Sodium ion battery soc estimation method based on cnn-resnet-lstm hybrid network

Abstract

The application relates to a sodium ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network. The method comprises the following steps: obtaining a battery operation data sequence of a target battery to be estimated in a historical time period, wherein the battery operation data sequence comprises battery operation data of the target battery at multiple historical time points in the historical time period; extracting features from the battery operation data in the battery operation data sequence, obtaining a battery operation feature sequence based on battery operation features extracted from each battery operation data; optimizing the features of the battery operation features in the battery operation feature sequence, obtaining a battery optimization feature sequence based on battery optimization features extracted from each battery operation feature; and predicting the state of charge of the target battery based on the battery optimization feature sequence to obtain the predicted state of charge of the target battery in a target time period. The method can improve the accuracy of battery state of charge estimation.

Description

Technical Field

[0001] This application relates to the field of power prediction technology, and in particular to a method for estimating the state of charge (SOC) of sodium-ion batteries based on a CNN-ResNet-LSTM hybrid network. Background Technology

[0002] Sodium-ion batteries have gradually become an important alternative to lithium-ion batteries due to their abundant resources, low cost, and good safety performance, and are suitable for scenarios such as grid energy storage, electric mobility, and home energy management. However, sodium-ion batteries still face a series of challenges in practical use, including limited cycle life, fluctuations in rate performance, and the complexity of battery state estimation, especially the problem of accurate assessment of state of charge.

[0003] In traditional techniques, the state of charge is dynamically estimated using various physical models, such as extended Kalman filtering or unscented Kalman filtering.

[0004] However, traditional technologies rely on accurate battery parameter identification, and their filtering performance degrades significantly when there is a high degree of nonlinearity or drastic changes in operating conditions, resulting in low accuracy in battery power prediction. Summary of the Invention

[0005] Therefore, it is necessary to provide a sodium-ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network that can improve the accuracy of power prediction, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a method for estimating the state of charge (SOC) of sodium-ion batteries based on a CNN-ResNet-LSTM hybrid network, including:

[0007] Obtain the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0008] Feature extraction is performed on the battery operation data in the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data.

[0009] The battery operation features in the battery operation feature sequence are optimized, and the battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operation feature.

[0010] Based on the battery optimization feature sequence, the state of charge of the target battery is predicted to obtain the predicted state of charge of the target battery in the target time period.

[0011] The predicted state of charge is obtained through a CNN-ResNet-LSTM hybrid network.

[0012] In one embodiment, feature extraction is performed on the battery operation data in the battery operation data sequence. Based on the battery operation features extracted for each battery operation data, a battery operation feature sequence is obtained, including:

[0013] For each battery operation data point in the battery operation data sequence, the battery operation data is normalized, and the normalized battery operation data is input into the convolutional neural network layer for feature extraction to obtain the battery operation features corresponding to the battery operation data.

[0014] Based on the battery operating characteristics corresponding to each battery operating data, a battery operating characteristic sequence is obtained.

[0015] In one embodiment, feature optimization is performed on the battery operating features in the battery operating feature sequence. Based on the battery optimization features extracted for each battery operating feature, a battery optimization feature sequence is obtained, including:

[0016] For each battery operating feature in the battery operating feature sequence, feature extraction is performed on the battery operating feature input to obtain the first optimized feature of the battery operating feature;

[0017] The battery operating characteristics are scaled to obtain the second optimized features of the battery operating characteristics. The scale of the first optimized features and the second optimized features are consistent.

[0018] The first optimization feature and the second optimization feature are fused to obtain the battery optimization feature corresponding to the battery operation feature;

[0019] Based on the battery optimization features corresponding to each battery's operating characteristics, a battery optimization feature sequence is obtained.

[0020] In one embodiment, the battery operating data of the target battery at a historical time includes voltage data and / or current data of the target battery at that historical time.

[0021] In one embodiment, the predicted state of charge (SOC) is obtained through a trained sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network. The training process of the sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network includes:

[0022] Obtain the sample battery operation data sequence corresponding to the sample battery. The sample battery operation data sequence is obtained during the open circuit voltage test of the sample battery.

[0023] Obtain the actual remaining battery power sequence corresponding to the sample battery operation data sequence;

[0024] The sample battery running data sequence is input into the sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network to be trained, and the predicted state of charge sequence output by the power prediction model is obtained.

[0025] Based on the difference between the actual remaining charge sequence and the predicted state of charge sequence, a sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network is trained to obtain the trained sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network.

[0026] In one embodiment, obtaining the sample battery operation data sequence corresponding to the sample battery includes:

[0027] The sample battery is subjected to open-circuit voltage test, and the test voltage sequence and test current sequence of the sample battery during the charging and discharging process are collected. Based on the test voltage sequence and test current sequence, the sample battery operation data sequence corresponding to the sample battery is generated.

[0028] The charging and discharging process includes a charging phase and a discharging phase. The discharging phase is initiated after the charging phase ends and the sample battery has been left to stand for a certain period of time.

[0029] Secondly, this application also provides a sodium-ion battery SOC estimation device based on a CNN-ResNet-LSTM hybrid network, comprising:

[0030] The acquisition module is used to acquire the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0031] The extraction module is used to extract features from the battery operation data sequence, and obtain a battery operation feature sequence based on the battery operation features extracted for each battery operation data.

[0032] The optimization module is used to optimize the battery operating features in the battery operating feature sequence. Based on the battery optimization features extracted for each battery operating feature, a battery optimization feature sequence is obtained.

[0033] The prediction module is used to predict the state of charge of the target battery based on the battery optimization feature sequence, so as to obtain the predicted state of charge of the target battery in the target time period; wherein, the predicted state of charge is obtained by a CNN-ResNet-LSTM hybrid network.

[0034] 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:

[0035] Obtain the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0036] Feature extraction is performed on the battery operation data in the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data.

[0037] The battery operation features in the battery operation feature sequence are optimized, and the battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operation feature.

[0038] Based on the battery optimization feature sequence, the state of charge of the target battery is predicted to obtain the predicted state of charge of the target battery in the target time period.

[0039] The predicted state of charge is obtained through a CNN-ResNet-LSTM hybrid network.

