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 the operation data of an energy storage battery, constructing a multi-dimensional time sequence feature matrix based on the operation data of the energy storage battery, and coding the multi-dimensional time sequence feature matrix, and carrying out multi-step decoding prediction on the encoded context feature sequence to obtain a prediction result of the service life of the energy storage battery. The method effectively captures the aging trend and potential mode of the battery, and improves the prediction accuracy of the service life of the energy storage battery.

Description

Technical Field

[0001] This invention relates to the field of energy storage battery life prediction technology, and specifically to a method and apparatus for predicting the lifespan of energy storage batteries. Background Technology

[0002] With the large-scale deployment of energy storage systems in photovoltaic, wind power, electric transportation, and power frequency regulation scenarios, the accurate prediction of the life state of electrochemical batteries, as the core energy unit, places higher demands on ensuring the reliability, safety, and economy of system operation.

[0003] However, the existing methods for predicting the lifespan of energy storage batteries cannot effectively capture battery aging trends and potential patterns, resulting in low accuracy in predicting the lifespan of energy storage batteries. Summary of the Invention

[0004] In view of this, the present invention provides a method and apparatus for predicting the lifespan of energy storage batteries, in order to solve the problem that related energy storage battery lifespan prediction methods cannot effectively capture battery aging trends and potential patterns, resulting in low accuracy in predicting the lifespan of energy storage batteries.

[0005] In a first aspect, the present invention provides a method for predicting the lifespan of an energy storage battery, the method comprising:

[0006] Acquire energy storage battery operation data and construct a multi-dimensional time series feature matrix based on the energy storage battery operation data;

[0007] Encode the multidimensional time series feature matrix to obtain the encoded context feature sequence;

[0008] Multi-step decoding and prediction are performed on the encoded context feature sequence to obtain the predicted lifespan of the energy storage battery.

[0009] The energy storage battery life prediction method provided in this embodiment acquires energy storage battery operation data, constructs a multi-dimensional time series feature matrix based on the energy storage battery operation data, and obtains the encoded context feature sequence by encoding the multi-dimensional time series feature matrix. This achieves the capture of complex time dependencies of the energy storage battery, and performs multi-step decoding and prediction on the encoded context feature sequence, which can effectively avoid the problem of prediction errors from the previous moment propagating to subsequent time points, ensuring the accuracy of energy storage battery life prediction, and finally obtaining the energy storage battery life prediction result. This method effectively captures the aging trend and potential patterns of energy storage batteries, and improves the accuracy of energy storage battery life prediction.

[0010] In one optional implementation, a multi-dimensional time-series feature matrix is ​​constructed based on the energy storage battery operation data, including:

[0011] Global time-series reconstruction of energy storage battery operation data yields global time series data of energy storage battery.

[0012] Feature extraction is performed on the global time series of energy storage batteries to obtain a multidimensional time series feature matrix.

[0013] The energy storage battery life prediction method provided in this embodiment obtains the energy storage battery time series by globally reconstructing the energy storage battery operation data, thereby enhancing the integrity and consistency of the energy storage battery time series and providing a more reliable basis for battery status assessment, life prediction and optimized control. Furthermore, by extracting features from the energy storage battery time series, a multi-dimensional time series feature matrix is ​​obtained, extracting key features from complex operation data and providing a structured and interpretable analytical basis for battery status monitoring, fault diagnosis and performance optimization, thereby improving the ability to capture battery characteristics and the application effect.

[0014] In one optional implementation, the multidimensional time series feature matrix is ​​encoded to obtain the encoded context feature sequence, including:

[0015] The multidimensional time series feature matrix is ​​mapped to a high-dimensional feature space to obtain the high-dimensional time series feature matrix;

[0016] The context features of the high-dimensional time series feature matrix in the time dimension are extracted to obtain the context feature sequence;

[0017] Feature representation enhancement is performed on the context feature sequence to obtain the encoded context feature sequence.

[0018] The energy storage battery life prediction method provided in this embodiment maps a multi-dimensional time series feature matrix to a high-dimensional feature space, which can amplify the subtle differences and correlations between potential features and enhance the ability to identify and learn complex battery operating modes. By extracting the bidirectional contextual features of the high-dimensional time series feature matrix in the time dimension, it fully captures the sequential dependencies of features on the time axis, improves the ability to analyze the dynamic changes in battery operating state, and enhances the feature representation of the contextual feature sequence, it can highlight the recognizability and stability of key features and improve the accuracy of capturing deep modes of battery operating state.

[0019] In one optional implementation, the bidirectional context features of the high-dimensional time series feature matrix in the time dimension are extracted to obtain a context feature sequence, including:

[0020] The high-dimensional time series feature matrix is ​​updated in forward time order to obtain the forward encoded hidden state.

[0021] The state of the high-dimensional time series feature matrix is ​​updated in reverse time order to obtain the reverse encoded hidden state.

[0022] The forward-encoded hidden state and the backward-encoded hidden state are concatenated to obtain the context feature sequence.

