A method, system, device and medium for online sodium precipitation detection of a sodium-ion battery

Abstract

The application discloses a kind of sodium ion battery online sodium precipitation detection method, system, equipment and medium, it is related to sodium ion battery energy storage technology field, comprising the following steps: real-time acquisition is detected sodium ion battery in charging process in original charging curve data;Original charging curve data is input to pre-trained auto-encoder and is handled to obtain reconstructed charging curve data, and reconstructed charging curve data is the charging curve data corresponding to normal aging state of sodium ion battery to be detected;Original charging curve data and reconstructed charging curve data are subtracted, and residual signal containing abnormal information is obtained;Residual signal is extracted and mapped to feature, and sodium precipitation probability is obtained;According to sodium precipitation probability, whether sodium precipitation anomaly occurs in real-time detection sodium ion battery to be detected.The application can capture abnormal sodium precipitation signal in real time during charging process, effectively overcome the problem of poor real-time performance and slow response speed in traditional detection method.

Description

Technical Field

[0001] This invention relates to the field of sodium-ion battery energy storage technology, and in particular to an online sodium deposition detection method, system, equipment, and medium for sodium-ion batteries. Background Technology

[0002] As a key development direction in the field of energy storage technology, sodium-ion batteries have become a focus of research in large-scale energy storage systems in recent years due to their unique cost advantages and resource adaptability. Compared with lithium-ion batteries, this technology not only has the advantages of abundant sodium resources and the use of aluminum foil as current collectors, but its electrochemical characteristics are also more adaptable to a wide temperature range operating environment. However, in the process of industrialization, sodium deposition has become a key technical bottleneck restricting its safety performance and cycle stability. Specifically, when the battery is under high-current charging or low-temperature conditions, the transport kinetics of sodium ions in the hard carbon anode material deteriorates significantly, causing sodium ions to be unable to embed in the complex sodium storage sites composed of disordered carbon layers and nanopores in time, resulting in heterogeneous deposition of metallic sodium on the anode surface. Experimental studies have shown that this irreversible dendrite growth phenomenon increases the rate of decline in battery cycle efficiency by 3-5 times, while also causing the interface SEI (Solid Electrolyte Interface) film to thicken continuously. More seriously, the penetration of local sodium dendrites can form micro-short-circuit channels, reducing the thermal runaway trigger temperature to below 80°C, which is about 40% lower than under normal operating conditions. Therefore, establishing an effective real-time sodium precipitation detection mechanism has become a core issue in overcoming the obstacles to the industrialization of sodium electrolysis technology.

[0003] For the detection of metal deposition in batteries, a series of in-situ and ex-situ techniques have been proposed. These techniques can be mainly divided into three categories: The first category is physical morphological characterization methods, such as optical microscopy, scanning electron microscopy, and transmission electron microscopy, which analyze the deposition process by observing the distribution and particle size of the deposits; the second category is chemical composition analysis methods, including X-ray photoelectron spectroscopy and Fourier transform infrared spectroscopy, which are used to study the chemical properties of the deposits and their interfacial environment; the third category is electrochemical measurement methods, which obtain reaction kinetic information by analyzing the interaction between the electrode and electrolyte interface. These methods include relaxation voltage analysis, incremental capacity analysis, and electrochemical impedance spectroscopy. While physical morphology characterization and chemical composition analysis methods can effectively characterize sediments in situ, they are not inherently suitable for continuous, online, and real-time detection. Advanced electrochemical technologies can provide valuable information on reaction kinetics, but they typically require high-precision specialized measurement equipment for online data acquisition. In particular, in sodium-ion batteries, the charge storage mechanism of the hard carbon anode relies on the adsorption and intercalation of sodium ions, rather than the stable intercalation of graphite. This makes the sodium precipitation signal weaker and harder to distinguish from other signals, easily masked by other factors, and difficult to detect in real time. Summary of the Invention

[0004] In view of the defects of the existing technology, the present invention provides a method, system, equipment and medium for online sodium precipitation detection in sodium-ion batteries, which solves the existing problems.

