State estimation method and device of energy storage unit, computer equipment and storage medium

Abstract

The invention relates to a state estimation method and device of an energy storage unit, computer equipment and a storage medium. The method comprises the following steps: acquiring running state sequence data of all energy storage units in the hybrid energy storage system to be analyzed; inputting the running state sequence data into a normalization model to obtain a normalization feature vector of each energy storage unit; taking the normalized feature vector of each energy storage unit as a node feature and the incidence relation between the energy storage units as an edge to construct a full connection graph; inputting the full connection graph into a graph neural network model to obtain state estimation information of each energy storage unit; the graph neural network model is used for updating the characteristics of each energy storage unit by aggregating the characteristics of at least one layer of similar energy storage unit, and obtaining the state estimation information of the energy storage unit by using the updated characteristics of the energy storage unit. The method can be applied to a hybrid energy storage system comprising different types of energy storage units.

Description

Technical Field

[0001] This application relates to the field of hybrid energy storage system technology, and in particular to a state estimation method, apparatus, computer equipment, and storage medium for an energy storage unit. Background Technology

[0002] In recent years, with the rapid expansion of the global new energy industry, the intermittency and volatility of new energy sources such as wind power and photovoltaics have posed challenges to grid stability and power quality. Against this backdrop, hybrid energy storage systems, due to their ability to integrate the advantages of different energy storage technologies, have become a key technological direction for promoting energy structure transformation. The efficient operation of hybrid energy storage systems depends on the precise management of each energy storage unit; therefore, it is necessary to monitor and estimate the status of each energy storage unit within the hybrid energy storage system.

[0003] In related technologies, energy storage system modeling and state estimation algorithms are mostly designed for single chemical systems, such as being only suitable for lithium-ion batteries. This results in the inability to be applied to hybrid energy storage systems that contain different types of energy storage units. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, and storage medium for state estimation of energy storage units that can be applied to hybrid energy storage systems containing different types of energy storage units, in order to address the above-mentioned technical problems.

[0005] In a first aspect, this application provides a state estimation method for an energy storage unit, including:

[0006] Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed;

[0007] The operating state sequence data is input into a normalization model to obtain the normalized feature vector of each energy storage unit;

[0008] A fully connected graph is constructed using the normalized feature vectors of each energy storage unit as node features and the relationships between each energy storage unit as edges.

[0009] The fully connected graph is input into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit using the updated features.

[0010] In one embodiment, the method further includes:

[0011] For each energy storage unit, the weighted cosine similarity between the energy storage unit and its neighboring energy storage units is determined using the feature vector of the energy storage unit, the feature vectors of the energy storage unit and its neighboring energy storage units, and a preset feature weight.

[0012] Adjacent energy storage units with a weighted cosine similarity greater than a similarity threshold are identified as similar energy storage units.

[0013] In one embodiment, the method further includes:

[0014] For each energy storage unit, based on the first The adjacent energy storage units of the energy storage unit are determined by considering the candidate energy storage units adjacent to the energy storage unit, the preset sampling quantity, and the data sampling strategy.

[0015] In one embodiment, the aggregation formula for feature aggregation of each energy storage unit is as follows: In the formula, This indicates the updated characteristics of the energy storage unit. Indicates the characteristics of the energy storage unit, For adjacent energy storage units of an energy storage unit, It is the identity matrix. For degree matrix, For learnable parameter matrix, This is the activation function.

[0016] In one embodiment, the normalization model includes a mask, an encoder, and a decoder; the training process of the normalization model includes:

[0017] The sample operating status sequence data of all energy storage units in the hybrid energy storage system is input into the masker. The masker performs masking processing on the sample operating status sequence data using a high-proportion random mask to obtain masked sample operating status sequence data. The masked sample operating status sequence data includes masked data and unmasked data.

[0018] The unmasked data and its position encoding are input into the encoder to obtain the latent feature information of the unmasked data;

[0019] The location encodings of the latent feature information and the mask data are input into the decoder for data reconstruction to obtain reconstructed data;

[0020] The mask and the encoder are trained with the goal of minimizing the mean square error between the reconstructed data and the sample running state sequence data to obtain the normalized model.

[0021] In one embodiment, acquiring the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed includes:

[0022] Collect the initial operating status sequence data of all energy storage units in the hybrid energy storage system;

[0023] The initial running state sequence data is continuously divided into multiple independent data segments according to a fixed-length time window;

[0024] The multiple data segments are split into multiple continuous and non-overlapping data blocks along the time dimension to obtain the running state sequence data.

[0025] Secondly, this application also provides a state estimation device for an energy storage unit, comprising:

[0026] The first acquisition module is used to acquire the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed;

[0027] The second acquisition module is used to input the operating state sequence data into the normalization model to obtain the normalized feature vector of each energy storage unit;

[0028] The construction module is used to construct a fully connected graph using the normalized feature vectors of each energy storage unit as node features and the relationships between each energy storage unit as edges.