[0023] The energy storage battery life prediction method provided in this embodiment updates the state of the high-dimensional time series feature matrix according to the forward and reverse time order, comprehensively integrates historical and future temporal correlation information, improves the global perception and dynamic modeling capability of battery operating state change trend, and enhances the comprehensive characterization capability of bidirectional temporal dependencies in battery operating state by splicing the forward and reverse encoded hidden states, thus providing a more comprehensive feature basis for energy storage battery life prediction.

[0024] In one optional implementation, the context feature sequence is enhanced with feature representation to obtain an encoded context feature sequence, including:

[0025] The context feature sequence is processed by convolutional pooling, and the context feature sequence after convolutional pooling is fused to obtain the channel attention weight vector.

[0026] The channel attention weight vector and the context feature sequence are multiplied to obtain the channel feature map;

[0027] Pooling is performed on the channel feature maps, and then convolution is performed on the pooled channel feature maps to obtain the spatial attention weight vector.

[0028] The channel feature map and the spatial attention weight vector are multiplied to obtain the encoded context feature sequence.

[0029] The energy storage battery life prediction method provided in this embodiment integrates effective information at different scales by performing convolutional pooling on the context feature sequence and then fusing the pooled context feature sequence, thereby enhancing the robustness and representativeness of the features. By multiplying the channel attention weight vector and the context feature sequence, key features are dynamically strengthened and noise in the features is suppressed, improving the discriminative ability of the features. By pooling the channel feature map and then performing convolution on the pooled channel feature map, key spatial location features are focused and interference from irrelevant features is suppressed, enhancing the spatial discriminativeness of the features. Furthermore, by multiplying the channel feature map and the spatial attention weight vector, effective features of key spatial locations are precisely strengthened and interference from redundant regions is suppressed, improving the spatial focus and discriminative power of the encoded context feature sequence.

[0030] In one optional implementation, multi-step decoding and prediction are performed on the encoded context feature sequence to obtain the energy storage battery lifetime prediction result, including:

[0031] The encoded context feature sequence is bidirectionally decoded according to the sequence time step to obtain the forward decoding hidden state and the reverse decoding hidden state;

[0032] By concatenating the forward-decoded hidden state and the reverse-decoded hidden state, the current state of the energy storage battery can be obtained.

[0033] A linear transformation is performed on the current state of the energy storage battery to obtain the predicted lifespan of the energy storage battery.

[0034] The energy storage battery life prediction method provided in this embodiment fully explores the sequential dependencies of the encoded context feature sequence by bidirectionally decoding the sequence according to the sequence time step, enhancing the ability to capture and fully represent the dynamic changes of the sequence. By splicing the forward and backward decoded hidden states, the forward and backward dependency information of the time series features is integrated to form a more comprehensive sequence representation, improving the ability to capture and deeply understand the time series context. Finally, by performing a linear transformation on the current state of the energy storage battery, the accurate prediction of the energy storage battery life is achieved.

[0035] In a second aspect, the present invention provides an energy storage battery life prediction device, the device comprising:

[0036] The module is used to acquire energy storage battery operation data and construct a multi-dimensional time series feature matrix based on the energy storage battery operation data;

[0037] The encoding module is used to encode the multidimensional time series feature matrix to obtain the encoded context feature sequence;

[0038] The prediction module is used to perform multi-step decoding and prediction on the encoded context feature sequence to obtain the predicted lifespan of the energy storage battery.

[0039] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the energy storage battery life prediction method of the first aspect or any corresponding embodiment described above.

[0040] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the energy storage battery life prediction method of the first aspect or any corresponding embodiment described above.

[0041] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the energy storage battery life prediction method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0042] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating a method for predicting the lifespan of an energy storage battery according to an embodiment of the present invention.

[0044] Figure 2 This is a flowchart illustrating another method for predicting the lifespan of an energy storage battery according to an embodiment of the present invention.

[0045] Figure 3 This is a flowchart illustrating another method for predicting the lifespan of an energy storage battery according to an embodiment of the present invention.

[0046] Figure 4 This is a flowchart illustrating another method for predicting the lifespan of an energy storage battery according to an embodiment of the present invention.

[0047] Figure 5 This is a structural block diagram of an energy storage battery life prediction device according to an embodiment of the present invention;

[0048] Figure 6 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] As the proportion of new energy power generation in the power system continues to rise, electrochemical energy storage systems, as an important support for peak shaving, valley filling, frequency regulation, voltage regulation, and load response, have become a core component in building new power systems. Among them, lithium-ion batteries, with their advantages of high specific energy, long cycle life, and fast response speed, are widely used in centralized energy storage power stations and distributed user-side energy storage scenarios. However, due to the combined effects of material aging, operating temperature, and other factors, the degradation process of energy storage batteries exhibits highly nonlinear, complex dynamic, and individual variability characteristics, making it difficult to accurately characterize the capacity decay rate through statistical models.