[0040] 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:

[0041] Obtain the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0042] Feature extraction is performed on the battery operation data in the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data.

[0043] The battery operation features in the battery operation feature sequence are optimized, and the battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operation feature.

[0044] Based on the battery optimization feature sequence, the state of charge of the target battery is predicted to obtain the predicted state of charge of the target battery in the target time period.

[0045] The predicted state of charge is obtained through a CNN-ResNet-LSTM hybrid network.

[0046] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0047] Obtain the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0048] Feature extraction is performed on the battery operation data in the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data.

[0049] The battery operation features in the battery operation feature sequence are optimized, and the battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operation feature.

[0050] Based on the battery optimization feature sequence, the state of charge of the target battery is predicted to obtain the predicted state of charge of the target battery in the target time period.

[0051] The predicted state of charge is obtained through a CNN-ResNet-LSTM hybrid network.

[0052] The aforementioned method, apparatus, equipment, medium, and product for estimating the state of charge (SOC) of sodium-ion batteries based on a CNN-ResNet-LSTM hybrid network provide a data foundation for subsequent analysis by acquiring battery operation data sequences of the target battery over historical time periods. Then, feature extraction is performed on the battery operation data to obtain a battery operation feature sequence, efficiently capturing local features related to the state of charge from the raw data. Subsequently, feature optimization is performed on the battery operation feature sequence to obtain an optimized battery feature sequence, enhancing the robustness of the features. Finally, based on the optimized battery features at multiple historical moments, the state of charge is predicted, achieving accurate estimation of the remaining battery capacity. Attached Figure Description

[0053] 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.

[0054] Figure 1 This is an application environment diagram of a sodium-ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network in one embodiment;

[0055] Figure 2 This is a flowchart illustrating a sodium-ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network in one embodiment.

[0056] Figure 3 This is a schematic diagram of the architecture of a CNN-ResNet-LSTM hybrid network in one embodiment;

[0057] Figure 4 This is a schematic diagram comparing the SOC estimation results of the training set with the actual SOC in another embodiment;

[0058] Figure 5 This is a schematic diagram comparing the SOC estimation results of the test set with the actual SOC in another embodiment;

[0059] Figure 6 This is a schematic diagram of the SOC estimation error for the training and test sets in another embodiment.

[0060] Figure 7 This is a block diagram of a sodium-ion battery SOC estimation device based on a CNN-ResNet-LSTM hybrid network in one embodiment.

[0061] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0062] 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.

[0063] 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.

[0064] The sodium-ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Terminal 102 sends a power prediction request to server 104. Server 104 responds to the power prediction request by acquiring the battery operation data sequence of the target battery to be estimated over a historical time period. The battery operation data sequence includes battery operation data of the target battery at multiple historical moments within the historical time period. Features are extracted from the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data. Features are optimized in the battery operation feature sequence, and a battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operation feature. The state of charge (SOC) of the target battery is predicted based on the battery optimization feature sequence to obtain the predicted SOC of the target battery in the target time period. Server 104 then sends the predicted SOC to terminal 102. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0065] Alternatively, sodium-ion batteries, due to their abundant resources, low cost, and good safety performance, are gradually becoming an important alternative to lithium-ion batteries, suitable for scenarios such as grid energy storage, electric mobility, and home energy management. However, sodium-ion batteries still face a series of challenges in practical use, including limited cycle life, fluctuations in rate performance, and the complexity of battery state estimation, especially the accurate assessment of the State of Charge (SOC). SOC, as a core parameter reflecting the remaining capacity of the battery, plays a decisive role in ensuring the safe operation of the battery system and extending battery life.

[0066] Compared to lithium-ion batteries, sodium-ion batteries exhibit more complex electrochemical behavior and significant polarization, making it difficult to establish an accurate mapping directly using simple mathematical models. Traditional SOC estimation methods, such as coulometric methods or table lookup methods based on open-circuit voltage, while simple to implement, are easily affected by integral errors, environmental disturbances, and other factors, making it difficult to meet the dual requirements of real-time performance and accuracy in practical engineering.

[0067] To improve estimation performance, researchers have proposed various physical model-based methods, such as the Extended Kalman Filter (EKF) or the Unscented Kalman Filter (UKF), which dynamically estimate the State of Charge (SOC) by combining it with the battery's equivalent circuit model. These methods can integrate model structure and measurement data to a certain extent, exhibiting good robustness and engineering applicability. However, these models rely on accurate battery parameter identification, and their filtering performance degrades significantly under high nonlinearity or drastic changes in operating conditions. Therefore, finding SOC estimation methods that do not rely on strong prior models and can flexibly adapt to diverse data characteristics has become a key direction for the development of intelligent battery management technology.

[0068] Based on this, in an exemplary embodiment, such as Figure 2 As shown, a method for estimating the state of charge (SOC) of sodium-ion batteries based on a CNN-ResNet-LSTM hybrid network is presented, and this method is applied to... Figure 1 Taking the server in the example of this, the explanation includes:

[0069] Step 202: Obtain the battery operation data sequence of the target battery to be estimated in the historical time period.

[0070] The target battery can be a sodium-ion battery; the historical time period can be the period during which battery operation data is collected, for example, the 30 seconds before the remaining power prediction; the battery operation data sequence includes battery operation data of the target battery at multiple historical moments within the historical time period; the battery operation data includes, but is not limited to, battery current and / or battery voltage, etc.

[0071] Optionally, voltage and current data of the target battery over a historical period can be collected through a battery management system (BMS) or sensors to obtain a battery operation data sequence.

[0072] Step 204: Extract features from the battery operation data in the battery operation data sequence. Based on the battery operation features extracted for each battery operation data, obtain the battery operation feature sequence.

[0073] Among them, battery operation characteristics can be time-series features obtained after feature extraction, which are characteristic expressions of battery operation data.

[0074] Optionally, pre-trained convolutional neural networks (CNNs) can be used to process battery operating data (such as battery current and battery voltage) to obtain features corresponding to battery current and battery voltage, i.e., to obtain battery operating features at each historical moment, and thus to obtain a battery operating feature sequence. Features can also be extracted using GRU (Gated Recurrent Unit) or Transformer encoders; this embodiment does not limit this approach.