[0029] An estimation module is used to input the fully connected graph into a graph neural network model to obtain state estimation information for each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit using the updated features.

[0030] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0031] Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed;

[0032] The operating state sequence data is input into a normalization model to obtain the normalized feature vector of each energy storage unit;

[0033] A fully connected graph is constructed using the normalized feature vectors of each energy storage unit as node features and the relationships between each energy storage unit as edges.

[0034] The fully connected graph is input into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit using the updated features.

[0035] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0036] Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed;

[0037] The operating state sequence data is input into a normalization model to obtain the normalized feature vector of each energy storage unit;

[0038] A fully connected graph is constructed using the normalized feature vectors of each energy storage unit as node features and the relationships between each energy storage unit as edges.

[0039] The fully connected graph is input into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit using the updated features.

[0040] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0041] Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed;

[0042] The operating state sequence data is input into a normalization model to obtain the normalized feature vector of each energy storage unit;

[0043] A fully connected graph is constructed using the normalized feature vectors of each energy storage unit as node features and the relationships between each energy storage unit as edges.

[0044] The fully connected graph is input into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit using the updated features.

[0045] The aforementioned energy storage unit state estimation method, device, computer equipment, and storage medium, by acquiring the operational state sequence data of all energy storage units in the hybrid energy storage system to be analyzed, can input the operational state sequence data of all energy storage units in the hybrid energy storage system into a normalization model. Through the normalization model, it learns the common characteristics and deep correlations between different energy storage units, obtaining the normalized feature vector of each energy storage unit. This solves the problem of heterogeneous data adaptation among different energy storage units in the hybrid energy storage system, achieving automated and unified representation of heterogeneous data. Then, using the normalized feature vector of each energy storage unit... The eigenvectors are node features, and the relationships between energy storage units are edges. A fully connected graph is constructed, which not only retains the fine electrochemical features of individual units but also incorporates system-level correlation information. This significantly improves the adaptability of the state estimation of each energy storage unit in the hybrid energy storage system to complex system scenarios. The fully connected graph is then input into a graph neural network model. The graph neural network model updates the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units. Using the updated features of the energy storage units, the state estimation information of the energy storage units is obtained, thus realizing the estimation of the state information of each energy storage unit in the hybrid energy storage system. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating the state estimation method for an energy storage unit in one embodiment;

[0048] Figure 2 This is a flowchart illustrating the state estimation method for an energy storage unit in another embodiment;

[0049] Figure 3 This is a flowchart illustrating the state estimation method for an energy storage unit in another embodiment;

[0050] Figure 4 This is a flowchart illustrating the state estimation method for an energy storage unit in another embodiment;

[0051] Figure 5 This is a structural block diagram of the state estimation device for an energy storage unit in one embodiment;

[0052] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0055] In one embodiment, such as Figure 1 As shown, a state estimation method for an energy storage unit is provided. This embodiment illustrates the application of this method to a computer device. It is understood that this method can also be applied to a server, and to a system including both a computer device and a server, and is implemented through the interaction between the computer device and the server. In this embodiment, the method includes the following steps:

[0056] S201, Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed.

[0057] Hybrid energy storage systems refer to systems composed of two or more different types and chemical systems of energy storage units (such as lithium-ion batteries, sodium-ion batteries, and supercapacitors). The operational status sequence data of the energy storage units can include voltage sequence data, current sequence data, and temperature sequence data over time.

[0058] For example, in this embodiment, sensors can be used to collect the operating status data of each energy storage unit in the hybrid energy storage system in the time dimension to obtain the operating status sequence data of all energy storage units in the hybrid energy storage system.

[0059] S202, input the operating status sequence data into the normalization model to obtain the normalized feature vector of each energy storage unit.

[0060] For example, in this embodiment, the operating status sequence data of all energy storage units are input into a normalization model. The normalization model can be used to mine the common features between different energy storage units and determine the normalized feature vector of each energy storage unit with a fixed dimension.

[0061] Optionally, the normalization model in this embodiment can be any one of the following: Masked Autoencoder (MAE) model, contrastive learning framework model, variational autoencoder model, and denoising autoencoder model. It is understood that the MAE model randomly masks (hides) part of the input data and trains the model to reconstruct the masked portion, thereby learning a deep and robust feature representation of the data; contrastive learning framework models such as SimCLR and MoCo construct positive samples by enhancing the same battery data and constructing negative samples with other battery data, learning cross-system universal features; variational autoencoders (VAE) normalize heterogeneous data by introducing probability distribution constraints, achieving feature extraction results consistent with MAE; and denoising autoencoders (DAE) force the model to learn robust features by adding noise, achieving feature extraction results consistent with MAE.

[0062] S203 uses the normalized feature vectors of each energy storage unit as node features and the relationships between energy storage units as edges to construct a fully connected graph.