[0051] In practical applications, energy storage batteries typically complete about 3 to 4 full charge-discharge cycles per day. The number of individual cells is large and their states are unevenly distributed. Once abnormal degradation or performance deterioration occurs, it will significantly affect the safety and economy of the system. Therefore, it is urgent to build a life prediction model with high timeliness, strong generalization ability and dynamic modeling ability. It should not only be able to track the degradation trend of batteries under the current operating conditions, but also support the prediction of cycle behavior at the "day" granularity, so as to provide a reliable basis for intelligent scheduling, maintenance early warning and life management of batteries.

[0052] Currently, extensive research has been conducted on predicting the remaining useful life (RUL) and state of health (SOH) of batteries. Many related methods rely on mechanistic modeling and statistical analysis. However, when faced with the complex nonlinear degradation behavior of batteries, these methods generally suffer from difficulties in modeling, poor real-time performance, and weak generalization ability. With the development of deep learning technology, data-driven time series prediction methods have gradually become mainstream. Among them, sequence models based on recurrent neural networks (RNNs) and long short-term memory networks (LSTMs) have shown excellent performance in processing time series data. However, these models still have performance bottlenecks when dealing with long sequence modeling and multi-step prediction.

[0053] To address the aforementioned technical issues, this invention provides a method for predicting the lifespan of energy storage batteries. Based on an improved Seq2Seq (Sequence-to-Sequence) architecture, it uses historical operational data such as charging and discharging current, voltage, temperature, and SOC (State of Charge) as input. An encoder-decoder structure extracts temporal features to predict the SOH value of the battery at multiple future time steps. Compared to regression prediction methods, the improved Seq2Seq architecture has significant advantages in long-sequence modeling, more effectively capturing battery aging trends and potential patterns. Furthermore, an attention mechanism is introduced to enhance the perception of key time segments, improving prediction accuracy and stability. To enhance the model's robustness and adaptability, and to outperform related methods in prediction accuracy and stability, it also possesses good deployment feasibility and system scalability. This provides strong technical support for the intelligent operation and maintenance of electrochemical energy storage systems, laying the foundation for widespread application in smart energy systems, grid-side energy storage, electric vehicles, and distributed energy storage nodes.

[0054] This invention provides a method for predicting the lifespan of energy storage batteries. It should be noted that the execution subject of this method can be a device for predicting the lifespan of energy storage batteries. This device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. The electronic device can be a server or a terminal. In this embodiment, the server can be a single server or a server cluster composed of multiple servers. The terminal in this embodiment can be a smartphone, personal computer, tablet computer, wearable device, or other intelligent hardware device such as a smart robot. The following method embodiments all use an electronic device as the execution subject for description.

[0055] According to an embodiment of the present invention, an embodiment of a method for predicting the lifespan of an energy storage battery is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0056] This embodiment provides a method for predicting the lifespan of energy storage batteries, which can be used in the aforementioned electronic devices. Figure 1 This is a flowchart of a battery life prediction method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0057] Step S101: Obtain the operating data of the energy storage battery and construct a multi-dimensional time series feature matrix based on the operating data of the energy storage battery.

[0058] Specifically, in the task of predicting the lifespan of energy storage batteries, the operating data of energy storage batteries usually comes from the periodic monitoring of experimental benches or actual operating systems. It covers multiple key operating parameters and forms a sequence feature in the time dimension, which can reflect the performance degradation trend of batteries in long-term operation. In order to fully explore the lifespan information contained in the key operating parameters, it is necessary to first perform systematic preprocessing and feature factor extraction on the operating data of energy storage batteries, and construct a representative multi-dimensional time series feature matrix to provide high-quality input for subsequent neural network model training.

[0059] Furthermore, the energy storage battery operation data originates from a high-precision data acquisition platform and has multiple raw features, such as Data Point, Test Time, Date Time, Step Time, and Step Index. Through a scientific data extraction and processing process, seven core feature variables were constructed from these raw measurement fields (i.e., energy storage battery operation data): Cycle, Disc Capacity, Heal Indicator, Resistance, CCCT (Constant Current Cycle Test), CVCT (Constant Voltage Cycle Test), and DISCT (Discharge Cycle Test). A multi-dimensional time series feature matrix was then constructed based on these core feature variables.

[0060] Step S102: Encode the multidimensional time series feature matrix to obtain the encoded context feature sequence.

[0061] Step S103: Perform multi-step decoding and prediction on the encoded context feature sequence to obtain the energy storage battery life prediction result.

[0062] Specifically, the encoded context feature sequence is decoded and predicted step-by-step by the bidirectional GRU network in the decoder part, realizing dynamic multi-step prediction of the life of the energy storage battery. That is, after the encoder completes feature extraction and context modeling of the multi-dimensional time series feature vector, it enters the decoding stage. The main task of the decoder is to generate prediction results for several future time steps based on the context feature sequence output by the encoder, thereby realizing dynamic multi-step prediction of the life of the energy storage battery. Unlike the related one-step prediction, which only predicts the health status value (such as SOH, remaining capacity) at a single future time point, multi-step prediction can provide the trend trajectory for a period of time in the future, which helps to identify degradation risks in advance and enhance the foresight and safety of system maintenance.