[0005] The present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides an online sodium deposition detection method for sodium-ion batteries, comprising the following steps:

[0007] Real-time acquisition of raw charging curve data of the sodium-ion battery under test during the charging process;

[0008] The original charging curve data is input into a pre-trained autoencoder to obtain reconstructed charging curve data, which is the charging curve data of the sodium-ion battery under test in normal aging state.

[0009] The original charging curve data and the reconstructed charging curve data are subtracted to obtain a residual signal containing abnormal information; the residual signal is subjected to feature extraction and feature mapping to obtain the sodium deposition probability; and the sodium deposition probability is used to detect in real time whether the sodium-ion battery under test has a sodium deposition abnormality.

[0010] Preferably, the autoencoder includes an encoder and a decoder. The encoder is used to extract aging features from the original charging curve data, and the decoder is used to reconstruct the aging features to obtain the reconstructed charging curve data.

[0011] Preferably, the residual signal is as follows:

[0012] ;

[0013] In the formula, For residual signals, The original charging curve data, To reconstruct the charging curve data.

[0014] Preferably, the sodium analysis probability is obtained through a pre-trained one-dimensional convolutional neural network, wherein the one-dimensional convolutional neural network includes an input layer, a first convolutional block, a second convolutional block, and a fully connected layer;

[0015] The input layer reconstructs the shape of the residual signal to obtain a two-dimensional tensor with shape [128,1].

[0016] The first convolutional block extracts features from the two-dimensional tensor, resulting in a first feature map with shape [64,4].

[0017] The second convolutional block extracts features from the first feature map, resulting in a second feature map with shape [32, 8].

[0018] The fully connected layer flattens and maps the second feature map to obtain the sodium precipitation probability.

[0019] Preferably, if the sodium deposition probability is greater than a set threshold, the sodium-ion battery under test is determined to have an abnormal sodium deposition.

[0020] Preferably, the original charging curve data covers a preset charging voltage range for the sodium-ion battery under test, including voltage data and charging capacity data at multiple time points.

[0021] Preferably, before inputting the original charging curve data into the pre-trained autoencoder for aging feature extraction, the original charging curve data is preprocessed by data filtering, noise reduction, interpolation, and normalization.

[0022] Secondly, the present invention provides an online sodium deposition detection system for sodium-ion batteries, comprising:

[0023] The acquisition module is used to acquire the raw charging curve data of the sodium-ion battery under test in real time during the charging process;

[0024] The reconstruction module is used to input the original charging curve data into a pre-trained autoencoder to process and obtain reconstructed charging curve data, wherein the reconstructed charging curve data is the charging curve data corresponding to the sodium-ion battery under normal aging state.

[0025] The detection module is used to subtract the original charging curve data and the reconstructed charging curve data to obtain a residual signal containing abnormal information; to perform feature extraction and feature mapping on the residual signal to obtain the sodium deposition probability; and to detect in real time whether the sodium-ion battery under test has experienced sodium deposition abnormality based on the sodium deposition probability.

[0026] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described online sodium deposition detection method for sodium-ion batteries.

[0027] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described online sodium deposition detection method for sodium-ion batteries.

[0028] Compared with the prior art, the above-mentioned at least one technical solution adopted by the present invention can achieve the following beneficial effects:

[0029] This invention first acquires the raw charging curve data of the sodium-ion battery under test in real time during the charging process. This raw charging curve data is then input into an autoencoder that reconstructs normal aging state data, accurately outputting the normal charging curve data under the corresponding aging state, laying the foundation for subsequent residual signal extraction. The difference between the raw and reconstructed charging curve data is calculated to obtain a residual signal containing abnormal information. This effectively separates the sodium deposition anomaly signal from the normal aging signal during sodium-ion battery charging, thereby improving the real-time performance and accuracy of the overall online sodium deposition detection. Finally, the residual signal is input into a pre-trained one-dimensional convolutional neural network for feature extraction and mapping, and the sodium deposition probability is used to detect in real time whether the sodium-ion battery under test has experienced sodium deposition anomalies. This invention can capture abnormal sodium deposition signals during charging in real time, effectively overcoming the problems of poor real-time performance and slow response speed in traditional detection methods, preventing safety accidents caused by sodium deposition, and providing a reliable decision-making basis for subsequent intelligent battery management. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the online sodium deposition detection method for sodium-ion batteries according to the present invention;