[0063] For example, in this embodiment, each energy storage unit in the hybrid energy storage system can be regarded as a node in a graph. It is assumed that each energy storage unit has potential connections with all other energy storage units, and an initial connection graph is constructed. Then, the connections between the energy storage units are determined. The normalized feature vectors of each energy storage unit are used to initialize the features of each node in the initial connection graph, and the connections between the energy storage units are used as edges to construct a fully connected graph. Optionally, the connections between the energy storage units can be the similarity between them; for example, the weighted cosine similarity between them can be used.

[0064] For example, when initializing the features of each node in the initial connection graph using the normalized feature vector of each energy storage unit, the initial feature matrix is ​​set according to time steps. Represented as: In the formula, for Original feature data matrix at each time step, For feature type constraint functions, The number of energy storage units. The feature dimension (including core parameters such as current and temperature).

[0065] S204. Input the fully connected graph into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit by using the updated features of the energy storage unit.

[0066] The graph neural network model can be a GraphSAGE (Graph Sample and Aggregate) graph neural network model, used to update the features of each energy storage unit by aggregating features from at least one layer of similar energy storage units. The updated features are then used to obtain the state estimation information of the energy storage unit. For example, in this embodiment, the aggregation formula for feature aggregation of each energy storage unit can be: In the formula, This indicates the updated characteristics of the energy storage unit. Indicates the characteristics of the energy storage unit, For adjacent energy storage units of an energy storage unit, It is the identity matrix. For degree matrix, For learnable parameter matrix, is the activation function. It can be understood that after one to multiple layers of aggregation, the final output features of each node in the fully connected graph contain not only its own initial features but also information from other similar or related units within the hybrid energy storage system. This effectively captures complex relationships such as collaborative aging and temperature cross-influence among energy storage units in the hybrid energy storage system, thereby optimizing and enhancing the features of the mask autoencoder.

[0067] As another optional implementation, the graph neural network model in this embodiment can also be any one of the graph convolutional network model, graph attention network model, and graph isomorphic network model. It should be noted that although the graph convolutional network model aggregates neighbor node features through local convolution of the adjacency matrix, the graph attention network assigns differentiated weights to neighbor nodes using an attention mechanism, and the graph isomorphic network captures graph isomorphic features through stacking-aggregation-update operations, these models, despite their different aggregation methods, can all uncover the relationships between energy storage units, such as collaborative aging and temperature effects, and obtain the updated features of each energy storage unit.

[0068] Furthermore, the state estimation information for each energy storage unit can include health status estimates, remaining useful life estimates, fault diagnosis estimates, and anomaly detection estimates. Depending on the specific application requirements, the graph neural network model can be trained with different outputs to obtain the corresponding state estimation information. It's understandable that different state estimations can be achieved by adjusting the output dimension of the regression head of the graph neural network model and retaining or replacing the loss function. For example, for fault diagnosis estimation, the downstream head can be changed to a classification network, and the loss function can be replaced with a cross-entropy loss function to achieve fault classification and early warning. For anomaly detection estimation, an unsupervised reconstruction network can be used to identify anomalies through reconstruction errors. Expanding the state estimation information for each energy storage unit does not require changing the preceding processes; only the downstream modules need adjustment.

[0069] In the aforementioned energy storage unit state estimation method, by acquiring the operating state sequence data of all energy storage units in the hybrid energy storage system to be analyzed, the operating state sequence data of all energy storage units in the hybrid energy storage system can be input into a normalization model. Through the normalization model, the common characteristics and deep correlations between different energy storage units are learned, resulting in the normalized feature vector of each energy storage unit. This solves the problem of heterogeneous data adaptation among different energy storage units in the hybrid energy storage system, achieving automated and unified representation of heterogeneous data. Then, the normalized feature vector of each energy storage unit is used as the node feature vector. The relationships between energy storage units are used as edges to construct a fully connected graph. This not only preserves the detailed electrochemical characteristics of individual units but also incorporates system-level correlation information, significantly improving the adaptability of state estimation of each energy storage unit in a hybrid energy storage system to complex system scenarios. This fully connected graph is then input into a graph neural network model. The graph neural network model updates the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units. Using the updated features of the energy storage units, the state estimation information of the energy storage units is obtained, thus realizing the estimation of the state information of each energy storage unit in the hybrid energy storage system.

[0070] This embodiment will explain in detail the process of determining similar energy storage units for an energy storage unit. In an exemplary embodiment, such as... Figure 2 As shown, the above method also includes:

[0071] S301, For each energy storage unit, the weighted cosine similarity between the energy storage unit and its neighboring energy storage units is determined by using the feature vector of the energy storage unit, the feature vector of the energy storage unit, and the preset feature weights.

[0072] The feature vector of the energy storage unit may include the current feature vector, temperature feature vector, capacity feature vector, voltage feature vector, etc. of the energy storage unit, and feature weights are assigned to each feature vector of the energy storage unit. That is to say, the preset feature weights in this embodiment may include the feature weights of various different feature vectors of the energy storage unit, and the feature weights of different feature vectors may be the same or different.