[0063] Furthermore, in terms of prediction strategy, to address the problem of error accumulation in multi-step prediction, a "parallel prediction (direct multi-output)" mechanism is adopted. This means that the prediction results for all future time steps are generated simultaneously in a single decoding stage. Compared with stepwise recursive prediction, this can effectively avoid the problem of prediction errors from the previous time step propagating to subsequent time points. It is particularly suitable for modeling highly nonlinear degradation processes such as the aging behavior of lithium batteries. In addition, to further improve prediction stability and generalization ability, the network end can be optimized by adding strategies such as Dropout (a regularization technique) and regularization loss.

[0064] In practical applications, multi-step prediction capabilities are crucial for the operation and management of energy storage systems. Battery life is subject to the nonlinear effects of multiple factors such as temperature, load, SOC, and rate, and its degradation process exhibits phased and abrupt characteristics. Therefore, by accurately predicting the trends of key indicators in multiple future cycles through a decoder, sufficient time windows can be provided for system scheduling, enabling early warning of "soft failures" and adjustments to operation and maintenance plans, thereby greatly reducing operating costs and safety risks.

[0065] The energy storage battery life prediction method provided in this embodiment acquires energy storage battery operation data, constructs a multi-dimensional time series feature matrix based on the energy storage battery operation data, and obtains the encoded context feature sequence by encoding the multi-dimensional time series feature matrix. This achieves the capture of complex time dependencies of the energy storage battery, and performs multi-step decoding and prediction on the encoded context feature sequence, which can effectively avoid the problem of prediction errors from the previous moment propagating to subsequent time points, ensuring the accuracy of energy storage battery life prediction, and finally obtaining the energy storage battery life prediction result. This method effectively captures the aging trend and potential patterns of energy storage batteries, and improves the accuracy of energy storage battery life prediction.

[0066] This embodiment provides a method for predicting the lifespan of energy storage batteries, which can be used in the aforementioned electronic devices. Figure 2 This is a flowchart of a battery life prediction method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:

[0067] Step S201: Obtain the operating data of the energy storage battery and construct a multi-dimensional time series feature matrix based on the operating data of the energy storage battery.

[0068] Specifically, step S201 includes:

[0069] Step S2011: Perform global time series reconstruction on the energy storage battery operation data to obtain the global time series of the energy storage battery.

[0070] Specifically, raw data is usually collected in the form of experimental sampling or sensor recording, including timestamps (such as test time and step time), cycle index, step sequence and corresponding electrochemical parameters such as voltage, current and internal resistance, and an initial global time series is constructed based on the raw steps.

[0071] Furthermore, to improve the temporal consistency and prediction stability of the modeling, it is necessary to extract global features from the initial global time series, that is, to reconstruct the global time series according to the actual occurrence order, and to build a unified periodic index based on the charge and discharge cycle, so that subsequent features have a strict time reference system. The above process is essentially to convert unstructured and redundant data into structured periodic time series (i.e., energy storage battery time series), so that each period constitutes an ordered sample unit (i.e., energy storage battery global time series).

[0072] Step S2012: Extract features from the global time series of the energy storage battery to obtain a multi-dimensional time series feature matrix.

[0073] Specifically, in the task of modeling and predicting the lifespan of energy storage batteries, the design and extraction of feature factors are the fundamental steps in the entire modeling process. Unlike general time series prediction problems, battery life prediction has significant characteristics such as physical driving, performance evolution, and gradual degradation. Therefore, before constructing the multidimensional time series feature matrix of the model input, it is necessary to combine the working mechanism of the battery, experimental paradigm, and time series characteristics to systematically preprocess the raw data and perform multi-scale feature mining (i.e., extract local features from the global time series of energy storage batteries) to ensure that the process of predicting the lifespan of energy storage batteries has a high information density input foundation.

[0074] Furthermore, the original features contained in the energy storage battery sequence after local feature extraction include: DataPoint, Test Time, Date Time, Step Time, and Step Index. Features are extracted from the energy storage battery time series according to different feature classifications, and a multi-dimensional time feature matrix is ​​constructed, which includes 7 core feature variables: Cycle, Disc Capacity, Heal Indicator, Resistance, CCCT (Constant Current Cycle Test), CVCT (Constant Voltage Cycle Test), and DISCT (Discharge Cycle Test).

[0075] Furthermore, the local feature extraction process adopts a single-channel, channel-isolated approach to adaptively determine the different periodic characteristics of the energy storage battery time series in order to extract the feature information of each local time period.