[0075] Step 206: Perform feature optimization on the battery operation features in the battery operation feature sequence. Based on the battery optimization features extracted for each battery operation feature, obtain the battery optimization feature sequence.

[0076] Among them, the battery optimization features can be the higher-order time-series features obtained after optimization.

[0077] Optionally, the battery operating feature sequence obtained earlier can be adjusted to a format consistent with that of the training phase to ensure that the time sequence length and number of channels meet the input requirements of the pre-trained residual network. The two trained cascaded residual blocks are then used to optimize the battery operating feature sequence to obtain optimized battery features at multiple historical time points, i.e., the battery optimized feature sequence. Alternatively, a densely connected network (DenseNet) can be used to optimize the battery operating features; this embodiment does not impose any limitations on this approach.

[0078] Step 208: Based on the battery optimization feature sequence, predict the state of charge of the target battery to obtain the predicted state of charge of the target battery in the target time period.

[0079] The target time period can be a future time period in which the remaining power of the battery to be estimated needs to be estimated; the predicted state of charge can be the estimated remaining power of the battery at each time point within the target time period, and the predicted state of charge is obtained by a hybrid network of CNN (Convolutional Neural Networks)-ResNet (Residual Network)-LSTM (Long Short-Term Memory).

[0080] Optionally, a pre-trained time-dependent capture layer can be invoked, and the battery optimization features from multiple historical moments after adaptation can be input into this layer. Then, the hidden state output by the time-dependent capture layer can be input into a pre-trained fully connected layer. Through spatial mapping of the trained weight matrix and bias vector, the high-dimensional hidden state can be transformed into a one-dimensional normalized prediction value, which corresponds to the normalized SOC of the target time period. Finally, the same inverse normalization method as in the training phase (such as Min-Max) is used, combined with the maximum and minimum SOC (Ymin) corresponding to the training data, to restore the normalized prediction value to the actual physical meaning of the remaining battery power, thus obtaining the estimated remaining battery power of the battery at each moment in the target time period.

[0081] In the aforementioned sodium-ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network, the battery operation data sequence of the target battery to be estimated over a historical period is obtained, providing a data foundation for subsequent analysis. Then, features are extracted from the battery operation data to obtain a battery operation feature sequence, enabling efficient capture of local features related to the state of charge from the raw data. Subsequently, the battery operation feature sequence is optimized to obtain an optimized battery feature sequence, enhancing the robustness of the features. Finally, based on the optimized battery feature sequence, the state of charge is predicted, achieving accurate estimation of the remaining battery capacity.

[0082] In an exemplary embodiment, feature extraction is performed on battery operation data in a battery operation data sequence. Based on the battery operation features extracted for each battery operation data, a battery operation feature sequence is obtained. This includes: normalizing the battery operation data for each battery operation data in the battery operation data sequence, and inputting the normalized battery operation data into a convolutional neural network layer for feature extraction to obtain the battery operation features corresponding to the battery operation data; and obtaining a battery operation feature sequence based on the battery operation features corresponding to each battery operation data.

[0083] Optionally, each battery operation data point in the battery operation data sequence can be input into a pre-trained CNN layer. One-dimensional convolution operations with a kernel size of 3 are used to capture local correlation features at consecutive time points in the data (such as the continuous voltage decrease pattern during the discharge phase and the coupling relationship between current fluctuations and voltage changes). Simultaneously, the weight matrix and bias terms learned during the training phase are combined to output the convolution calculation results. Batch normalization is performed on the convolution calculation results, and then the normalized results are input into the ReLU activation function to perform a non-linear transformation on the feature data. Finally, the battery operation features at each historical time point are output, thus obtaining the battery operation feature sequence.

[0084] In this embodiment, the raw time-series data is transformed into a battery operation feature sequence through the synergistic effect of convolutional feature extraction, normalization stabilization, and nonlinear activation. This alleviates the deficiency of key information loss in traditional feature extraction and provides accurate input data for subsequent feature optimization and temporal dependency capture.

[0085] In an exemplary embodiment, feature optimization is performed on battery operating features in a battery operating feature sequence. Based on the battery optimization features extracted for each battery operating feature, a battery optimization feature sequence is obtained. This includes: for each battery operating feature in the battery operating feature sequence, feature extraction is performed on the battery operating feature input to obtain a first optimized feature of the battery operating feature; scaling is performed on the battery operating feature to obtain a second optimized feature of the battery operating feature, wherein the scale of the first optimized feature and the second optimized feature is consistent; the first optimized feature and the second optimized feature are fused to obtain the battery optimization feature corresponding to the battery operating feature; and a battery optimization feature sequence is obtained based on the battery optimization features corresponding to each battery operating feature.

[0086] The first optimized feature can be deep feature data obtained after processing the battery operation feature sequence; the second optimized feature can be feature data obtained after adjusting the number of channels in the battery operation feature sequence; the number of channels is used to characterize the number of battery feature data of different dimensions extracted by the residual network.

[0087] Optionally, the battery operating feature sequence is input into the main branch of the residual block to obtain the deep feature data obtained after processing the battery operating feature sequence, which is output by the main branch of the residual block; then, the battery operating feature sequence is input into the shortcut branch of the residual block to obtain the feature data obtained after adjusting the number of channels of the battery operating feature sequence, which is output by the shortcut branch of the residual block; the first optimized feature and the second optimized feature are added and fused to obtain the battery optimized feature corresponding to the battery operating feature. If the residual network contains multiple cascaded residual blocks (e.g., 2), the output of the current residual block is used as the input of the next residual block, and the above steps are repeated; finally, after processing by all residual blocks, the battery optimized feature sequence is output.

[0088] In this embodiment, the robustness of the features is improved by combining deep feature extraction and channel number adaptation with additive fusion and activation enhancement. The final output battery-optimized feature sequence provides high-quality input for subsequent time dependency capture and accurate battery prediction, thereby improving prediction stability.

