integrates temporal location information pos The expression is:

[0098]

[0099] Step S202, time sequence dependence relationship extraction is performed on the feature vector fused with time sequence position information, to obtain a multi-dimensional context feature sequence. For details, please refer to Figure 1 Step S102 of the embodiment shown will not be repeated here.

[0100] Step S203, multi-time scale information fusion is performed on the multi-dimensional context feature sequence, to obtain a multi-source joint feature vector. For details, please refer to Figure 1 Step S103 of the embodiment shown will not be repeated here.

[0101] Step S204, based on the multi-source joint feature vector, the energy storage battery life is predicted, to obtain an energy storage battery life prediction result. For details, please refer to Figure 1 Step S104 of the embodiment shown will not be repeated here.

[0102] The energy storage battery life prediction method provided in this embodiment expands the energy storage battery multi-dimensional time sequence feature vector into an energy storage battery multi-dimensional time sequence feature map, converts the implicit correlation between features into an explicit spatial structure, and lays a foundation for subsequent convolution operations, while retaining the time sequence continuity of the features. By performing convolution on the energy storage battery multi-dimensional time sequence feature map, a multi-granularity local spatio-temporal feature map is obtained, which comprehensively captures the local spatio-temporal patterns of different time spans and feature combinations, improves the feature abstraction and discriminability, and through position encoding of the multi-granularity local spatio-temporal feature map, a feature vector fused with time sequence position information is obtained, which restores the time sequence logic lost in the convolution processing, strengthens the long time sequence dependence, and unifies the time sequence benchmarks of multi-granularity features.

[0103] In this embodiment, an energy storage battery life prediction method is provided, which can be used in the electronic device described above, Figure 3 is a flowchart of the energy storage battery life prediction method according to the embodiment of the present application, as shown in Figure 3 The flowchart includes the following steps:

[0104] Step S301, an energy storage battery multi-dimensional time sequence feature vector is obtained, and based on the energy storage battery multi-dimensional time sequence feature vector, local spatio-temporal features are extracted, to obtain a feature vector fused with time sequence position information. For details, please refer to Figure 2 Step S201 of the embodiment shown will not be repeated here.

[0105] Step S302, time sequence dependence relationship extraction is performed on the feature vector fused with time sequence position information, to obtain a multi-dimensional context feature sequence.

[0106] Specifically, the fused features enhance the representation ability of key dimensions through the CBAM attention mechanism, and are further input into the bidirectional attention LSTM to extract deep time sequence dependence relationships, to form a multi-dimensional context feature sequence.

[0107] Further, the multi-dimensional context feature sequence output by the bidirectional attention LSTM not only retains the time sequence context information before and after, but also further highlights the contribution of key time points in the sequence to the target task, thereby constructing a high-expression semantic feature with context awareness capability, laying a solid foundation for subsequent memory fusion and multi-scale modeling.

[0108] Specifically, the above step S302 includes:

[0109] Step S3021, the feature vector fused with the time sequence position information is enhanced by attention to obtain an attention-enhanced feature vector.

[0110] Specifically, after the preliminary spatio-temporal convolution feature extraction and position encoding are completed, the feature vector fused with the time sequence position information still contains a large amount of redundant or low-contribution information. In order to further highlight the key dimensions and improve the network's perception of important time sequence and channel features, a convolution block attention module (CBAM) is introduced to adaptively enhance the fused features. CBAM is a lightweight attention mechanism that realizes feature selection by concatenating two sub-modules, namely a channel attention (Channel Attention) module and a spatial attention (Spatial Attention) module.

[0111] Further, the feature vector fused with the time sequence position information is pooled, and the feature vector fused with the time sequence position information after the pooling is fused to obtain a channel attention weight vector; wherein the expression of the channel attention weight vector is:

[0112]

[0113] Wherein, σ is an activation function, MLP represents a perception machine, which is used to pool the pooled context feature sequence, AvgPool is an average pooling, MaxPool is a global maximum pooling, M c (F pos ) is the channel attention weight vector, is the dimension of the channel attention weight vector.

[0114] Further, the channel attention weight vector and the feature vector fused with the time sequence position information are multiplied to obtain a channel feature map; wherein the expression of the channel feature map is:

[0115]

[0116] Wherein, F c is the channel feature map.