[0032] Figure 2 This is a schematic diagram of the structure of the self-encoder of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Example 1

[0035] To address the shortcomings of existing technologies, this invention discloses an online sodium deposition detection method for sodium-ion batteries, such as... Figure 1 As shown, the specific steps include the following:

[0036] S1: Data Acquisition and Preprocessing: Real-time acquisition of raw charging curve data of the sodium-ion battery under test during the charging process, followed by data smoothing and standardization.

[0037] For the actual operating conditions of online testing of sodium-ion batteries, the collected data mainly covers a preset charging voltage range, such as 3.2V to 3.98V. Within this voltage range, an appropriate sampling frequency is set, for example, collecting voltage and charging capacity data at several times per second during charging, or ensuring at least 128 time points per charging cycle, to fully reflect the subtle changes and dynamic characteristics of the charging curve. The horizontal axis of the charging curve represents voltage, and the vertical axis represents charging capacity.

[0038] Battery charging curve data is collected within a preset charging voltage range, including voltage data and corresponding charging capacity data. One charging curve is collected for each charging process. The collected data undergoes smoothing, filtering, interpolation, and normalization. Specifically, firstly, digital filtering (using low-pass filtering or wavelet transform) and noise reduction are performed to ensure continuous and smooth data. Then, cubic B-spline interpolation is used to reconstruct the data, achieving a uniform distribution of data within the 3.2V to 3.98V range. Afterward, the charging capacity data is normalized to eliminate dimensional differences between different test conditions and samples, obtaining preprocessed original charging curve data, providing accurate input for subsequent feature extraction.

[0039] Furthermore, in step S1, this embodiment first conducts a long-term cycle aging test on the sodium-ion battery. Specifically, four commercial sodium-ion batteries, model NaCY26700-2850mAh, with a nominal voltage of 3.0V and an operating voltage range of 2V to 4V, are selected. Charge-discharge tests are performed using an Arbin LBT testing system in a 25℃ temperature-controlled chamber. The charge-discharge current rate is set to 0.5C, and a 0.2C constant current discharge is used as a diagnostic step every fourth cycle. Parameters such as charge-discharge voltage, current, and temperature are recorded. To verify the effectiveness of the proposed method, four batteries were selected and labeled SIB (Sodium-ion Battery) #1, SIB#2, SIB#3, and SIB#4. During the experiment, SIB#1 served as the baseline sample, and no sodium deposition abnormalities were detected throughout the cycle test; however, SIB#2, SIB#3, and SIB#4 exhibited sodium deposition in subsequent cycle stages.

[0040] Charging curve data was collected within a preset charging voltage range (3.2V to 3.98V), ensuring that voltage and charging capacity data were collected at least 128 time points per charging cycle. The collected raw charging curve data was preprocessed.

[0041] Data filtering and noise reduction: After acquiring the original charging curve data, the data is first digitally filtered using methods such as low-pass filters or wavelet transforms to remove high-frequency noise and electromagnetic interference, and eliminate instantaneous outliers during the acquisition process. The original charging data is then smoothed to ensure good continuity and smoothness in subsequent processing.

[0042] Data Smoothing and Interpolation: To obtain more continuous and refined charging curve data, this invention employs cubic B-spline interpolation to reconstruct the collected original charging curve data, compensating for discontinuities caused by sampling intervals. The interpolation formula is as follows:

[0043] (1);

[0044] in, B k ( v ) represents the B-spline basis function. w k The interpolation weights are determined by the least squares method. v The voltage variable is used. This method allows the charging curve data to be evenly distributed within a preset voltage window, further improving the smoothness and accuracy of the data.