[0089] In an exemplary embodiment, the state of charge (SOC) prediction of the battery to be estimated is performed based on the battery optimization feature sequence to obtain the predicted SOC of the battery to be estimated in a target time period. This includes: inputting the optimization feature sequence into a time dependency capture layer, capturing the time dependency relationship in the battery optimization feature sequence through the gating mechanism of the time dependency capture layer, and outputting the hidden state; inputting the hidden state into a fully connected layer, and performing spatial mapping of the hidden state through the weight matrix and bias vector of the fully connected layer to obtain the predicted SOC of the battery to be estimated in the target time period.

[0090] The time-dependent capture layer can be a network layer used to capture temporal feature dependencies, such as a pre-trained Long Short-Term Memory (LSTM) network; the time dependency can be the correlation between features at different times in the battery optimization feature sequence, such as the correlation between voltage features at the beginning of charging and power consumption in the middle of discharging; the hidden state can be the output of the time-dependent capture layer, which is a high-dimensional feature vector containing complete temporal dependency rules.

[0091] Optionally, the adapted feature sequence can be input into a pre-trained LSTM layer. This involves filtering features strongly correlated with remaining battery power through the input gate; filtering redundant noise through the forget gate; focusing on the temporal patterns of battery power correlation between the current time and the target time period through the output gate, updating the memory cell state and hidden state step-by-step; and outputting a hidden state containing complete long- and short-term temporal dependencies. The hidden state is then input into a pre-trained fully connected layer, where a linear transformation is performed using the learned weight matrix and bias vector to map the high-dimensional hidden state to a one-dimensional normalized predicted value, corresponding to the normalized SOC for the target time period. Finally, the normalized SOC is de-normalized to obtain the predicted state of charge (SOC) of the battery to be estimated within the target time period.

[0092] In this embodiment, the problem of insufficient long-term time-dependent capture in traditional methods is alleviated by using the gating mechanism of the time-dependent capture layer and the precise mapping with the fully connected layer. This not only fully exploits the temporal correlation patterns in the battery optimization feature sequence, but also improves the accuracy of prediction by outputting the predicted state of charge through standardized spatial mapping and inverse normalization.

[0093] In an exemplary embodiment, the training process of the sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network includes: obtaining the sample battery operation data sequence corresponding to the sample battery; obtaining the actual remaining charge sequence corresponding to the sample battery operation data sequence; inputting the sample battery operation data sequence into the sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network to be trained, and obtaining the predicted state of charge sequence output by the charge prediction model; training the sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network based on the difference between the actual remaining charge sequence and the predicted state of charge sequence, and obtaining the trained sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network.

[0094] The sample battery can be a sodium-ion battery used to conduct open-circuit voltage testing and provide training data; the sample battery operation data sequence is obtained during the open-circuit voltage testing of the sample battery; the predicted state of charge sequence is obtained through a sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network, which includes a convolutional neural network layer, a residual network, a time-dependent capture layer, and a fully connected layer; the actual remaining charge sequence is the actual remaining charge of the sample battery at each sampling time during the charging and discharging process.

[0095] Optionally, voltage and current sequences are obtained by testing the open-circuit voltage of sample batteries as training samples. The actual SOC is used as a label and input into the initial model for forward propagation to obtain the predicted charge. Then, the parameters are adjusted in reverse based on the difference between the prediction and the label, and finally, an optimized sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network is obtained.

[0096] In this embodiment, the sodium-ion battery SOC estimation model based on the trained CNN-ResNet-LSTM hybrid network can fully capture the local correlation, deep features and time-series dependence features of the battery operation data, reduce the dependence of traditional models on strong prior knowledge, and alleviate the feature extraction limitations of a single network structure, thereby achieving accurate prediction of the remaining power of the sodium-ion battery.

[0097] In an exemplary embodiment, obtaining the sample battery operation data sequence corresponding to the sample battery includes: performing an open-circuit voltage test on the sample battery, collecting the test voltage sequence and test current sequence of the sample battery during the charging and discharging process, and generating the sample battery operation data sequence corresponding to the sample battery based on the test voltage sequence and test current sequence.

[0098] The charging and discharging process includes a charging phase and a discharging phase. The discharging phase is initiated after the charging phase ends and the sample battery has been left to stand for a first period of time.

[0099] The charging phase includes: charging the sample battery for a first duration according to a first constant current, then resting for a second resting duration, and charging the sample battery according to a first constant voltage until the charging current is lower than the first cutoff current.

[0100] The discharge phase includes discharging the sample battery for a second duration according to a second constant current, and then allowing it to stand for a third duration until the terminal voltage of the sample battery drops to the cutoff voltage.

[0101] Wherein, the first resting time can be the resting time of the sample battery after the charging phase ends, used to eliminate the polarization effect generated during the charging process; the first constant current can be the constant current charging current during the charging phase; the second resting time can be the resting time after the constant current charging during the charging phase; the first constant voltage can be the target voltage of the constant voltage charging during the charging phase; the first duration can be the duration of the constant current charging during the charging phase; the second constant current can be the constant current discharging current during the discharging phase; the second duration can be the duration of the constant current discharging during the discharging phase; the third resting time can be the resting time after the constant current discharging during the discharging phase; and the cutoff voltage can be the termination voltage of the discharging phase.

[0102] Optionally, an open-circuit voltage (OCV) test is performed on the sodium-ion battery to obtain charge / discharge voltage and current data. The specific experimental steps are as follows: Under constant temperature conditions (e.g., 25°C), the battery is first charged with a first constant current (e.g., 0.5C constant current) for a first duration (e.g., 6 minutes), followed by a second resting duration (e.g., 30 minutes), until the voltage reaches a first constant voltage (e.g., 3.95V). Then, the battery is charged with the first constant voltage until the current is less than the first cutoff current (e.g., 0.05C). After fully charging, the battery is rested for a first resting duration (e.g., 30 minutes) and then begins to discharge. The battery is then discharged with a second constant current (e.g., 0.5C constant current) for a second duration (e.g., 6 minutes), followed by a first resting duration (e.g., 30 minutes), until the voltage drops to the cutoff voltage (e.g., 2V). This completes one round of charge and discharge. In this embodiment, three rounds of charge and discharge can be performed to obtain the test voltage sequence and test current sequence.