[0117] Further, the channel feature map is subjected to a pooling process, and the channel feature map after the pooling is subjected to a convolution operation to obtain a spatial attention weight vector; wherein the expression of the spatial attention weight vector is:

[0118]

[0119] wherein Mt(Fc) is the spatial attention weight vector, ConvlD is the convolution processing, the dimension of the spatial attention weight vector is

[0120] Further, the channel feature map and the spatial attention weight vector are subjected to a multiplication operation to obtain a feature vector after attention enhancement; wherein the expression of the feature vector after attention enhancement X3 is:

[0121]

[0122] Step S3022, based on the feature vector after attention enhancement, bidirectional state prediction is performed to obtain the hidden state of each time step.

[0123] Specifically, in order to model the time sequence dependent information, the feature vector after attention enhancement is input into a bidirectional attention LSTM module. Unlike LSTM, the bidirectional attention LSTM module encodes the feature vector after attention enhancement through two directions of forward and backward to obtain the hidden state of each time step, and introduces attention mechanism in the hidden state of each time step to highlight the historical state which contributes most to the prediction task.

[0124] Step S3023, based on the hidden state of each time step, a context attention weight is calculated.

[0125] Specifically, the CBAM output sequence is X3=[x1,x2,…,x T ], the hidden state of each time step is h t , and the context attention weight of each time step is calculated through a dot-product attention (Dot-Product Attention) similar to the Transformer, and the calculation formula of the context attention weight a t，j is:

[0126]

[0127] wherein W a is a learnable parameter, h j is the hidden state of the jth time step, and h k is the hidden state of the kth time step.

[0128] In step S3024, the hidden states of each time step are weighted and summed based on the context attention weight to obtain a context feature vector.

[0129] Specifically, the context feature vector c t is calculated according to the following formula:

[0130]

[0131] In step S3025, the hidden states of each time step and the context feature vector are fused to obtain a multi-dimensional context feature sequence.

[0132] Specifically, the expression of the multi-dimensional context feature sequence c is as follows:

[0133]

[0134] wherein tanh is an activation function, W h is a learnable weight matrix.

[0135] In step S303, the multi-dimensional context feature sequence is subjected to multi-time scale information fusion to obtain a multi-source joint feature vector. For details, please refer to step S203 of the embodiment shown in Figure 2 , which will not be described here again.

[0136] In step S304, the energy storage battery life is predicted based on the multi-source joint feature vector to obtain an energy storage battery life prediction result. For details, please refer to step S204 of the embodiment shown in Figure 1 , which will not be described here again.

[0137] The energy storage battery life prediction method provided in this embodiment suppresses noise interference by attention-enhancing the feature vector fused with time sequence position information, enhances the discriminability of the feature, performs bidirectional state prediction based on the attention-enhanced feature vector to obtain the hidden state of each time step, captures the correlation between each time step and the previous and subsequent time sequences, enriches the time sequence context information of the hidden state, calculates the context attention weight based on the hidden state of each time step, quantifies the contribution degree of each time step to the overall context, assigns a higher weight to the key time sequence information, weights and sums the hidden states of each time step based on the context attention weight to obtain a context feature vector, aggregates the global key time sequence information through the weighted sum to form a compact feature representation containing the overall context, and fuses the hidden states of each time step and the context feature vector to obtain a multi-dimensional context feature sequence, so that each time step feature contains both local time sequence details and global context information, and the context correlation of the sequence is enhanced.

[0138] The application provides a method for predicting the service life of an energy storage battery, which can be used for the electronic device, Figure 4 The method for predicting the service life of the energy storage battery according to the embodiment of the application is shown in a flowchart as Figure 4 The flowchart includes the following steps:

[0139] In step S401, a multi-dimensional time sequence feature vector of the energy storage battery is obtained, local space-time features are extracted based on the multi-dimensional time sequence feature vector of the energy storage battery, and a feature vector fused with time sequence position information is obtained. For details, refer to step S301 of the embodiment shown in Figure 3 which will not be repeated here.

[0140] In step S402, time sequence dependence relationship extraction is performed on the feature vector fused with time sequence position information, and a multi-dimensional context feature sequence is obtained. For details, refer to step S302 of the embodiment shown in Figure 3 which will not be repeated here.

[0141] In step S403, multi-time scale information fusion is performed on the multi-dimensional context feature sequence, and a multi-source joint feature vector is obtained.

[0142] Specifically, the above step S403 includes:

[0143] In step S4031, the multi-dimensional context feature sequence is divided into multiple windows according to the time scale, and multiple time scale energy storage battery life feature vectors are obtained.