[0076] Furthermore, in the processing of energy storage battery operation data, to ensure the consistency between global features and local features at different scales, time alignment is required first. Specifically, using the global time index as a reference, the extracted local features are mapped onto a unified time axis through interpolation, forward padding, or Dynamic Time Warping (DTW) methods to obtain a complete sequence representation. The aligned local features are then normalized and dimensionally projected to eliminate dimensional differences. Subsequently, weighted concatenation or attention mechanisms are used to fuse global and local features, enabling the model to capture both overall trends and reflect local fluctuations, ultimately yielding a multidimensional time series feature matrix, providing comprehensive input for subsequent prediction tasks. The expression for the multidimensional time series feature matrix is ​​as follows:

[0077]

[0078] Among them, g t Represents global features. Let x represent the k-th local feature after alignment, b be a learnable parameter, and x be a local feature. t This is the output multidimensional time series feature matrix.

[0079] Step S202: Encode the multidimensional time series feature matrix to obtain the encoded context feature sequence. For details, please refer to [link to relevant documentation]. Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0080] Step S203 involves performing multi-step decoding and prediction on the encoded context feature sequence to obtain the energy storage battery lifetime prediction result. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0081] The energy storage battery life prediction method provided in this embodiment obtains the energy storage battery time series by globally reconstructing the energy storage battery operation data, thereby enhancing the integrity and consistency of the energy storage battery time series and providing a more reliable basis for battery status assessment, life prediction and optimized control. Furthermore, by extracting features from the energy storage battery time series, a multi-dimensional time series feature matrix is ​​obtained, extracting key features from complex operation data and providing a structured and interpretable analytical basis for battery status monitoring, fault diagnosis and performance optimization, thereby improving the ability to capture battery characteristics and the application effect.

[0082] This embodiment provides a method for predicting the lifespan of energy storage batteries, which can be used in the aforementioned electronic devices. Figure 3 This is a flowchart of a battery life prediction method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0083] Step S301: Obtain energy storage battery operation data and construct a multi-dimensional time series feature matrix based on the energy storage battery operation data. For details, please refer to [link to relevant documentation]. Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0084] Step S302: Encode the multidimensional time series feature matrix to obtain the encoded context feature sequence.

[0085] Specifically, an encoder is used to encode the multidimensional time-series feature matrix. To effectively enhance the expressive power of the input features and fully explore their contextual dependencies in the time dimension, the encoder structure consists of three key modules: a linear embedding layer, a bidirectional gated recurrent unit (BiGRU), and a convolutional block attention module (CBAM). While maintaining good trainability, it can effectively improve the ability to model the deep structural relationships between time-series features, thereby providing a high-quality representation foundation for the energy storage lifetime prediction task at the decoder end.

[0086] Furthermore, in the sequence modeling task of predicting the lifespan of energy storage batteries, the input multidimensional time series feature vectors are often low in dimensionality (e.g., 7-dimensional), and there are complex time series features such as nonlinearity, non-stationarity, and non-Gaussianity among the data. Therefore, in the encoder part (i.e., encoding the multidimensional time series feature matrix), the original feature factors are first mapped to the high-dimensional feature space through a linear embedding layer. Then, a bidirectional gated recurrent unit is used to extract the bidirectional context features of the sequence data in the time dimension, thereby capturing the complex time dependencies.

[0087] Specifically, step S302 includes:

[0088] Step S3021: Map the multidimensional time series feature matrix to a high-dimensional feature space to obtain the high-dimensional time series feature matrix.

[0089] Specifically, after standardization, the multidimensional time series feature matrix has the following dimensions: Where L represents the time step length and D is the feature dimension of the multidimensional time series feature matrix. Since D is small and insufficient for direct high-order time series modeling, a linear embedding layer is introduced in the first layer of the encoder to embed the feature vectors in the multidimensional time series feature matrix at each time step. A linear transformation is performed to map the multidimensional time series feature matrix to a larger embedding space, resulting in a high-dimensional time series feature matrix. The expression for the linear transformation of the eigenvectors in the multidimensional time series feature matrix is ​​as follows:

[0090]

[0091] Among them, W e Let be a trainable weight matrix, and the dimension of the trainable weight matrix is ​​. d e The embedding dimension can be set according to the complexity of the downstream model, b e The bias term has the following dimensions:

[0092] Furthermore, the significance of linear embedding lies not only in dimensionality expansion but also in constructing a new feature space, allowing the original nonlinear structure to be better captured by subsequent networks; where the expression for the high-dimensional time series feature matrix E is:

[0093]

[0094] Step S3022: Extract the bidirectional context features of the high-dimensional time series feature matrix in the time dimension to obtain the context feature sequence.

[0095] Specifically, the high-dimensional time series feature matrix E output by the linear embedding layer is fed into a bidirectional gated recurrent unit network for context feature extraction. When processing sequence data, unidirectional RNN-type network models only utilize historical information, i.e., forward modeling. In practice, they may ignore the potential impact of future time steps on the current state. Therefore, a bidirectional GRU (Gated Recurrent Unit) is used to perform parallel computation through two independent forward and backward GRU layers, effectively fusing contextual information.

[0096] Furthermore, the basic unit structure of GRU has a stronger long-term memory capability compared to RNN structure. Its update gate and reset gate control the retention and discard of information flow, respectively. BiGRU not only improves the flexibility and expressiveness of sequence modeling, but also takes into account historical and future information without significantly increasing computational complexity, and has a good modeling effect on the gradual change time problem of battery life.