[0073] For example, for each energy storage unit, the formula can be used. Determine the weighted cosine similarity between the energy storage unit and its adjacent energy storage units, where, This represents the weighted cosine similarity between an energy storage unit and its adjacent energy storage units. Indicates energy storage unit eigenvectors, These represent the current, temperature, capacity, and voltage of the energy storage unit, respectively. Indicates energy storage unit The L2 norm, The feature weights represent the feature vectors of each energy storage unit. By introducing the feature weights of the feature vectors, the influence of key features on the weighted cosine similarity calculation can be amplified.

[0074] Optionally, in this embodiment, for each energy storage unit, it can be based on the first The adjacent candidate energy storage cells of the energy storage unit are determined by considering the candidate energy storage cells adjacent to the current energy storage unit, the preset sampling quantity, and the data sampling strategy. The sampling logic for adjacent energy storage units in a layer is as follows: In the formula, For the first Adjacent energy storage units of the layered energy storage unit, For the first Candidate energy storage cells adjacent to the layer and energy storage cell. This is the preset number of samples.

[0075] S302, adjacent energy storage units with a weighted cosine similarity greater than the similarity threshold are identified as similar energy storage units.

[0076] For example, in this embodiment, it can be based on a similarity matrix. ,Pick Quantiles are adaptive thresholds: Construct an adjacency matrix In the formula, The indicator function is used, and then isolated nodes of adjacent matrices are removed. The adjacent energy storage units whose weighted cosine similarity is greater than the similarity threshold are identified as similar energy storage units of that energy storage unit.

[0077] In this embodiment, for each energy storage unit, the weighted cosine similarity between the energy storage unit and its neighboring energy storage units can be determined relatively quickly using the feature vector of the energy storage unit, the feature vectors of the adjacent energy storage units, and the preset feature weights. This allows for the rapid identification of adjacent energy storage units with a weighted cosine similarity greater than the similarity threshold as similar energy storage units, thus improving the efficiency of identifying similar energy storage units.

[0078] In the scenario described above, where the operating state sequence data of all energy storage units in a hybrid energy storage system are input into a normalization model to obtain the normalized feature vector of each energy storage unit, this normalization model is a pre-trained model. This embodiment will explain the training process of the normalization model. In one embodiment, the normalization model includes a mask, an encoder, and a decoder, such as... Figure 3 As shown, the training process of the above normalization model includes:

[0079] S401, input the sample operating status sequence data of all energy storage units in the hybrid energy storage system into the masker, and use the masker to perform masking processing on the sample operating status sequence data with a high proportion of random masking to obtain the masked sample operating status sequence data; the masked sample operating status sequence data includes masked data and unmasked data.

[0080] In this embodiment, the masker can use a high-proportion random mask to mask the sample operating state sequence data of all energy storage units in the hybrid energy storage system, obtaining masked sample operating state sequence data including both masked and unmasked data. It is understood that a high-proportion random mask is used when masking the sample operating state sequence data; therefore, most of the data in the sample operating state sequence data is masked, while a small portion remains unmasked.

[0081] S402, input the unmasked data and the position encoding of the unmasked data into the encoder to obtain the potential feature information of the unmasked data.

[0082] In this embodiment, the unmasked data and the position encoding of the unmasked data in the sample running state sequence data can be input into the encoder. The encoder learns the deep correlation of the visible data that is not masked, and obtains the potential feature information of the unmasked data.

[0083] S403 inputs the location encoding of the latent feature information and mask data into the decoder for data reconstruction to obtain the reconstructed data.

[0084] In this embodiment, the decoder can reconstruct the masked operating state sequence data in the hybrid energy storage system based on the potential feature information of the unmasked data and the position encoding of all masked data, thus completing the reconstruction task of the original complete data segment and obtaining the reconstructed data.

[0085] S404 trains the mask and encoder with the goal of minimizing the mean square error between the reconstructed data and the sample running state sequence data, and obtains the normalized model.

[0086] In this embodiment, the training objective can be to minimize the mean square error (MSE) between the reconstructed data and the original data, forcing the encoder to focus on the essential electrochemical dynamic characteristics behind the sample operating state sequence data (e.g., voltage sequence data, current sequence data) rather than the surface numerical differences. The mask and encoder are trained, and the encoder is retained after training to obtain a normalized model. This allows the encoder to stably output a normalized feature vector of fixed dimensions for any input data segment (lithium battery or sodium battery), serving as a unified and universal representation for different batteries.

[0087] In this embodiment, the sample operating state sequence data of all energy storage units in the hybrid energy storage system is input into the mask of the normalization model. The mask can be used to mask the sample operating state sequence data with a high proportion of random masking to obtain masked sample operating state sequence data including masked data and unmasked data. Then, the unmasked data and the position encoding of the unmasked data are input into the encoder of the normalization model to obtain the latent feature information of the unmasked data. Then, the latent feature information of the unmasked data and the position encoding of the masked data are input into the decoder of the normalization model to reconstruct the masked sample operating state sequence data to obtain reconstructed data. Then, the mask and encoder can be trained with the goal of minimizing the mean square error between the reconstructed data and the sample operating state sequence data to obtain a trained normalization model. This allows the trained normalization model to stably output a normalized feature vector of fixed dimension for any input data segment (lithium battery or sodium battery), serving as a unified and universal representation for different batteries.