[0144] Specifically, the multiple multi-dimensional context feature sequences are divided into multiple windows according to the time scale, the dependence structure in different length ranges is captured respectively, and three typical scales are set: a short-term window w s , a medium-term window w m and a long-term window w l , which correspond to the information extraction of local fluctuations, trend changes and long-term memory respectively. In each window, a sliding window operation and a shared parameter convolution or a gating unit are used to extract the corresponding segment representation (i.e., the multiple time scale energy storage battery life feature vectors); wherein the expression of the multiple time scale energy storage battery life feature vectors M (k) is as follows:

[0145]

[0146] wherein Encoder (k) is a feature extraction submodule for each time scale, which can be set as a lightweight convolution layer, a GRU unit or a Self-Attention block, and is output as a local representation under the scale

[0147] Step S4032, calculating the contribution weight based on the energy storage battery life feature vectors of multiple time scales.

[0148] Specifically, in order to effectively integrate the semantic differences and relative importance between different time scales, a scale attention mechanism based on context vectors (Scale Attention) is introduced to automatically learn the contribution weight (i.e. contribution weight) of the energy storage battery life feature vectors of multiple time scales to the current prediction task. For each time step t, the contribution weight The calculation formula is as follows:

[0149]

[0150] Wherein, u T is the attention projection vector, b m , W m is the bias term.

[0151] Step S4033, weighted sum of the multi-dimensional context feature sequence based on the contribution weight, to obtain the multi-source joint feature vector.

[0152] Specifically, based on the contribution weight, the local representation of all scales (i.e. multi-dimensional context feature sequence) is summed by weighted sum to obtain the time step feature representation; wherein, the expression of the time step feature representation F t is as follows:

[0153]

[0154] Further, after combining all time step feature representations, the final fusion sequence representation (i.e. multi-source joint feature vector) is constructed, which encodes multi-dimensional time sequence semantic information such as short-term disturbance, periodic trend and long-term degradation process; wherein, the expression of the multi-source joint feature vector is as follows:

[0155]

[0156] Step S404, predicting the energy storage battery life based on the multi-source joint feature vector to obtain the energy storage battery life prediction result. For details, please refer to step S304 of the embodiment shown in Figure 3 , which will not be repeated here.

[0157] The energy storage battery life prediction method provided in the embodiment, through multi-window division of the multi-dimensional context feature sequence according to a time scale, obtains an energy storage battery life feature vector of multiple time scales, the multi-window division covers life features of different time scales, captures multi-dimensional life-related patterns of short-term fluctuations and long-term trends, calculates a contribution weight based on the energy storage battery life feature vector of multiple time scales, quantifies the importance of features of different time scales to life evaluation, highlights key scale information, and performs weighted summation on the multi-dimensional context feature sequence based on the contribution weight, to obtain a multi-source joint feature vector, the weighted summation integrates multi-scale key features, forms a comprehensive life feature representation that takes into account the importance of each time scale, and improves the integrity and discriminativeness of the multi-source joint feature vector.

[0158] In the embodiment, an energy storage battery life prediction method is provided, which can be used for the electronic device described above, Figure 5 is a flowchart of the energy storage battery life prediction method according to the embodiment of the present application, as shown in the figure, the flowchart includes the following steps: Figure 5

[0159] In step S501, a multi-dimensional time sequence feature vector of the energy storage battery is obtained, local space-time features are extracted based on the multi-dimensional time sequence feature vector of the energy storage battery, and a feature vector fused with time sequence position information is obtained. For details, please refer to step S401 of the embodiment shown in Figure 4 , which will not be repeated here.

[0160] In step S502, time sequence dependence relationship extraction is performed on the feature vector fused with time sequence position information, to obtain a multi-dimensional context feature sequence. For details, please refer to step S402 of the embodiment shown in Figure 4 , which will not be repeated here.

[0161] In step S503, information fusion of multiple time scales is performed on the multi-dimensional context feature sequence, to obtain a multi-source joint feature vector. For details, please refer to step S403 of the embodiment shown in Figure 4 , which will not be repeated here.

[0162] In step S504, the life of the energy storage battery is predicted based on the multi-source joint feature vector, to obtain an energy storage battery life prediction result.

[0163] Specifically, after the extraction and fusion of multi-scale memory information, the model has obtained a joint representation sequence (i.e. multi-source joint feature vector) with rich time sequence semantics, spatial features and context dependence. To realize the final target sequence prediction (i.e. energy storage battery life prediction), the multi-source joint feature vector needs to be further compressed and fused to extract global key information and map to the prediction output space, complete life estimation and running state regression, and obtain an energy storage battery life prediction result.