[0097] In some optional implementations, step S3022 above includes:

[0098] Step a1: Update the state of the high-dimensional time series feature matrix according to the forward time order to obtain the forward encoded hidden state.

[0099] Specifically, the expression for forward encoding the hidden state is:

[0100]

[0101] in, For the current forward-encoded hidden state, GRU f For the forward GRU layer, e t These are the eigenvectors in the high-dimensional time series feature matrix. This is the forward-encoded hidden state from the previous time step.

[0102] Step a2: Update the state of the high-dimensional time series feature matrix in reverse time order to obtain the reverse encoded hidden state.

[0103] Specifically, the expression for the reverse-encoded hidden state is:

[0104]

[0105] in, For the reverse-encoded hidden state at the current time, GRU f For backward GRU layer, This is the reverse-encoded hidden state for the next time step.

[0106] Step a3: Concatenate the forward-encoded hidden state and the reverse-encoded hidden state to obtain the context feature sequence.

[0107] Specifically, the expression for the context feature sequence is:

[0108]

[0109] Among them, h t The context feature sequence has a dimension of . h represents the number of hidden units.

[0110] Step S3023: Enhance the feature representation of the context feature sequence to obtain the encoded context feature sequence.

[0111] Specifically, to further enhance the encoder's ability to focus on key time steps and feature dimensions, a CBAM attention mechanism is introduced as a post-processing module for the encoder output. CBAM includes channel attention and temporal attention.

[0112] In some optional implementations, step S3023 above includes:

[0113] Step b1 involves performing convolutional pooling on the context feature sequence and then fusing the convolutional pooling context feature sequence to obtain the channel attention weight vector.

[0114] Specifically, the expression for the channel attention weight vector is:

[0115]

[0116] Where σ is the activation function, MLP represents the perceptron used to perform pooling on the pooled context feature sequence, AvgPool is average pooling, MaxPool is global max pooling, and M... c (h t ) represents the channel attention weight vector. is the dimension of the channel attention weight vector.

[0117] Step b2: Multiply the channel attention weight vector and the context feature sequence to obtain the channel feature map.

[0118] Specifically, the expression for the channel feature map is:

[0119] F c =M c (h t )⊕h t (8)

[0120] Among them, F c This is a channel feature map.

[0121] Step b3: Pool the channel feature maps and then perform convolution on the pooled channel feature maps to obtain the spatial attention weight vector.

[0122] Specifically, the expression for the spatial attention weight vector is:

[0123]

[0124] Where Mt(Fc) is the spatial attention weight vector, ConvlD is the convolution process, and the dimension of the spatial attention weight vector is...

[0125] Step b4 involves multiplying the channel feature map and the spatial attention weight vector to obtain the encoded context feature sequence.

[0126] Specifically, the expression for the encoded context feature sequence is:

[0127]

[0128] Among them, F att This is the encoded context feature sequence.

[0129] Step S303 involves performing multi-step decoding and prediction on the encoded context feature sequence to obtain the energy storage battery lifetime prediction result. For details, please refer to [link to relevant documentation]. Figure 2 Step S203 of the illustrated embodiment will not be described again here.

[0130] The energy storage battery life prediction method provided in this embodiment maps a multi-dimensional time series feature matrix to a high-dimensional feature space, which can amplify the subtle differences and correlations between potential features and enhance the ability to identify and learn complex battery operating modes. By extracting the bidirectional contextual features of the high-dimensional time series feature matrix in the time dimension, it fully captures the sequential dependencies of features on the time axis, improves the ability to analyze the dynamic changes in battery operating state, and enhances the feature representation of the contextual feature sequence, it can highlight the recognizability and stability of key features and improve the accuracy of capturing deep modes of battery operating state.

[0131] This embodiment provides a method for predicting the lifespan of energy storage batteries, which can be used in the aforementioned electronic devices. Figure 4 This is a flowchart of a battery life prediction method according to an embodiment of the present invention, such as... Figure 4 As shown, the process includes the following steps:

[0132] Step S401: Obtain energy storage battery operation data and construct a multi-dimensional time series feature matrix based on the energy storage battery operation data. For details, please refer to [link to relevant documentation]. Figure 3 Step S301 of the illustrated embodiment will not be described again here.

[0133] Step S402: Encode the multidimensional time series feature matrix to obtain the encoded context feature sequence. For details, please refer to [link to relevant documentation]. Figure 3 Step S302 of the illustrated embodiment will not be described again here.

[0134] Step S403: Perform multi-step decoding and prediction on the encoded context feature sequence to obtain the energy storage battery life prediction result.