[0045] Data normalization: To eliminate the dimensional differences in data collected under different batteries or test conditions, this invention normalizes the processed charging capacity data, mapping the data to a fixed interval. The normalization formula is as follows:

[0046] (2);

[0047] in, q The charging capacity value collected at a certain moment. and These represent the minimum and maximum values ​​during the charging process, respectively. Normalization ensures that subsequent feature extraction and comparison of different battery data are performed at the same scale.

[0048] S2: Deep Feature Decomposition: The preprocessed raw charging curve data is modeled for aging features through an autoencoder neural network, its inherent aging features (aging components) are extracted, and the data is reconstructed based on the aging features to obtain the reconstructed charging curve data.

[0049] Autoencoder Network Structure Design: This embodiment employs a multi-layer fully connected autoencoder, structurally divided into an encoder and a decoder. The encoder maps the preprocessed raw charging curve data to a low-dimensional latent space. Its typical structure can be designed as follows: 128-dimensional input layer, 64-dimensional first hidden layer, 32-dimensional second hidden layer, 16-dimensional third hidden layer, and 8-dimensional encoder output layer. The decoder adopts a symmetrical structure with the encoder, progressively decoding and inversely mapping the 8-dimensional latent variables to the same dimensions as the original input, thus obtaining the reconstructed charging curve data. This structure aims to capture the main features implicit in the charging process during normal battery aging, enabling the network to learn the feature representation of the battery's health state.

[0050] Training Strategy and Objective Function: In training the autoencoder, this embodiment uses charging data with and without sodium deposition as input. When the input to the autoencoder is normal charging data without sodium deposition, its output is itself; when the input is charging data with abnormal sodium deposition, its output is normal charging data with the same capacity without abnormal sodium deposition. This training strategy ensures that the network only "memorizes" the dynamic process of normal battery aging. The training objective is the mean squared error, which is used as the loss function. The calculation formula is as follows:

[0051] (3);

[0052] in, x i This indicates the first [item] on the original charging curve data. i Data points, The corresponding reconstructed charging curve data points output by the network. N This represents the total number of data points. By minimizing this objective function, the network weights are optimized, enabling the aging behavior in the normal charging curve data to be efficiently encoded into low-dimensional latent variables.

[0053] By training on a large amount of normal and abnormal data, the autoencoder has formed a stable database of normal charging patterns. The trained autoencoder model can accurately reconstruct the corresponding charging curve data when normal charging data is input; and when abnormal charging data is input, it can accurately output the normal charging curve data under the corresponding aging state, laying the foundation for the extraction of residual signals.

[0054] In step S2, the data obtained in S1 is input into the autoencoder model for deep feature decomposition, such as... Figure 2As shown, the autoencoder employs a multi-layer fully connected network structure. The encoder layer progressively reduces the 128-dimensional data to 64, 32, 16, and finally an 8-dimensional bottleneck layer. The decoder layer uses a symmetrical structure to reconstruct the 8-dimensional latent variables into 128-dimensional data, resulting in reconstructed charging curve data representing the normal aging state. During training, mean squared error is used as the loss function, and normal data without sodium deposition is used as the primary training sample to ensure the model learns only normal aging features. For input charging curve data containing sodium deposition anomalies, since the autoencoder can only reconstruct the normal portion, the difference between the original charging curve data and the reconstructed charging curve data reveals the anomaly. Specifically, the residual signal mainly reflects the abnormal deviation caused by sodium deposition, serving as an effective input for the next step of online detection.

[0055] S3: Extract the residual signal by subtracting the preprocessed original charging curve data from the reconstructed charging curve data.

[0056] As a further technical solution, based on the modeling of normal aging characteristics, this invention further proposes a residual signal extraction scheme to isolate abnormal signals caused by sodium precipitation. The specific technical solution is as follows:

[0057] Residual signal calculation: Using a trained autoencoder model, preprocess the charging curve data for each input. x The reconstructed charging curve data is obtained through network reconstruction. . residual signal The calculation formula is:

[0058] (4);

[0059] Because autoencoders can only accurately reconstruct the normal aging portion, when sodium precipitation anomalies exist in the input data... This part of the information cannot be reproduced, therefore The value mainly reflects the abnormal component (sodium content), becoming an important basis for subsequent anomaly detection.