[0164] ​Specifically, the step S504 includes:

[0165] In step S5041, the attention weight is calculated based on the multi-source joint feature vector.

[0166] Specifically, a feature fusion layer is introduced, which is tasked with converting the multi-source joint feature vector into a context vector with fixed dimensions. Various methods can be used, including global average pooling, attention pooling, or last time step extraction. If a weighted attention pooling mechanism is used, the attention weight β t is calculated as follows:

[0167]

[0168] where W f is a learnable projection matrix, b f is a bias term, and v T is the query vector.

[0169] In step S5042, the multi-source joint feature vector is weighted and summed based on the attention weight, resulting in a battery context feature vector.

[0170] Specifically, the calculation formula of the battery context feature vector z is as follows:

[0171]

[0172] In step S5043, the battery context feature vector is mapped to the target regression output space, resulting in a battery life prediction result.

[0173] Specifically, the obtained battery context feature vector is input into a fully connected regression prediction module, which is composed of a group of linear layers, nonlinear activation functions, and Dropout (a regularization mechanism). The fully connected regression prediction module is used to map the battery context feature vector to the target regression output space. The expression of the battery life prediction result is as follows:

[0174]

[0175] where W o is the output layer weight, and b o is the bias term.

[0176] Further, the feature fusion layer and the regression prediction module play a key role in the whole model, and their performance directly determines the accuracy and stability of the final prediction result. Combined with the attention enhancement, sequence modeling and multi-scale fusion structure in the foregoing, this step ensures that the model has strong representation ability for the life degradation pattern in complex time and space sequences, and provides a reliable quantitative basis for intelligent maintenance and state evaluation.

[0177] The energy storage battery life prediction method provided in the embodiment calculates attention weight based on a multi-source joint feature vector, focuses on the core information in the multi-source features that is strongly related to life, enhances the pertinence and relevance of the features, weights and sums the multi-source joint feature vectors based on the attention weight, obtains an energy storage battery context feature vector, integrates key multi-source information, forms a more compact and discriminative context feature, improves the representation ability of the features, and maps the energy storage battery context feature vector to a target regression output space to obtain an energy storage battery life prediction result, thereby realizing accurate conversion from features to life prediction values and ensuring that the output result meets the quantitative task requirements of life prediction.

[0178] The specific steps of the energy storage battery life prediction method will be described below through a specific embodiment.

[0179] The specific steps of the energy storage battery life prediction method include:

[0180] 1) Input the energy storage battery multi-dimensional time sequence feature vector into the spatio-temporal gating convolution feature extraction module to extract local spatio-temporal features and obtain a feature vector that fuses time sequence position information; wherein the process of extracting local spatio-temporal features includes: expanding the energy storage battery multi-dimensional time sequence feature into an energy storage battery multi-dimensional time sequence feature map; performing convolution on the energy storage battery multi-dimensional time sequence feature map to obtain a multi-granularity local spatio-temporal feature map; performing position coding on the multi-granularity local spatio-temporal feature map, and then performing feature splicing to obtain a feature vector that fuses time sequence position information.

[0181] 2) Input the feature vector that fuses time sequence position information into the CBAM module for attention enhancement to obtain an attention-enhanced feature vector, including: performing padding calculation in the attention mechanism on the feature vector that fuses time sequence position information; and performing memory fusion on the feature vector after padding calculation through a cross-fusion mechanism; performing memory interaction on the feature vector after completing memory fusion to obtain an attention-enhanced feature vector.

[0182] 3) Input the attention-enhanced feature vector into the bidirectional LSTM to extract time sequence dependency relationships and obtain a multi-dimensional context feature sequence.

[0183] 4) Perform multi-time scale information fusion on the multi-dimensional context feature sequence to obtain a multi-source joint feature vector.

[0184] 5) predicting the energy storage battery life based on the multi-source joint feature vector to obtain an energy storage battery life prediction result.

[0185] The embodiment also provides an energy storage battery life prediction device for implementing the above embodiment and preferred implementation, which has been described above. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiment is preferably implemented in software, hardware implementation or a combination of software and hardware implementation is also possible and contemplated.

[0186] The embodiment provides an energy storage battery life prediction device, as shown in the accompanying drawings, comprising: Figure 6

[0187] The first extraction module 601 is configured to obtain an energy storage battery multi-dimensional time sequence feature vector, extract local space-time features based on the energy storage battery multi-dimensional time sequence feature vector, and obtain a feature vector fused with time sequence position information.