[0135] Specifically, the decoder performs multi-step decoding and prediction on the encoded context feature sequence. The decoder uses a bidirectional gated recurrent unit network that is symmetrical to the encoder structure to perform multi-step decoding and prediction. In the Seq2Seq architecture, the decoder often uses a unidirectional GRU or LSTM structure to generate the output from left to right. However, in the weakly supervised and strongly time-dependent problem of energy storage system lifetime prediction, the introduction of a bidirectional modeling strategy can more comprehensively characterize the internal structure of the target sequence and improve the fitting accuracy of future trends.

[0136] Furthermore, the encoded context feature sequence output by the encoder serves as the initial input state for the decoder. The decoder then generates a prediction sequence of length T based on the encoded context feature sequence. The prediction at each time step depends on the historical predicted values ​​and the encoded state. By introducing BiGRU, which is consistent with the encoder structure, the bidirectional dependencies between future states can be modeled synchronously during the prediction process.

[0137] Specifically, step S403 includes:

[0138] Step S4031: The encoded context feature sequence is bidirectionally decoded according to the sequence time step to obtain the forward decoding hidden state and the reverse decoding hidden state.

[0139] Specifically, the expressions for the forward decoding hidden state and the backward decoding hidden state are:

[0140]

[0141] in, This represents the current forward decoding hidden state. This is the hidden state of the reverse decoding at the current moment. This is the predicted output from the previous time step. This is the predicted output for the next time step.

[0142] Step S4032: Concatenate the forward decoding hidden state and the reverse decoding hidden state to obtain the current state of the energy storage battery.

[0143] Specifically, the current state s of the energy storage battery t The expression is:

[0144]

[0145] Step S4033: Perform a linear transformation on the current state of the energy storage battery to obtain the predicted lifespan of the energy storage battery.

[0146] Specifically, the expression for the predicted lifespan of energy storage batteries is as follows:

[0147]

[0148] in, W represents the predicted state of battery life at the current time step in the future. o b o This is for outputting the parameters of the mapping layer.

[0149] The energy storage battery life prediction method provided in this embodiment fully explores the sequential dependencies of the encoded context feature sequence by bidirectionally decoding the sequence according to the sequence time step, enhancing the ability to capture and fully represent the dynamic changes of the sequence. By splicing the forward and backward decoded hidden states, the forward and backward dependency information of the time series features is integrated to form a more comprehensive sequence representation, improving the ability to capture and deeply understand the time series context. Finally, by performing a linear transformation on the current state of the energy storage battery, the accurate prediction of the energy storage battery life is achieved.

[0150] The following specific embodiment illustrates the detailed steps of a method for predicting the lifespan of an energy storage battery:

[0151] Example 1:

[0152] 1) Preprocess the raw running data and extract feature factors to construct a multidimensional time series as the basic feature representation (i.e., multidimensional time series feature matrix) as the input of the neural network.

[0153] 2) An encoder-decoder structure is used for modeling. In the encoder part, the original feature factors are first mapped to a high-dimensional feature space through a linear embedding layer to enhance the feature representation capability. Then, bidirectional gated recurrent units are used to extract bidirectional contextual features of the sequence data in the time dimension, thereby capturing complex temporal dependencies.

[0154] 3) To further enhance the model's ability to focus on key features, a convolutional attention module (CBAM) was introduced to significantly enhance the encoded features in both the channel and spatial dimensions, thereby uncovering deep interactions and dependencies between features and improving the discriminative power of feature representation.

[0155] 4) Finally, the encoded context feature sequence is decoded and predicted step by step by the bidirectional GRU network of the decoder part to realize dynamic, multi-step rolling prediction of the life of the energy storage battery and obtain the prediction result of the life of the energy storage battery.

[0156] This embodiment also provides an energy storage battery life prediction device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. 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 embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0157] This embodiment provides a device for predicting the lifespan of an energy storage battery, such as... Figure 5 As shown, it includes:

[0158] Module 501 is used to acquire energy storage battery operation data and construct a multi-dimensional time series feature matrix based on the energy storage battery operation data.

[0159] The encoding module 502 is used to encode the multidimensional time series feature matrix to obtain the encoded context feature sequence.

[0160] The prediction module 503 is used to perform multi-step decoding and prediction on the encoded context feature sequence to obtain the prediction result of the energy storage battery life.

[0161] In some alternative implementations, the construction module 501 includes:

[0162] The reconstruction unit is used to perform global time-series reconstruction of the energy storage battery's operating data to obtain the global time series of the energy storage battery.

[0163] The first extraction unit is used to extract features from the global time series of the energy storage battery to obtain a multi-dimensional time series feature matrix.

[0164] In some alternative implementations, the encoding module 502 includes:

[0165] The mapping unit is used to map the multidimensional time series feature matrix to a high-dimensional feature space to obtain the high-dimensional time series feature matrix.

[0166] The second extraction unit is used to extract the bidirectional context features of the high-dimensional time series feature matrix in the time dimension to obtain the context feature sequence.

[0167] The enhancement unit is used to enhance the feature representation of the context feature sequence to obtain the encoded context feature sequence.

[0168] In some optional implementations, the second extraction unit includes:

[0169] The first update subunit updates the state of the high-dimensional time series feature matrix according to the forward time order to obtain the forward encoded hidden state.