[0103] In this embodiment, the open-circuit voltage testing process alleviates the environmental interference problem in traditional testing, ensuring the completeness and consistency of the collected test voltage and current sequences. Data supplementation from multiple rounds of testing further enriches the sample diversity, ensuring that the training samples comprehensively cover the battery's operating state. This provides highly reliable training data for the sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network, effectively improving the model's ability to learn battery features and the accuracy of subsequent predictions.

[0104] In one embodiment, taking a CNN-ResNet-LSTM hybrid network as an example, the method of predicting the remaining battery capacity using a CNN-ResNet-LSTM hybrid network is illustrated, including: acquiring sodium-ion battery voltage and current test data; dividing the sodium-ion battery voltage and current test data into training and test sets, and performing data normalization preprocessing; constructing a CNN-ResNet-LSTM hybrid network, which consists of an input layer, a CNN (Convolutional Neural Networks) layer, a ResNet residual network, an LSTM (Long Short-Term Memory) layer, a fully connected layer, and a regression output layer from the input to the output; training the CNN-ResNet-LSTM hybrid network; using the trained hybrid network for SOC estimation and inverse normalization processing; and evaluating model performance and visualizing the results.

[0105] Optionally, the discharge voltage and discharge current data of the sodium-ion battery obtained in the experiment are divided into training and test sets. The first 90% of the data is assigned to the training set, and the last 10% is assigned to the test set. After dividing the data into training and test sets, the Min-Max method is used to perform linear normalization, mapping the feature values ​​to the interval [-1, 1], as shown in the following formula:

[0106]

[0107] Among them, X t This refers to raw data, such as battery operating data of the target battery at multiple historical moments within a historical time period, including discharge voltage and / or discharge current, all of which are column vectors; X min and X max These refer to the minimum and maximum values ​​of the original data, respectively. This refers to the data after normalization.

[0108] Optionally, the constructed hybrid network uses residual connections to optimize the feature extraction capability of the CNN and combines it with LSTM to capture temporal information, thereby improving the accuracy and robustness of SOC estimation. The overall network construction idea is as follows: data flows from the input layer to the CNN layer, then from the residual block output to the LSTM layer, then into the fully connected layer, and finally the regression layer outputs the final prediction result and calculates the loss. The network architecture is as follows: Figure 3 As shown. The specific construction process is as follows: The input layer data of the hybrid network consists of one-dimensional discharge current and discharge voltage time series data with 2 features. This data needs to be divided into training and testing sets and normalized. After the input layer, a one-dimensional CNN layer is constructed to initially extract the voltage and current data features:

[0109]

[0110] Where Z0 represents the extracted voltage or current data features, i.e., the battery operating characteristics at each historical moment; ReLU refers to the activation function, * indicates one-dimensional convolution calculation; W0 is the convolution kernel matrix with a kernel size of 3; x represents the normalized data of discharge voltage and discharge current, and b0 is the corresponding bias term; BN represents the batch normalization operation, as shown in the following formula:

[0111]

[0112] Where γ and β are learnable parameters that control the scale and offset of the normalized output, respectively; μ and σ 2 These are the mean and variance of a small batch of samples, and ε is a small constant to prevent division by zero.

[0113] Optionally, a ResNet residual network is constructed. The ResNet residual network consists of two cascaded residual blocks, each including a main branch and a shortcut branch. The output of each residual block is a fusion of the outputs of its main branch and shortcut branch. In the main branch, two one-dimensional convolutional layers are used sequentially to extract data features. Each convolutional layer performs batch normalization and activation function operations, with a kernel size of 3. The operation of the first convolutional layer is as follows:

[0114]

[0115] The operation of the second convolutional layer is as follows:

[0116]

[0117] Where Z1 and F(x) represent the outputs of the first and second convolutional layers of the main branch, respectively, Z0 represents the input of residual block 1 (i.e. the output of the previous CNN layer); W1 and W2 are the corresponding convolutional kernel matrices, and b1 and b2 are the corresponding bias terms.

[0118] Within shortcut branches, the processing method differs depending on the number of channels. Details are as follows:

[0119] When the number of input and output channels is the same, use the identity mapping:

[0120]

[0121] When the number of channels is different, use a 1×1 convolution to adjust the number of channels:

[0122]

[0123] Where x and Shortcut(x) represent the input and output of the shortcut branch, respectively, and W s It is a 1×1 convolution.

[0124] like Figure 3 As shown, the shortcut branch of residual block 1 receives input from the output of the preceding CNN layer (32 channels) and outputs 16 channels. The shortcut branch of residual block 2 receives input from the output of residual block 1 (16 channels) and outputs 32 channels. Therefore, the number of input and output channels for the shortcut branches of the two residual blocks are different. Residual fusion within each residual block involves fusing the main branch and the shortcut branch through addition, and then applying an activation function to obtain the output Y of each residual block, i.e., the battery optimization feature sequence.

[0125]

[0126] Optionally, after the output of residual block 2, it is connected to an LSTM layer to capture the temporal dependencies in the sequence. The core mechanism of the LSTM layer includes the input gate, forget gate, candidate memory gate, and output gate, which are computed in parallel. Let the input of the LSTM layer at time t be x. t The recursive formula is as follows:

[0127] Input Gate: ;

[0128] Forgotten Gate: ;

[0129] Candidate state: ;

[0130] Cell status update: ;

[0131] Output gate: ;

[0132] Hidden state: ;

[0133] Where σ refers to the sigmoid activation function, ⊙ refers to the Hadamard element-wise product, and bf b i b c and b o The bias term, W f W i W c and W o The weight matrix, C t The state of the memory unit at the current moment, h t-1 This refers to the hidden state at time t-1. The fully connected layer stores the last hidden state h of the LSTM layer. t Mapping to the output space, we can obtain the predicted SOC output of this hybrid network:

[0134]

[0135] Among them, W out Let b be the weight matrix of the fully connected layer. out This is the bias vector.

[0136] Since the estimated SOC above is a normalized result, it needs to be denormalized:

[0137]

[0138] Among them, Y t The final SOC estimate is the predicted state of charge; the predicted state of charge, Y max Y min These are the maximum and minimum SOC corresponding to the training data, respectively.