[0088] This embodiment will explain in detail the process of acquiring the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed. In one embodiment, such as... Figure 4 As shown, the above S201 includes:

[0089] S501 collects the initial operating status sequence data of all energy storage units in the hybrid energy storage system.

[0090] Optionally, in this embodiment, various sensors can be used to collect the initial operating status data of all energy storage units in the hybrid energy storage system over time, thereby obtaining the initial operating status sequence data of each energy storage unit. For example, the operating status data of the energy storage unit may include key parameters such as voltage, current, and temperature of the energy storage unit.

[0091] S502 divides the continuous data of each energy storage unit into multiple independent data segments according to a fixed-length time window.

[0092] In this embodiment, the initial running state sequence data can be continuously divided into multiple independent data segments according to a fixed-length time window. It is understood that when continuously segmenting the initial running state sequence data according to a fixed-length time window, the initial running state sequence data may be segmented into multiple independent data segments sequentially according to the segmentation order, and there may be some overlap in the data segments during segmentation. This embodiment does not limit the segmentation method of the initial running state sequence data.

[0093] S503 splits multiple data segments of each energy storage unit into multiple continuous and non-overlapping data blocks in the time dimension to obtain the operating status sequence data.

[0094] In this embodiment, since there may be overlapping data segments among the multiple independent data segments obtained by continuously dividing the initial operating state sequence data, as an optional implementation method, the multiple data segments can also be split into multiple continuous and non-overlapping data blocks in the time dimension to obtain the operating state sequence data of all energy storage units in the hybrid energy storage system, thereby avoiding overlapping data blocks in the obtained operating state sequence data.

[0095] In this embodiment, by collecting the initial operating state sequence data of all energy storage units in the hybrid energy storage system, the initial operating state sequence data can be continuously divided into multiple independent data segments according to a fixed time length. Furthermore, the multiple data segments can be divided into multiple continuous and non-overlapping data blocks in the time dimension. By removing overlapping data blocks, the accuracy of the final operating state sequence data of all energy storage units in the hybrid energy storage system is ensured.

[0096] It should be noted that the core of this application's embodiments lies in constructing a unified end-to-end framework through a normalization model (masked autoencoder) to normalize heterogeneous data and a cascaded architecture of a graph neural network model. This framework addresses the problem of heterogeneous data adaptation and collaborative state estimation for different chemical system units (such as lithium-ion batteries and sodium-ion batteries) in hybrid energy storage systems, achieving high-precision and universal estimation of energy storage unit state information (e.g., estimation of energy storage unit health status information). Specifically, this application addresses the issue of numerical differences in the original data such as voltage and current of lithium-ion batteries and sodium-ion batteries in hybrid energy storage systems. It innovatively employs a masked autoencoder (MAE) for self-supervised pre-training: through a high-proportion random mask and data block reconstruction task, the MAE encoder is forced to learn the common electrochemical dynamic characteristics behind the data of different chemical systems, generating a fixed-dimensional normalized feature vector independent of specific chemical systems. This method overcomes the limitations of traditional manual feature engineering, achieving automated and unified representation of heterogeneous data, laying the foundation for subsequent cross-unit collaborative modeling, and directly improving the model's adaptability to different types of energy storage units. This paper innovatively designs a two-level cascaded architecture of feature normalization and relationship modeling: Normalized features are extracted from heterogeneous data of energy storage units using Modeling Effects (MAE); a graph neural network is introduced, using the normalized feature vectors as graph node features. Through hierarchical neighbor sampling and feature aggregation, potential correlations such as co-aging and temperature coupling between units in the hybrid system are captured. The two-level modules work synergistically, preserving the fine electrochemical characteristics of energy storage units while incorporating system-level correlation information. Compared to a single MAE or graph neural network model, this significantly improves the adaptability of energy storage unit state estimation to complex system scenarios. Addressing the diverse unit types and complex relationships in hybrid energy storage systems, this application innovatively constructs a fully connected initial graph structure and optimizes node features through the dynamic feature aggregation mechanism of the GraphSAGE graph neural network model: during aggregation, the contribution weights of neighboring nodes are dynamically adjusted using a learnable parameter matrix and activation function, ensuring that the final node features include both their own MAE normalized features and aging mode information from similar units. This strategy overcomes the limitations of traditional independent unit estimation, using system-wide correlation information to correct single-unit feature biases, directly improving the accuracy and robustness of energy storage unit state information estimation. Furthermore, this application embodiment achieves synchronous state information estimation for all heterogeneous units in a hybrid system through modular integration of MAE and GraphSAGE graph neural network models, combined with a lightweight MLP regression head. The fine-tuning stage employs a hierarchical learning rate strategy, ensuring the reuse of general features while adapting to the state information estimation task. This framework eliminates the need for separate modeling for different types of units, simplifying the deployment process of the battery management system for hybrid energy storage systems, improving the model's engineering practicality and versatility, and reducing system modeling and maintenance costs.