[0170] The second update subunit is used to update the state of the high-dimensional time series feature matrix in reverse time order to obtain the reverse encoded hidden state.

[0171] The splicing subunit is used to splice the forward-encoded hidden state and the reverse-encoded hidden state to obtain the context feature sequence.

[0172] In some alternative implementations, the enhancement unit includes:

[0173] The first pooling subunit is used to perform convolutional pooling on the context feature sequence and to fuse the context feature sequence after convolutional pooling to obtain the channel attention weight vector.

[0174] The first multiplication subunit is used to multiply the channel attention weight vector and the context feature sequence to obtain the channel feature map.

[0175] The second pooling subunit is used to perform pooling on the channel feature map and convolution on the pooled channel feature map to obtain the spatial attention weight vector.

[0176] The second multiplication subunit is used to multiply the channel feature map and the spatial attention weight vector to obtain the encoded context feature sequence.

[0177] In some alternative implementations, the prediction module 503 includes:

[0178] The decoding unit is used to perform bidirectional decoding on the encoded context feature sequence according to the sequence time step to obtain the forward decoding hidden state and the reverse decoding hidden state.

[0179] The splicing unit is used to splice the forward-decoded hidden state and the reverse-decoded hidden state to obtain the current state of the energy storage battery.

[0180] The transformation unit is used to perform a linear transformation on the current state of the energy storage battery to obtain the predicted lifespan of the energy storage battery.

[0181] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0182] 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.

[0183] This invention also provides a computer device having the above-described features. Figure 5 The device shown is for predicting the lifespan of an energy storage battery.

[0184] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 6As 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 6 Take a processor 10 as an example.

[0185] 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.

[0186] 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.

[0187] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0188] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0189] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0190] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0191] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0192] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0193] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all 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: Acquire energy storage battery operation data, and construct a multi-dimensional time series feature matrix based on the energy storage battery operation data; The multidimensional time series feature matrix is ​​encoded to obtain the encoded context feature sequence; The encoded context feature sequence is decoded and predicted in multiple steps to obtain the energy storage battery life prediction result.

2. The method according to claim 1, characterized in that, The construction of a multi-dimensional time series feature matrix based on the energy storage battery operation data includes: The global time series of the energy storage battery operation data is reconstructed to obtain the global time series of the energy storage battery. Feature extraction is performed on the global time series of the energy storage battery to obtain the multidimensional time series feature matrix.

3. The method according to claim 1, characterized in that, The process of encoding the multidimensional time series feature matrix to obtain the encoded context feature sequence includes: The multidimensional time series feature matrix is ​​mapped to a high-dimensional feature space to obtain a high-dimensional time series feature matrix; The bidirectional context features of the high-dimensional time series feature matrix in the time dimension are extracted to obtain the context feature sequence. The context feature sequence is enhanced with feature representation to obtain the encoded context feature sequence.

4. The method according to claim 3, characterized in that, The step of extracting bidirectional contextual features in the time dimension of the high-dimensional time series feature matrix to obtain a contextual feature sequence includes: The high-dimensional time series feature matrix is ​​updated according to the forward time sequence to obtain the forward encoded hidden state. The high-dimensional time series feature matrix is ​​updated in reverse time order to obtain the reverse encoded hidden state. The forward encoded hidden state and the reverse encoded hidden state are concatenated to obtain the context feature sequence.

5. The method according to claim 3, characterized in that, The step of enhancing the feature representation of the context feature sequence to obtain the encoded context feature sequence includes: The context feature sequence is subjected to convolutional pooling, and the context feature sequence after convolutional pooling is fused to obtain the channel attention weight vector. Multiply the channel attention weight vector and the context feature sequence to obtain the channel feature map; The channel feature map is pooled, and the pooled channel feature map is convolved to obtain the spatial attention weight vector. The channel feature map and the spatial attention weight vector are multiplied together to obtain the encoded context feature sequence.

6. The method according to claim 1, characterized in that, The step of performing multi-step decoding and prediction on the encoded context feature sequence to obtain the energy storage battery life prediction result includes: The encoded context feature sequence is bidirectionally decoded according to the sequence time step to obtain the forward decoding hidden state and the reverse decoding hidden state; The current state of the energy storage battery is obtained by concatenating the forward decoding hidden state and the reverse decoding hidden state. A linear transformation is performed on the current state of the energy storage battery to obtain the predicted lifespan of the energy storage battery.

7. A device for predicting the lifespan of an energy storage battery, characterized in that, The device includes: The module is used to acquire energy storage battery operation data and construct a multi-dimensional time series feature matrix based on the energy storage battery operation data; The encoding module is used to encode the multidimensional time series feature matrix to obtain the encoded context feature sequence; The prediction module is used to perform multi-step decoding and prediction on the encoded context feature sequence to obtain the energy storage battery life prediction result.

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 energy storage battery life prediction method according to any one of claims 1 to 6 by executing the computer instructions.

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, It 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