[0139] Optionally, the hybrid network constructed in this embodiment also includes a regression output layer connected after the fully connected layers. This layer outputs the final SOC estimate and is responsible for calculating the mean squared error loss between the SOC estimate and the true value, thus guiding the training process of the hybrid network. When training the hybrid network, the Adam optimizer can be used to optimize the weights of all layers in the network architecture. For example, the maximum number of training epochs is set to 500, with 64 samples used per training session and an initial learning rate of 0.0005. The Adam optimizer optimizes the network parameters based on the mean squared error (MSE) of the regression loss function.

[0140] Optionally, this embodiment can use RMSE to evaluate the performance of the hybrid network, and plot comparison charts of the SOC estimation results between the real SOC and the training and test sets. Figure 4 and Figure 5 , and, such as Figure 6 The image shows the SOC estimation error plots for the training and test sets. Figure 4The results of the CNN-ResNet-LSTM hybrid network on the training set show that the model has high fitting accuracy to the training set data, and the estimated SOC value almost coincides with the true value, demonstrating good learning and fitting ability and accurately capturing the SOC change pattern. Figure 5 The SOC estimation results of this hybrid network on the test set show that it can follow the key changes of the true value as a whole, and the trend fit in the low SOC range is improved, indicating that the model has a certain generalization ability on the test set data. Although there are error fluctuations, the overall prediction effect is good, and it has the potential to be applied in actual battery SOC estimation scenarios and provide technical support for battery management systems. Figure 6 The graph compares the errors in SOC estimation between the training and test sets. The error on the training set fluctuates slightly around the 0 axis, indicating that the model has high prediction accuracy and stable error on the training set. The error on the test set fluctuates slightly more than that on the training set, but it follows a regular pattern, providing direction for subsequent model optimization.

[0141] To more comprehensively demonstrate this solution, this embodiment presents a method for estimating the state of charge (SOC) of sodium-ion batteries based on a CNN-ResNet-LSTM hybrid network, specifically including:

[0142] 1. Obtain the battery operation data sequence of the target battery to be estimated over a historical time period;

[0143] 2. For each battery operation data in the battery operation data sequence, the battery operation data is normalized, and the normalized battery operation data is input into the convolutional neural network layer for feature extraction to obtain the battery operation features corresponding to the battery operation data.

[0144] 3. Based on the battery operating characteristics corresponding to each battery's operating data, obtain the battery operating characteristic sequence;

[0145] 4. For each battery operating feature in the battery operating feature sequence, perform feature extraction on the battery operating feature input to obtain the first optimized feature of the battery operating feature;

[0146] 5. Perform scaling transformation on the battery operating characteristics to obtain the second optimized features of the battery operating characteristics. The scale of the first optimized features and the second optimized features are consistent.

[0147] 6. The first optimization feature and the second optimization feature are fused together to obtain the battery optimization feature corresponding to the battery operation feature;

[0148] 7. Based on the battery optimization features corresponding to each battery's operating characteristics, obtain the battery optimization feature sequence;

[0149] 8. Based on the battery optimization feature sequence, predict the state of charge of the target battery to obtain the predicted state of charge of the target battery in the target time period.

[0150] The specific process of the above steps can be found in the description of the above method embodiments. The implementation principle and technical effect are similar, and will not be repeated here.

[0151] It should be understood that although the steps in the flowcharts of the embodiments described above 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 embodiments described above 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.

[0152] Based on the same inventive concept, this application also provides a sodium-ion battery SOC estimation device based on a CNN-ResNet-LSTM hybrid network for implementing the aforementioned sodium-ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network. The solution provided by this device is similar to the implementation described in the above method. Therefore, the specific limitations of one or more embodiments of the sodium-ion battery SOC estimation device based on a CNN-ResNet-LSTM hybrid network provided below can be found in the limitations of the sodium-ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network described above, and will not be repeated here.

[0153] In one exemplary embodiment, such as Figure 7 As shown, a sodium-ion battery SOC estimation device based on a CNN-ResNet-LSTM hybrid network is provided, including: an acquisition module 71, an extraction module 72, an optimization module 73, and a prediction module 74, wherein:

[0154] The acquisition module 71 is used to acquire the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0155] Extraction module 72 is used to extract features from battery operation data in battery operation data sequence, and obtain battery operation feature sequence based on the battery operation features extracted for each battery operation data.

[0156] Optimization module 73 is used to optimize the battery operating features in the battery operating feature sequence, and obtain the battery optimized feature sequence based on the battery optimization features extracted for each battery operating feature.

[0157] The prediction module 74 is used to predict the state of charge of the target battery based on the battery optimization feature sequence, so as to obtain the predicted state of charge of the target battery in the target time period; wherein, the predicted state of charge is obtained by the CNN-ResNet-LSTM hybrid network.

[0158] In one embodiment, the extraction module 72 is further configured to:

[0159] For each battery operation data point in the battery operation data sequence, the battery operation data is normalized, and the normalized battery operation data is input into the convolutional neural network layer for feature extraction to obtain the battery operation features corresponding to the battery operation data.

[0160] Based on the battery operating characteristics corresponding to each battery operating data, a battery operating characteristic sequence is obtained.

[0161] In one embodiment, the optimization module 73 is further configured to:

[0162] For each battery operating feature in the battery operating feature sequence, feature extraction is performed on the battery operating feature input to obtain the first optimized feature of the battery operating feature;

[0163] The battery operating characteristics are scaled to obtain the second optimized features of the battery operating characteristics. The scale of the first optimized features and the second optimized features are consistent.

[0164] The first optimization feature and the second optimization feature are fused to obtain the battery optimization feature corresponding to the battery operation feature;

[0165] Based on the battery optimization features corresponding to each battery's operating characteristics, a battery optimization feature sequence is obtained.

[0166] In one embodiment, the acquisition module 71 is further configured to:

[0167] Obtain the sample battery operation data sequence corresponding to the sample battery. The sample battery operation data sequence is obtained during the open circuit voltage test of the sample battery.

[0168] Obtain the actual remaining battery power sequence corresponding to the sample battery operation data sequence;

[0169] The sample battery running data sequence is input into the sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network to be trained, and the predicted state of charge sequence output by the power prediction model is obtained.