[0188] The second extraction module 602 is configured to extract time sequence dependency relationship of the feature vector fused with time sequence position information to obtain a multi-dimensional context feature sequence.

[0189] The fusion module 603 is configured to perform information fusion of the multi-dimensional context feature sequence in multiple time scales to obtain a multi-source joint feature vector.

[0190] The prediction module 604 is configured to predict the energy storage battery life based on the multi-source joint feature vector to obtain an energy storage battery life prediction result.

[0191] In some optional implementations, the first extraction module 601 comprises:

[0192] The expansion unit is configured to expand the energy storage battery multi-dimensional time sequence feature vector into an energy storage battery multi-dimensional time sequence feature map.

[0193] The convolution unit is configured to perform convolution on the energy storage battery multi-dimensional time sequence feature map to obtain a multi-granularity local space-time feature map.

[0194] The encoding unit is configured to perform position encoding on the multi-granularity local space-time feature map to obtain the feature vector fused with time sequence position information.

[0195] In some optional implementations, the encoding unit comprises:

[0196] The mapping subunit is configured to obtain a time step corresponding to the multi-granularity local space-time feature map, and map the time step corresponding to the multi-granularity local space-time feature map into a position encoding vector.

[0197] ​The splicing subunit is configured to perform element-by-element splicing on the position coding vector and the multi-granularity local space-time feature map to obtain a feature vector with fused time-series position information.

[0198] In some optional embodiments, the second extraction module 602 includes:

[0199] The enhancement unit is configured to perform attention enhancement on the feature vector with fused time-series position information to obtain an attention-enhanced feature vector.

[0200] The prediction unit is configured to perform bidirectional state prediction based on the attention-enhanced feature vector to obtain hidden states of each time step.

[0201] The first calculation unit is configured to calculate context attention weights based on the hidden states of each time step.

[0202] The first summation unit is configured to perform weighted summation on the hidden states of each time step based on the context attention weights to obtain a context feature vector.

[0203] The fusion unit is configured to fuse the hidden states of each time step and the context feature vector to obtain a multi-dimensional context feature sequence.

[0204] In some optional embodiments, the fusion module 603 includes:

[0205] The division unit is configured to perform multi-window division on the multi-dimensional context feature sequence according to a time scale to obtain energy storage battery life feature vectors of multiple time scales.

[0206] The second calculation unit is configured to calculate contribution weights based on the energy storage battery life feature vectors of multiple time scales.

[0207] The second summation unit is configured to perform weighted summation on the multi-dimensional context feature sequence based on the contribution weights to obtain a multi-source joint feature vector.

[0208] In some optional embodiments, the prediction module 604 includes:

[0209] The third calculation unit is configured to calculate attention weights based on the multi-source joint feature vector.

[0210] The third summation unit is configured to perform weighted summation on the multi-source joint feature vector based on the attention weights to obtain an energy storage battery context feature vector.

[0211] The mapping unit is configured to map the energy storage battery context feature vector to a target regression output space to obtain an energy storage battery life prediction result.

[0212] Further function descriptions of the above-mentioned modules and units are the same as those of the corresponding embodiments, and will not be described here again.

[0213] In this embodiment, the energy storage battery life prediction device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0214] This invention also provides a computer device having the above-described features. Figure 6 The energy storage battery life prediction device shown.

[0215] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 7 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 7 Take a processor 10 as an example.

[0216] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GPA), or any combination thereof.

[0217] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0218] The memory 20 can include a program storage area and a data storage area, where the program storage area can store an operating system, application programs required for at least one function, and the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some alternative embodiments, the memory 20 can optionally include a memory disposed remotely from the processor 10, which can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0219] The memory 20 can include a volatile memory, such as a random access memory, and can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid state disk, and can also include a combination of the above-mentioned types of memories.

[0220] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 can be connected through a bus or other means, Figure 7 For example, by way of example, through a bus connection.

[0221] The input device 30 can receive inputted digital or character information, and generate key signal inputs related to the user settings and function controls of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), a tactile feedback device (e.g., a vibration motor), etc. The display device includes, but is not limited to, a liquid crystal display, a light-emitting diode, a display, and a plasma display. In some alternative embodiments, the display device can be a touch screen.