[0060] The residual signal serves as the input for anomaly detection: the value range and dynamic variation characteristics of the residual signal can effectively reflect sodium precipitation anomaly information. This signal will be used as the input to a subsequent one-dimensional convolutional neural network, which will further extract features and perform pattern recognition on the input signal. Theoretically, the numerical differences and temporal variations contained in the residual signal are caused by sodium precipitation behavior, thus providing the model with sufficient information to determine the fault state.

[0061] It should be noted that by using an autoencoder to perform aging modeling on sodium-ion battery charging data and extracting residual signals that characterize abnormal sodium deposition, the changes in charging curve data caused by sodium deposition can be amplified, making the abnormal sodium deposition characteristics more obvious.

[0062] S4: Online Sodium Deposition Detection: The residual signal obtained from deep feature decomposition is used as input, and artificial intelligence algorithms are used to extract features and perform binary classification on the residual signal, thereby determining in real time whether sodium deposition occurs in the battery.

[0063] One-dimensional convolutional neural networks (CNNs) are used as the online detection algorithm. CNNs can fully utilize the temporal information and local features in the residual signal to achieve efficient capture and binary classification of abnormal patterns. The one-dimensional CNN used in this invention mainly consists of an input layer, a first convolutional block, a second convolutional block, and a fully connected layer. The structure of each module is detailed below:

[0064] The input data for the input layer is the residual signal obtained after deep feature decomposition, with an original length of 128 time steps. To meet the requirements of convolution operation, it is first reconstructed into a two-dimensional tensor with a shape of [128,1], where 128 represents the time step and 1 represents single-channel data.

[0065] The first convolutional block aims to initially extract local temporal features and downsample the signal to obtain the first feature map. The detailed structure of this convolutional block includes:

[0066] Convolutional layer: One-dimensional convolution operation is used, with a kernel size of 5, a stride of 1, and padding of 2 to keep the input length constant. In this layer, the number of input channels is expanded from 1 to 4, that is, the shape of the output feature map is [128, 4].

[0067] Pooling layer: One-dimensional max pooling is used, with a kernel size of 2 and a stride of 2, which halves the time dimension. The output shape after pooling is [64,4], which reduces the data dimensionality and enhances local features.

[0068] The second convolutional block further extracts deeper local features while increasing the number of channels, thereby enriching the feature representation. The specific structure of this convolutional block is as follows:

[0069] Convolutional layer: A one-dimensional convolution operation is performed on the output of the pooled first convolutional block. The kernel size is 5, the stride is 1, the padding is 2, and the number of output channels is set to 8, so that the output shape after convolution is [64,8].

[0070] Pooling layer: A one-dimensional max pooling operation is performed on the output of the convolutional layer. The kernel size is 2 and the stride is 2. The time step is further reduced, and the output shape is a second feature map of [32,8].

[0071] The output shape of the second convolutional block is [32, 8]. First, it is flattened to become a one-dimensional vector (length 32 * 8 = 256). Then, this vector passes through a series of fully connected layers for feature fusion and high-level information extraction.

[0072] The first fully connected layer maps the flattened vector to 64 dimensions and uses the ReLU activation function for nonlinear transformation.

[0073] The second fully connected layer (output layer): maps the features to an output node, outputting a real number. z Using the Sigmoid activation function Convert the output z into a scalar, where This represents the confidence level of the model's prediction for the category of "abnormal sodium precipitation," i.e. p The closer it is to 0, the lower the probability of sodium deposition in the battery; p The closer the value is to 1, the higher the probability of abnormal sodium deposition in the battery. If the probability of sodium deposition exceeds a set threshold, the sodium-ion battery under test is determined to have an abnormal sodium deposition. In this embodiment, the set threshold can be set to 0.5.