[0097] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0098] Based on the same inventive concept, this application also provides a state estimation device for an energy storage unit to implement the state estimation method for the energy storage unit described above. The solution provided by this device is similar to the implementation described in the above method. Therefore, the specific limitations in one or more state estimation device embodiments for energy storage units provided below can be found in the limitations of the state estimation method for energy storage units described above, and will not be repeated here.

[0099] In one exemplary embodiment, such as Figure 5 As shown, a state estimation device for an energy storage unit is provided, comprising: a first acquisition module, a second acquisition module, a construction module, and an estimation module, wherein:

[0100] The first acquisition module is used to acquire the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed.

[0101] The second acquisition module is used to input the operating status sequence data into the normalization model to obtain the normalized feature vector of each energy storage unit.

[0102] The module is used to construct a fully connected graph using the normalized feature vectors of each energy storage unit as node features and the relationships between energy storage units as edges.

[0103] The estimation module is used to input the fully connected graph into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit using the updated features.

[0104] The state estimation device for the energy storage unit provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0105] Based on the above embodiments, optionally, the above apparatus further includes: a first determining module and a second determining module, wherein:

[0106] The first determining module is used to determine the weighted cosine similarity between an energy storage unit and its neighboring energy storage units for each energy storage unit by using the feature vector of the energy storage unit, the feature vector of the energy storage unit's neighboring energy storage units, and the preset feature weights.

[0107] The second determining module is used to determine adjacent energy storage units with a weighted cosine similarity greater than a similarity threshold as similar energy storage units.

[0108] The state estimation device for the energy storage unit provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0109] Based on the above embodiments, optionally, the above apparatus further includes: a third determining module, wherein:

[0110] The third determining module is used to determine, for each energy storage unit, based on the first... The adjacent energy storage units of the energy storage unit are determined by selecting candidate energy storage units adjacent to the energy storage unit, setting the preset sampling quantity and data sampling strategy.

[0111] The state estimation device for the energy storage unit provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0112] Based on the above embodiments, optionally, the aggregation formula for feature aggregation of each energy storage unit is as follows: In the formula, This indicates the updated characteristics of the energy storage unit. Indicates the characteristics of the energy storage unit, For adjacent energy storage units of an energy storage unit, It is the identity matrix. For degree matrix, For learnable parameter matrix, This is the activation function.

[0113] The state estimation device for the energy storage unit provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0114] Based on the above embodiments, optionally, the normalization model includes a mask, an encoder, and a decoder; the apparatus further includes: a first processing module, a second processing module, a reconstruction module, and a training module, wherein:

[0115] The first processing module is used to input the sample operating status sequence data of all energy storage units in the hybrid energy storage system into the masker, and to perform masking processing on the sample operating status sequence data by using a high-ratio random mask to obtain the masked sample operating status sequence data; the masked sample operating status sequence data includes masked data and unmasked data.

[0116] The second processing module is used to input the unmasked data and the position encoding of the unmasked data into the encoder to obtain the potential feature information of the unmasked data.

[0117] The reconstruction module is used to input the location encoding of potential feature information and mask data into the decoder for data reconstruction to obtain reconstructed data.

[0118] The training module is used to train the mask and encoder with the goal of minimizing the mean square error between the reconstructed data and the sample running state sequence data, so as to obtain a normalized model.

[0119] The state estimation device for the energy storage unit provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0120] Based on the above embodiments, optionally, the first acquisition module includes: a collection unit, a segmentation unit, and a splitting unit, wherein:

[0121] The acquisition unit is used to acquire the initial operating status sequence data of all energy storage units in the hybrid energy storage system.

[0122] The segmentation unit is used to continuously divide the initial running state sequence data into multiple independent data segments according to a fixed-length time window.

[0123] The splitting unit is used to split multiple data segments into multiple continuous and non-overlapping data blocks in the time dimension to obtain the running state sequence data.

[0124] The state estimation device for the energy storage unit provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0125] The various modules in the state estimation device of the aforementioned energy storage unit can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware within or independently of the processor in a computer device, or stored in software within the memory of a computer device, so that the processor can call and execute the operations corresponding to each module.

[0126] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores the operating state sequence data of all energy storage units in the hybrid energy storage system to be analyzed. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for estimating the state of energy storage units.

[0127] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0128] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0129] Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed;

[0130] Input the operating status sequence data into the normalization model to obtain the normalized feature vector of each energy storage unit;

[0131] A fully connected graph is constructed using the normalized feature vectors of each energy storage unit as node features and the relationships between energy storage units as edges.

[0132] The fully connected graph is input into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and then use the updated features of the energy storage unit to obtain the state estimation information of the energy storage unit.