[0170] Based on the difference between the actual remaining charge sequence and the predicted state of charge sequence, a sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network is trained to obtain the trained sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network.

[0171] In one embodiment, the acquisition module 71 is further configured to:

[0172] An open-circuit voltage test is performed on the sample battery, and the test voltage sequence and test current sequence of the sample battery during the charging and discharging process are collected. Based on the test voltage sequence and test current sequence, the corresponding sample battery operation data sequence is generated.

[0173] The modules in the sodium-ion battery SOC estimation device based on the CNN-ResNet-LSTM hybrid network described above can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independent of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0174] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. 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 and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a sodium-ion battery SOC estimation method based on a CNN-ResNet-LSTM hybrid network. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0175] Those skilled in the art will understand that Figure 7 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.

[0176] 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:

[0177] Obtain the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0178] Feature extraction is performed on the battery operation data in the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data.

[0179] The battery operation features in the battery operation feature sequence are optimized, and the battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operation feature.

[0180] The target battery's state of charge (SOC) is predicted based on the battery's optimized feature sequence, resulting in a predicted SOC for the target battery within a target time period. The predicted SOC is obtained through a CNN-ResNet-LSTM hybrid network.

[0181] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0182] For each battery operation data point in the battery operation data sequence, the battery operation data is normalized, and the normalized battery operation data is input into the convolutional neural network layer for feature extraction to obtain the battery operation features corresponding to the battery operation data.

[0183] Based on the battery operating characteristics corresponding to each battery operating data, a battery operating characteristic sequence is obtained.

[0184] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0185] For each battery operating feature in the battery operating feature sequence, feature extraction is performed on the battery operating feature input to obtain the first optimized feature of the battery operating feature;

[0186] The battery operating characteristics are scaled to obtain the second optimized features of the battery operating characteristics. The scale of the first optimized features and the second optimized features are consistent.

[0187] The first optimization feature and the second optimization feature are fused to obtain the battery optimization feature corresponding to the battery operation feature;

[0188] Based on the battery optimization features corresponding to each battery's operating characteristics, a battery optimization feature sequence is obtained.

[0189] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0190] Obtain the sample battery operation data sequence corresponding to the sample battery. The sample battery operation data sequence is obtained during the open circuit voltage test of the sample battery.

[0191] Obtain the actual remaining battery power sequence corresponding to the sample battery operation data sequence;

[0192] The sample battery running data sequence is input into the state of charge prediction model to be trained to obtain the predicted state of charge sequence output by the power prediction model.

[0193] Based on the difference between the actual remaining charge sequence and the predicted state of charge sequence, a state of charge prediction model is trained to obtain the trained state of charge prediction model.

[0194] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0195] The sample battery is subjected to open-circuit voltage test, and the test voltage sequence and test current sequence of the sample battery during the charging and discharging process are collected. Based on the test voltage sequence and test current sequence, the sample battery operation data sequence corresponding to the sample battery is generated.

[0196] The charging and discharging process includes a charging phase and a discharging phase. The discharging phase is initiated after the charging phase ends and the sample battery has been left to stand for a certain period of time.

[0197] 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:

[0198] Obtain the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0199] Feature extraction is performed on the battery operation data in the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data.

[0200] The battery operation features in the battery operation feature sequence are optimized, and the battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operation feature.

[0201] The target battery's state of charge (SOC) is predicted based on the battery's optimized feature sequence, resulting in a predicted SOC for the target battery within a target time period. The predicted SOC is obtained through a CNN-ResNet-LSTM hybrid network.

[0202] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0203] For each battery operation data point in the battery operation data sequence, the battery operation data is normalized, and the normalized battery operation data is input into the convolutional neural network layer for feature extraction to obtain the battery operation features corresponding to the battery operation data.

[0204] Based on the battery operating characteristics corresponding to each battery operating data, a battery operating characteristic sequence is obtained.

[0205] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0206] For each battery operating feature in the battery operating feature sequence, feature extraction is performed on the battery operating feature input to obtain the first optimized feature of the battery operating feature;

[0207] The battery operating characteristics are scaled to obtain the second optimized features of the battery operating characteristics. The scale of the first optimized features and the second optimized features are consistent.

[0208] The first optimization feature and the second optimization feature are fused to obtain the battery optimization feature corresponding to the battery operation feature;

[0209] Based on the battery optimization features corresponding to each battery's operating characteristics, a battery optimization feature sequence is obtained.

[0210] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0211] Obtain the sample battery operation data sequence corresponding to the sample battery. The sample battery operation data sequence is obtained during the open circuit voltage test of the sample battery.

[0212] Obtain the actual remaining battery power sequence corresponding to the sample battery operation data sequence;

[0213] The sample battery running data sequence is input into the state of charge prediction model to be trained to obtain the predicted state of charge sequence output by the power prediction model.

[0214] Based on the difference between the actual remaining charge sequence and the predicted state of charge sequence, a state of charge prediction model is trained to obtain the trained state of charge prediction model.

[0215] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0216] The sample battery is subjected to open-circuit voltage test, and the test voltage sequence and test current sequence of the sample battery during the charging and discharging process are collected. Based on the test voltage sequence and test current sequence, the sample battery operation data sequence corresponding to the sample battery is generated.

[0217] The charging and discharging process includes a charging phase and a discharging phase. The discharging phase is initiated after the charging phase ends and the sample battery has been left to stand for a certain period of time.

[0218] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0219] Obtain the battery operation data sequence of the target battery to be estimated in the historical time period. The battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period.

[0220] Feature extraction is performed on the battery operation data in the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data.

[0221] The battery operation features in the battery operation feature sequence are optimized, and the battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operation feature.

[0222] Based on the battery optimization feature sequence, the state of charge of the target battery is predicted to obtain the predicted state of charge of the target battery in the target time period.

[0223] The predicted state of charge is obtained through a CNN-ResNet-LSTM hybrid network.

[0224] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0225] For each battery operation data point in the battery operation data sequence, the battery operation data is normalized, and the normalized battery operation data is input into the convolutional neural network layer for feature extraction to obtain the battery operation features corresponding to the battery operation data.