[0074] To enable the one-dimensional convolutional neural network model to accurately distinguish sodium anomalies, this invention employs supervised training using a large number of residual signals labeled as "normal" and "abnormal sodium analysis." During model training, a binary cross-entropy loss function is used as the objective function, with the following formula:

[0075] (5);

[0076] in, t i Indicates the first i True label of the sample (0 indicates normal, 1 indicates abnormal sodium precipitation). p i The model represents the "first" i The predicted probability of a sample falling into the category of "sodium precipitation" is calculated. The network parameters are iteratively updated using the backpropagation algorithm until the model achieves the expected accuracy and robustness on the validation set.

[0077] After training, the one-dimensional convolutional neural network model is embedded into an online detection system to achieve real-time processing of the residual signals collected in each charging cycle. When the model outputs the probability of abnormal sodium deposition... p When the preset threshold is exceeded, the system immediately triggers an early warning and transmits the alarm information to the battery management system and remote monitoring center via a standard communication interface. To ensure detection stability, the system can statistically fuse and analyze detection results from multiple consecutive cycles, further reducing the risk of accidental misjudgments.

[0078] This invention employs a one-dimensional convolutional neural network for feature extraction and binary classification of residual signals. First, the residual signal (length 128) is adjusted into a two-dimensional tensor of shape [128, 1]. Then, the signal passes through two convolutional blocks: the first convolutional block uses a kernel size of 5 (stride 1, padding 2) to expand a single channel into four feature maps, and reduces the time step to 64 using 1D max pooling (kernel size 2, stride 2); the second convolutional block uses similar convolutional parameters on the output of the first block to expand the number of channels to 8, and then uses pooling to reduce the time step to 32. Finally, the convolutional output is flattened into a 256-dimensional vector, mapped to 64 dimensions via a fully connected layer, and then mapped to 16 dimensions. The final output layer uses a Softmax activation function to output the sodium deposition probability. If the sodium deposition probability is greater than a set threshold, the sodium-ion battery under test is determined to have an abnormal sodium deposition; otherwise, it is considered normal aging.

[0079] It should be noted that using a one-dimensional convolutional neural network to extract features and perform binary classification on the residual signal, compared to existing methods for sodium deposition detection based on charging data, is not affected by battery aging factors. It can directly extract features and perform binary classification on residual signals that only contain abnormal sodium deposition information, making online sodium deposition detection more accurate.

[0080] S5: Model Training and Evaluation: The charging curve data of SIB#1 and SIB#2 are used as the training set, and the trained model is tested on the independent test sets SIB#3 and SIB#4.

[0081] In step S5, charging curve data from SIB#1 and SIB#2 are selected as the training set. Specifically, an autoencoder is used to model the aging features of the original charging curve data to obtain the "healthy" charging curve corresponding to each sample. The residual between the original charging curve data and the reconstructed charging curve data is calculated to form the residual signal data for training. In the training data, the charging curves from SIB#1 are all in a sodium-free state, while the charging data from SIB#2 gradually shows sodium-free abnormalities in the later stages. Therefore, they are manually labeled as "normal" and "abnormal sodium-free," respectively, to construct a supervised training set. The Adam optimizer and mean squared error are used as the objective function of the autoencoder. After about 100 epochs of training, the autoencoder achieves a low reconstruction error. Subsequently, the 1D convolutional neural network model is trained using the cross-entropy loss function for about 200 epochs. On the validation set, the learning rate is reduced when there is no significant progress within 30 consecutive epochs.

[0082] In step S5, the trained model was evaluated using independent test sets (SIB#3 and SIB#4). The proposed detection model achieved an overall accuracy of approximately 97.2%, precision of 98.1%, recall of 85.3%, and an F1 score of 91.2%. Further periodic statistical analysis of the detection results from SIB#3 and SIB#4 revealed that the detection accuracy was particularly high in the early and late stages of battery charging, effectively capturing subtle changes in sodium deposition anomalies. The model's stability and robustness were validated through repeated experiments, demonstrating the effectiveness of this method in detecting sodium deposition anomalies in sodium-ion batteries in practical applications.

[0083] Based on the same concept, the present invention also provides an online sodium deposition detection system for sodium-ion batteries, including a data acquisition module, a reconstruction module and a detection module.

[0084] The acquisition module is used to acquire raw charging curve data of the sodium-ion battery under test in real time during the charging process.