[0133] Optionally, in this embodiment, the aggregation formula for feature aggregation of each energy storage unit is as follows: In the formula, This indicates the updated characteristics of the energy storage unit. Indicates the characteristics of the energy storage unit, For adjacent energy storage units of an energy storage unit, It is the identity matrix. For degree matrix, For learnable parameter matrix, This is the activation function.

[0134] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0135] For each energy storage unit, the weighted cosine similarity between the energy storage unit and its neighboring energy storage units is determined by using the feature vector of the energy storage unit, the feature vector of the energy storage unit, and the preset feature weights.

[0136] Adjacent energy storage units with a weighted cosine similarity greater than the similarity threshold are identified as similar energy storage units.

[0137] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0138] For each energy storage unit, based on the first The adjacent energy storage units of the energy storage unit are determined by selecting candidate energy storage units adjacent to the energy storage unit, setting the preset sampling quantity and data sampling strategy.

[0139] In one embodiment, the normalization model includes a mask, an encoder, and a decoder; the processor also performs the following steps when executing the computer program:

[0140] The sample operating status sequence data of all energy storage units in the hybrid energy storage system is input into a masker. The masker performs masking processing on the sample operating status sequence data using a high-ratio random mask, resulting in masked sample operating status sequence data. The masked sample operating status sequence data includes masked data and unmasked data.

[0141] The unmasked data and its position encoding are input into the encoder to obtain the latent feature information of the unmasked data;

[0142] The location encodings of latent feature information and mask data are input into the decoder for data reconstruction to obtain reconstructed data.

[0143] The mask and encoder are trained with the goal of minimizing the mean square error between the reconstructed data and the sample running state sequence data, resulting in a normalized model.

[0144] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0145] Collect initial operating status sequence data of all energy storage units in the hybrid energy storage system;

[0146] The initial running state sequence data is continuously divided into multiple independent data segments according to a fixed-length time window;

[0147] Multiple data segments are split into multiple continuous and non-overlapping data blocks along the time dimension to obtain the running state sequence data.

[0148] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0149] Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed;

[0150] Input the operating status sequence data into the normalization model to obtain the normalized feature vector of each energy storage unit;

[0151] A fully connected graph is constructed using the normalized feature vectors of each energy storage unit as node features and the relationships between energy storage units as edges.

[0152] The fully connected graph is input into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and then use the updated features of the energy storage unit to obtain the state estimation information of the energy storage unit.

[0153] Optionally, in this embodiment, the aggregation formula for feature aggregation of each energy storage unit is as follows: In the formula, This indicates the updated characteristics of the energy storage unit. Indicates the characteristics of the energy storage unit, For adjacent energy storage units of an energy storage unit, It is the identity matrix. For degree matrix, For learnable parameter matrix, This is the activation function.

[0154] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0155] For each energy storage unit, the weighted cosine similarity between the energy storage unit and its neighboring energy storage units is determined by using the feature vector of the energy storage unit, the feature vector of the energy storage unit, and the preset feature weights.

[0156] Adjacent energy storage units with a weighted cosine similarity greater than the similarity threshold are identified as similar energy storage units.

[0157] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0158] For each energy storage unit, based on the first The adjacent energy storage units of the energy storage unit are determined by selecting candidate energy storage units adjacent to the energy storage unit, setting the preset sampling quantity and data sampling strategy.

[0159] In one embodiment, the normalization model includes a mask, an encoder, and a decoder; when the computer program is executed by a processor, it also performs the following steps:

[0160] The sample operating status sequence data of all energy storage units in the hybrid energy storage system is input into a masker. The masker performs masking processing on the sample operating status sequence data using a high-ratio random mask, resulting in masked sample operating status sequence data. The masked sample operating status sequence data includes masked data and unmasked data.

[0161] The unmasked data and its position encoding are input into the encoder to obtain the latent feature information of the unmasked data;

[0162] The location encodings of latent feature information and mask data are input into the decoder for data reconstruction to obtain reconstructed data.

[0163] The mask and encoder are trained with the goal of minimizing the mean square error between the reconstructed data and the sample running state sequence data, resulting in a normalized model.

[0164] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0165] Collect initial operating status sequence data of all energy storage units in the hybrid energy storage system;

[0166] The initial running state sequence data is continuously divided into multiple independent data segments according to a fixed-length time window;

[0167] Multiple data segments are split into multiple continuous and non-overlapping data blocks along the time dimension to obtain the running state sequence data.

[0168] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0169] Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed;

[0170] Input the operating status sequence data into the normalization model to obtain the normalized feature vector of each energy storage unit;

[0171] A fully connected graph is constructed using the normalized feature vectors of each energy storage unit as node features and the relationships between energy storage units as edges.

[0172] The fully connected graph is input into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and then use the updated features of the energy storage unit to obtain the state estimation information of the energy storage unit.

[0173] Optionally, in this embodiment, the aggregation formula for feature aggregation of each energy storage unit is as follows: In the formula, This indicates the updated characteristics of the energy storage unit. Indicates the characteristics of the energy storage unit, For adjacent energy storage units of an energy storage unit, It is the identity matrix. For degree matrix, For learnable parameter matrix, This is the activation function.