[0226] Based on the battery operating characteristics corresponding to each battery operating data, a battery operating characteristic sequence is obtained.

[0227] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0228] For each battery operating feature in the battery operating feature sequence, feature extraction is performed on the battery operating feature input to obtain the first optimized feature of the battery operating feature;

[0229] The battery operating characteristics are scaled to obtain the second optimized features of the battery operating characteristics. The scale of the first optimized features and the second optimized features are consistent.

[0230] The first optimization feature and the second optimization feature are fused to obtain the battery optimization feature corresponding to the battery operation feature;

[0231] Based on the battery optimization features corresponding to each battery's operating characteristics, a battery optimization feature sequence is obtained.

[0232] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0233] Obtain the sample battery operation data sequence corresponding to the sample battery. The sample battery operation data sequence is obtained during the open circuit voltage test of the sample battery.

[0234] Obtain the actual remaining battery power sequence corresponding to the sample battery operation data sequence;

[0235] The sample battery running data sequence is input into the state of charge prediction model to be trained to obtain the predicted state of charge sequence output by the power prediction model.

[0236] Based on the difference between the actual remaining charge sequence and the predicted state of charge sequence, a state of charge prediction model is trained to obtain the trained state of charge prediction model.

[0237] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0238] The sample battery is subjected to open-circuit voltage test, and the test voltage sequence and test current sequence of the sample battery during the charging and discharging process are collected. Based on the test voltage sequence and test current sequence, the sample battery operation data sequence corresponding to the sample battery is generated.

[0239] The charging and discharging process includes a charging phase and a discharging phase. The discharging phase is initiated after the charging phase ends and the sample battery has been left to stand for a certain period of time.

[0240] 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.

[0241] 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.

[0242] 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 patent 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 method for estimating the state of charge (SOC) of sodium-ion batteries based on a CNN-ResNet-LSTM hybrid network, characterized in that, The method includes: Obtain the battery operation data sequence of the target battery to be estimated in a historical time period, the battery operation data sequence including the battery operation data of the target battery at multiple historical moments in the historical time period; Feature extraction is performed on the battery operation data in the battery operation data sequence, and a battery operation feature sequence is obtained based on the battery operation features extracted for each battery operation data. The battery operating features in the battery operating feature sequence are optimized, and a battery optimization feature sequence is obtained based on the battery optimization features extracted for each battery operating feature. Based on the battery optimization feature sequence, the state of charge of the target battery is predicted to obtain the predicted state of charge of the target battery in the target time period. The predicted state of charge is obtained through a CNN-ResNet-LSTM hybrid network.

2. The method according to claim 1, characterized in that, The step of extracting features from the battery operation data sequence, and obtaining a battery operation feature sequence based on the battery operation features extracted for each battery operation data, includes: For each battery operation data in the battery operation data sequence, the battery operation data is normalized, and the normalized battery operation data is input into a convolutional neural network layer for feature extraction to obtain the battery operation features corresponding to the battery operation data. Based on the battery operating characteristics corresponding to each of the battery operating data, a battery operating characteristic sequence is obtained.

3. The method according to claim 1, characterized in that, The step of optimizing the battery operating features in the battery operating feature sequence, based on the battery optimization features extracted for each battery operating feature, to obtain a battery optimization feature sequence includes: For each battery operating feature in the battery operating feature sequence, feature extraction is performed on the battery operating feature input to obtain the first optimized feature of the battery operating feature; The battery operating characteristics are scaled to obtain a second optimized feature of the battery operating characteristics, wherein the scale of the first optimized feature and the second optimized feature are consistent. The first optimization feature and the second optimization feature are fused together to obtain the battery optimization feature corresponding to the battery operation feature; Based on the battery optimization features corresponding to each of the battery operating features, a battery optimization feature sequence is obtained.

4. The method according to any one of claims 1 to 3, characterized in that, The battery operation data of the target battery at a historical time includes the voltage data and / or current data of the target battery at that historical time.

5. The method according to any one of claims 1 to 3, characterized in that, The predicted state of charge (SOC) is obtained through a trained sodium-ion battery SOC estimation model based on a CNN-ResNet-LSTM hybrid network. The training process of the sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network includes: Obtain the sample battery operation data sequence corresponding to the sample battery, which is obtained during the open circuit voltage test of the sample battery; Obtain the actual remaining power sequence corresponding to the sample battery operating data sequence; The sample battery running data sequence is input into the sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network to be trained, and the predicted state of charge sequence output by the power prediction model is obtained. Based on the difference between the actual remaining charge sequence and the predicted state of charge sequence, the sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network is trained to obtain the trained sodium-ion battery SOC estimation model based on the CNN-ResNet-LSTM hybrid network.

6. The method according to claim 5, characterized in that, The acquisition of the sample battery operation data sequence corresponding to the sample battery includes: An open-circuit voltage test is performed on the sample battery, and the test voltage sequence and test current sequence of the sample battery during the charging and discharging process are collected. Based on the test voltage sequence and the test current sequence, a sample battery operation data sequence corresponding to the sample battery is generated. The charging and discharging process includes a charging phase and a discharging phase. The discharging phase is initiated after the charging phase ends and the sample battery has been left to stand for a first settling time.

7. A sodium-ion battery SOC estimation device based on a CNN-ResNet-LSTM hybrid network, characterized in that, The device includes: The acquisition module is used to acquire the battery operation data sequence of the target battery to be estimated in a historical time period, wherein the battery operation data sequence includes the battery operation data of the target battery at multiple historical moments in the historical time period; The extraction module is used to extract features from the battery operation data in the battery operation data sequence, and obtain a battery operation feature sequence based on the battery operation features extracted for each battery operation data. The optimization module is used to perform feature optimization on the battery operating features in the battery operating feature sequence, and obtain a battery optimization feature sequence based on the battery optimization features extracted for each battery operating feature. The prediction module is used to predict the state of charge of the target battery based on the battery optimization feature sequence, so as to obtain the predicted state of charge of the target battery in the target time period; wherein the predicted state of charge is obtained by predicting through a CNN-ResNet-LSTM hybrid network.

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