[0085] Furthermore, in the data acquisition module, the system uses high-precision sensors installed in the sodium-ion battery or BMS, with voltage accuracy of 0.01V, current accuracy of 0.1mA, and temperature accuracy of ±0.1℃, to collect various electrochemical data generated during battery charging in real time. Multi-channel synchronous data transmission is achieved using CAN (Controller Area Network) bus, SPI (Serial Peripheral Interface), or wireless interfaces (such as Wi-Fi and Bluetooth). At the same time, a unified timestamp is assigned to the data, and it is initially stored in a local buffer to ensure data timing consistency and integrity.

[0086] The reconstruction module is used to input the original charging curve data into a pre-trained autoencoder to process and obtain reconstructed charging curve data. The reconstructed charging curve data is the charging curve data corresponding to the sodium-ion battery under test in the normal aging state.

[0087] Furthermore, an autoencoder model is used to model the aging features of the original charging curve data. The encoder reduces the data from 128 dimensions to an 8-dimensional latent space layer by layer, while the decoder uses a symmetrical structure to reconstruct the latent variables into a 128-dimensional "healthy" charging curve. Thus, the difference between the original charging curve data and the reconstructed charging curve data is calculated to obtain a residual signal reflecting abnormal sodium precipitation information. This signal, after necessary processing, serves as the input to the anomaly detection module.

[0088] The detection module is used to subtract the original charging curve data and the reconstructed charging curve data to obtain a residual signal containing abnormal information; the residual signal is subjected to feature extraction and feature mapping to obtain the sodium deposition probability; and the sodium deposition probability is used to detect in real time whether the sodium-ion battery under test has a sodium deposition abnormality.

[0089] Furthermore, a one-dimensional convolutional neural network is used to extract features and perform binary classification on the residual signal. The data is converted into two states, "normal aging" and "abnormal sodium precipitation", according to a predetermined algorithm, and the detection results are output.

[0090] The online sodium precipitation detection system for sodium-ion batteries of the present invention also includes a data preprocessing module and a control and early warning module.

[0091] The data preprocessing module performs digital filtering, low-pass filtering, and wavelet denoising on the collected raw charging curve data to remove high-frequency noise and abnormal fluctuations. Then, cubic B-spline interpolation is used to smooth and reconstruct the data, and normalization is used to standardize the data to form a standardized charging curve that is uniformly distributed within the preset charging voltage range, providing accurate input for the deep feature decomposition module.

[0092] Based on the results output by the anomaly detection module, the control and early warning module statistically analyzes the detection data from multiple consecutive charging cycles. Once the probability of abnormal sodium precipitation exceeds a preset threshold of 0.5, an early warning signal is sent via a local alarm system or remote communication (such as CAN bus, Wi-Fi, or Bluetooth). Simultaneously, charging parameters are automatically adjusted to ensure battery safety. All collected data, processing results, and alarm information are stored in real-time in local and cloud databases for subsequent system optimization and fault diagnosis.

[0093] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described online sodium deposition detection method for sodium-ion batteries.

[0094] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described online sodium deposition detection method for sodium-ion batteries.

[0095] This invention provides an online sodium deposition detection solution for sodium-ion batteries. By non-invasively and online detecting abnormal sodium deposition, it reduces safety risks caused by sodium deposition and extends battery life. The system features integrated design, good scalability and industrial application prospects, and is easy to integrate with existing battery management systems for widespread application.

[0096] This invention enables the effective separation and real-time detection of abnormal sodium deposition signals and normal aging signals during the charging process of sodium-ion batteries, providing high-quality, low-noise input data for subsequent anomaly detection modules, thereby improving the real-time performance and accuracy of the overall online sodium deposition detection.

[0097] This invention employs a detection framework that combines deep feature decomposition with a one-dimensional convolutional neural network, which can capture abnormal sodium deposition signals during charging in real time. This effectively overcomes the problems of poor real-time performance and slow response speed in traditional detection methods, prevents safety accidents caused by sodium deposition, and provides a reliable decision-making basis for subsequent intelligent battery management.