[0174] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0175] For each energy storage unit, the weighted cosine similarity between the energy storage unit and its neighboring energy storage units is determined by using the feature vector of the energy storage unit, the feature vector of the energy storage unit, and the preset feature weights.

[0176] Adjacent energy storage units with a weighted cosine similarity greater than the similarity threshold are identified as similar energy storage units.

[0177] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0178] For each energy storage unit, based on the first The adjacent energy storage units of the energy storage unit are determined by selecting candidate energy storage units adjacent to the energy storage unit, setting the preset sampling quantity and data sampling strategy.

[0179] In one embodiment, the normalization model includes a mask, an encoder, and a decoder; when the computer program is executed by a processor, it also performs the following steps:

[0180] The sample operating status sequence data of all energy storage units in the hybrid energy storage system is input into a masker. The masker performs masking processing on the sample operating status sequence data using a high-ratio random mask, resulting in masked sample operating status sequence data. The masked sample operating status sequence data includes masked data and unmasked data.

[0181] The unmasked data and its position encoding are input into the encoder to obtain the latent feature information of the unmasked data;

[0182] The location encodings of latent feature information and mask data are input into the decoder for data reconstruction to obtain reconstructed data.

[0183] The mask and encoder are trained with the goal of minimizing the mean square error between the reconstructed data and the sample running state sequence data, resulting in a normalized model.

[0184] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0185] Collect initial operating status sequence data of all energy storage units in the hybrid energy storage system;

[0186] The initial running state sequence data is continuously divided into multiple independent data segments according to a fixed-length time window;

[0187] Multiple data segments are split into multiple continuous and non-overlapping data blocks along the time dimension to obtain the running state sequence data.

[0188] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0189] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0190] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A state estimation method for an energy storage unit, characterized in that, The method includes: Obtain the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed; The operating state sequence data is input into a normalization model to obtain the normalized feature vector of each energy storage unit; A fully connected graph is constructed using the normalized feature vectors of each energy storage unit as node features and the relationships between each energy storage unit as edges. The fully connected graph is input into the graph neural network model to obtain the state estimation information of each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit using the updated features.

2. The method according to claim 1, characterized in that, The method further includes: For each energy storage unit, the weighted cosine similarity between the energy storage unit and its neighboring energy storage units is determined using the feature vector of the energy storage unit, the feature vectors of the energy storage unit and its neighboring energy storage units, and a preset feature weight. Adjacent energy storage units with a weighted cosine similarity greater than a similarity threshold are identified as similar energy storage units.

3. The method according to claim 2, characterized in that, The method further includes: For each energy storage unit, based on the first The adjacent energy storage units of the energy storage unit are determined by considering the candidate energy storage units adjacent to the energy storage unit, the preset sampling quantity, and the data sampling strategy.

4. The method according to claim 2, characterized in that, The aggregation formula for feature aggregation of each energy storage unit is as follows: In the formula, This indicates the updated characteristics of the energy storage unit. Indicates the characteristics of the energy storage unit, For adjacent energy storage units of an energy storage unit, It is the identity matrix. For degree matrix, For learnable parameter matrix, This is the activation function.

5. The method according to any one of claims 1 to 4, characterized in that, The normalization model includes a mask, an encoder, and a decoder; the training process of the normalization model includes: The sample operating status sequence data of all energy storage units in the hybrid energy storage system is input into the masker. The masker performs masking processing on the sample operating status sequence data using a high-proportion random mask to obtain masked sample operating status sequence data. The masked sample operating status sequence data includes masked data and unmasked data. The unmasked data and its position encoding are input into the encoder to obtain the latent feature information of the unmasked data; The location encodings of the latent feature information and the mask data are input into the decoder for data reconstruction to obtain reconstructed data; The mask and the encoder are trained with the goal of minimizing the mean square error between the reconstructed data and the sample running state sequence data to obtain the normalized model.

6. The method according to any one of claims 1 to 4, characterized in that, The acquisition of the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed includes: Collect the initial operating status sequence data of all energy storage units in the hybrid energy storage system; The initial running state sequence data is continuously divided into multiple independent data segments according to a fixed-length time window; The multiple data segments are split into multiple continuous and non-overlapping data blocks along the time dimension to obtain the running state sequence data.

7. A state estimation device for an energy storage unit, characterized in that, The device includes: The first acquisition module is used to acquire the operating status sequence data of all energy storage units in the hybrid energy storage system to be analyzed; The second acquisition module is used to input the operating state sequence data into the normalization model to obtain the normalized feature vector of each energy storage unit; The construction module is used to construct a fully connected graph using the normalized feature vectors of each energy storage unit as node features and the relationships between each energy storage unit as edges. An estimation module is used to input the fully connected graph into a graph neural network model to obtain state estimation information for each energy storage unit. The graph neural network model is used to update the features of each energy storage unit by aggregating the features of at least one layer of similar energy storage units, and to obtain the state estimation information of the energy storage unit using the updated features.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images