[0222] The embodiments of the present application further provide a computer readable storage medium, and the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium through network, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned memories. It can be understood that the computer, the processor, the microprocessor controller, or the programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0223] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, through the operation of the computer, the method and / or technical solutions according to the present application can be called or provided. Those skilled in the art should understand that the form of computer program instructions in a computer readable medium includes but is not limited to source files, executable files, installation package files, etc. Correspondingly, the way of executing computer program instructions by computer includes but is not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0224] Although the embodiments of the present application are described in conjunction with the accompanying drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A method for predicting the lifespan of an energy storage battery, characterized in that, The method includes: Obtain the multi-dimensional time series feature vector of the energy storage battery, and extract local spatiotemporal features based on the multi-dimensional time series feature vector of the energy storage battery to obtain a feature vector that integrates time-series location information; Temporal dependencies are extracted from the feature vectors containing the fused temporal location information to obtain a multidimensional context feature sequence. The multi-dimensional context feature sequence is fused at multiple time scales to obtain a multi-source joint feature vector; The lifespan of the energy storage battery is predicted based on the multi-source joint feature vector, and the prediction result of the energy storage battery lifespan is obtained.

2. The method according to claim 1, characterized in that, The step of extracting local spatiotemporal features based on the multidimensional time series feature vector of the energy storage battery to obtain a feature vector that integrates time-series location information includes: The multi-dimensional time series feature vector of the energy storage battery is extended into a multi-dimensional time series feature map of the energy storage battery. Convolution is performed on the multidimensional time series feature map of the energy storage battery to obtain a multi-granularity local spatiotemporal feature map; The multi-granularity local spatiotemporal feature map is positionally encoded to obtain the feature vector of the fused temporal position information.

3. The method according to claim 2, characterized in that, The step of performing position encoding on the multi-granularity local spatiotemporal feature map to obtain the feature vector fused with temporal position information includes: Obtain the time step corresponding to the multi-granularity local spatiotemporal feature map, and map the time step corresponding to the multi-granularity local spatiotemporal feature map into a position encoding vector; The location encoding vector is concatenated element-wise with the multi-granularity local spatiotemporal feature map to obtain the feature vector of the fused temporal location information.

4. The method according to claim 1, characterized in that, The step of extracting temporal dependencies from the feature vectors containing the fused temporal location information to obtain a multi-dimensional context feature sequence includes: The feature vector fused with temporal location information is enhanced with attention to obtain the attention-enhanced feature vector. Based on the attention-enhanced feature vector, bidirectional state prediction is performed to obtain the hidden state at each time step; Calculate the context attention weights based on the hidden states at each time step; The hidden states at each time step are weighted and summed based on the context attention weights to obtain the context feature vector; The hidden states at each time step and the context feature vectors are fused to obtain the multidimensional context feature sequence.

5. The method according to claim 1, characterized in that, The step of fusing information at multiple time scales on the multi-dimensional context feature sequence to obtain a multi-source joint feature vector includes: The multidimensional context feature sequence is divided into multiple windows according to the time scale to obtain energy storage battery lifetime feature vectors at multiple time scales. The contribution weights are calculated based on the energy storage battery lifetime feature vectors at the multiple time scales. The multi-source joint feature vector is obtained by weighting and summing the multi-dimensional context feature sequence based on the contribution weights.

6. The method according to claim 1, characterized in that, The prediction of energy storage battery life based on the multi-source joint feature vector, to obtain the energy storage battery life prediction result, includes: The attention weight is calculated based on the multi-source joint feature vector. The multi-source joint feature vector is weighted and summed based on the attention weights to obtain the energy storage battery context feature vector. The energy storage battery context feature vector is mapped to the target regression output space to obtain the energy storage battery lifetime prediction result.

7. A device for predicting the lifespan of an energy storage battery, characterized in that, The device includes: The first extraction module is used to obtain the multi-dimensional time series feature vector of the energy storage battery, and extract local spatiotemporal features based on the multi-dimensional time series feature vector of the energy storage battery to obtain a feature vector that integrates time-series location information. The second extraction module is used to extract temporal dependencies from the feature vector of the fused temporal location information to obtain a multi-dimensional context feature sequence. The fusion module is used to perform multi-time-scale information fusion on the multi-dimensional context feature sequence to obtain a multi-source joint feature vector; The prediction module is used to predict the lifespan of the energy storage battery based on the multi-source joint feature vector, and obtain the prediction result of the energy storage battery lifespan.

8. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the energy storage battery life prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to execute the energy storage battery life prediction method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes computer instructions for causing a computer to execute the energy storage battery life prediction method according to any one of claims 1 to 6.

Images