[0098] This invention achieves rapid and accurate detection of sodium deposition anomalies by preprocessing, aging feature modeling, residual signal extraction, and artificial intelligence detection of partial charging curve data collected during the charging process of sodium-ion batteries. Specifically, the method involves collecting charging curve data within a preset charging voltage range and obtaining standardized data through filtering, interpolation, and normalization; using an autoencoder to model aging features of normal charging data and obtaining a residual signal by calculating the difference between the original and reconstructed data; subsequently, using a one-dimensional convolutional neural network to extract temporal features from the residual signal and determining the abnormal sodium deposition state through a fully connected layer. The detection system of this invention also integrates modules for data acquisition, preprocessing, deep feature decomposition, anomaly detection, and alarm feedback, enabling real-time online monitoring, improving detection accuracy, effectively reducing safety risks caused by sodium deposition, and extending the service life of sodium-ion batteries.

[0099] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0100] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for detecting sodium deposition in an online sodium-ion battery, characterized in that, Includes the following steps: Real-time acquisition of raw charging curve data of the sodium-ion battery under test during the charging process; The original charging curve data is input into a pre-trained autoencoder to obtain reconstructed charging curve data, which is the charging curve data of the sodium-ion battery under test in normal aging state. The original charging curve data and the reconstructed charging curve data are subtracted to obtain a residual signal containing abnormal information; the residual signal is subjected to feature extraction and feature mapping to obtain the sodium deposition probability; the sodium deposition probability is used to detect in real time whether the sodium-ion battery under test has a sodium deposition abnormality. The sodium analysis probability is obtained by a pre-trained one-dimensional convolutional neural network, which includes an input layer, a first convolutional block, a second convolutional block, and a fully connected layer. The input layer reconstructs the shape of the residual signal to obtain a two-dimensional tensor with shape [128,1]. The first convolutional block extracts features from the two-dimensional tensor, resulting in a first feature map with shape [64,4]. The second convolutional block extracts features from the first feature map, resulting in a second feature map with shape [32, 8]. The fully connected layer flattens and maps the second feature map to obtain the sodium precipitation probability. The residual signal is as follows: ； In the formula, For residual signals, The original charging curve data, To reconstruct the charging curve data; If the sodium deposition probability is greater than a set threshold, the sodium-ion battery under test is determined to have an abnormal sodium deposition.

2. The method for online sodium deposition detection in a sodium-ion battery as described in claim 1, characterized in that, The autoencoder includes an encoder and a decoder. The encoder is used to extract aging features from the original charging curve data, and the decoder is used to reconstruct the aging features to obtain the reconstructed charging curve data.

3. The method for online sodium deposition detection in a sodium-ion battery as described in claim 1, characterized in that, The original charging curve data covers the preset charging voltage range of the sodium-ion battery under test, including voltage data and charging capacity data at multiple time points.

4. The method for online sodium deposition detection in a sodium-ion battery as described in claim 1, characterized in that, Before inputting the original charging curve data into the pre-trained autoencoder for aging feature extraction, the original charging curve data is preprocessed by data filtering, noise reduction, interpolation, and normalization.

5. A detection system for the online sodium deposition detection method of sodium-ion batteries according to claim 1, characterized in that, include: The acquisition module is used to acquire the raw charging curve data of the sodium-ion battery under test in real time during the charging process; The reconstruction module is used to input the original charging curve data into a pre-trained autoencoder to process and obtain reconstructed charging curve data, wherein the reconstructed charging curve data is the charging curve data corresponding to the sodium-ion battery under normal aging state. The detection module is used to subtract the original charging curve data and the reconstructed charging curve data to obtain a residual signal containing abnormal information; to perform feature extraction and feature mapping on the residual signal to obtain the sodium deposition probability; and to detect in real time whether the sodium-ion battery under test has experienced sodium deposition abnormality based on the sodium deposition probability.

6. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the online sodium deposition detection method for sodium-ion batteries according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the online sodium deposition detection method for sodium-ion batteries according to any one of claims 1-4.

Images