A lithium-sulfur battery shuttle effect inhibition method and system based on deep learning

Abstract

The application provides a lithium-sulfur battery shuttle effect inhibition method and system based on deep learning, and belongs to the technical field of lithium-sulfur batteries. The inhibition method comprises the following steps: S1, collecting real-time charging current, voltage, temperature data and electrolyte light transmittance data, and performing pretreatment; S2, sequentially calculating an initial node feature matrix, a first layer graph feature, an enhanced graph feature and a shuttle effect feature; S3, calculating a shuttle dynamic gain feature, extracting a shuttle speed feature and predicting a polysulfide shuttle speed; S4, constructing a dynamic compensation decision function and calculating a first pulse current amplitude; then, judging whether it is in a strong shuttle active period and triggering a pulse charging mode switching condition; and S5, encoding pulse charging parameters and conventional charging parameters into a data packet and sending the data packet to a charging device. The application can solve the problem that, under the working condition of high sulfur loading and high rate charging, the traditional method cannot dynamically compensate the battery capacity decay caused by the loss of polysulfide shuttle.

Description

Technical Field

[0001] This invention belongs to the field of lithium-sulfur battery technology, specifically relating to a method and system for suppressing the shuttle effect in lithium-sulfur batteries based on deep learning. Background Technology

[0002] In the development of lithium-sulfur battery technology, suppressing the shuttle effect of polysulfides is a key challenge to improve its cycle stability and practical application potential. The shuttle effect originates from soluble intermediates generated during charge and discharge—lithium polysulfides (Li₂S₂) x The repeated migration and irreversible deposition of polysulfides (4≤x≤8) between the positive and negative electrodes leads to the loss of active material, a decrease in coulombic efficiency, and rapid capacity decay, especially under high sulfur loading (>5 mg / cm²) and high-rate charging (≥1C) conditions. Traditional suppression strategies mainly rely on material structure design, such as constructing porous conductive frameworks, introducing polar host materials or functional membrane coatings, to slow down polysulfide diffusion through physical confinement or chemical adsorption. In addition, constant current-constant voltage charging management strategies are also used to try to compensate for capacity loss by extending charging time. However, these methods are mostly static or empirical designs, making it difficult to sense and respond to the dynamic shuttle behavior inside the battery in real time, especially under high load and high rate conditions, and failing to achieve accurate tracking and dynamic compensation for active material loss.

[0003] In recent years, with the expansion of deep learning technology in electrochemical systems, researchers have attempted to use neural network models to analyze battery operating data to predict performance degradation trends. A common approach is to construct fully connected networks or recurrent neural networks for state estimation based on cyclic voltammetry curves, impedance spectra, or voltage-time series. However, these methods generally rely on single-mode signal input, lack effective fusion of multi-source data, and most models are only used for post-hoc diagnosis or lifetime prediction, failing to form a closed-loop control mechanism with charging strategies. More importantly, existing deep learning models often neglect the nonlinear dynamic relationship between polysulfide concentration evolution and electrochemical response, resulting in insufficient prediction accuracy and an inability to trigger protective charging modes in a timely manner during critical phase transition stages (such as the Li2S2 / Li2S deposition period).

[0004] Therefore, there is an urgent need for an intelligent control method that can integrate multi-source real-time sensing information, analyze shuttle dynamics, and dynamically optimize charging behavior to overcome the problems of slow response and poor adaptability of traditional strategies, and to achieve the identification and active suppression of the shuttle effect of high-load lithium-sulfur batteries. Summary of the Invention

[0005] In view of the problems existing in the background technology, the purpose of this invention is to provide a method and system for suppressing the shuttle effect in lithium-sulfur batteries based on deep learning.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] A deep learning-based method for suppressing the shuttle effect in lithium-sulfur batteries includes the following steps:

[0008] S1: Collect real-time charging current, voltage, temperature data and electrolyte transmittance data, and perform preprocessing to obtain preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence and preprocessed electrolyte transmittance sequence.

[0009] S2: Based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence, and preprocessed electrolyte transmittance sequence, the initial node feature matrix, first layer graph features, enhanced graph features, and shuttle effect features are calculated sequentially; the first layer graph features are constructed based on the initial node feature matrix, sulfur loading, and charging rate, and a high load adaptation gate is constructed, which is then combined with the adjacency matrix to modulate the message passing matrix.

[0010] S3: Based on the characteristics of the shuttle effect and the preprocessed voltage sequence, calculate the shuttle dynamic gain characteristics, extract the shuttle rate characteristics, and predict the polysulfide shuttle rate.

[0011] S4: Based on the polysulfide shuttle rate, construct a dynamic compensation decision function and calculate the amplitude of the first pulse current; at the same time, based on the polysulfide shuttle rate and shuttle effect characteristics, determine whether to enter the strong shuttle active period. If the strong shuttle active period is entered, trigger the pulse charging mode and generate pulse charging parameters in the pulse charging mode; otherwise, maintain the current constant current charging state and generate conventional charging parameters.

[0012] The calculation process for the amplitude of the first pulse current is as follows: combining the non-negative cutoff term of the voltage deviation with the temperature-related Sigmoid adjustment term, the polysulfide shuttle rate is nonlinearly mapped through the hyperbolic tangent function, and finally summed to obtain the value.

[0013] S5: Encode the pulse charging parameters or regular charging parameters into a data packet and send it to the charging device.

[0014] Furthermore, the specific process of step S2 is as follows:

[0015] S21: Based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence, and preprocessed electrolyte transmittance sequence, calculate the initial node feature matrix. The calculation method is as follows:

[0016]

[0017]

[0018] in, Let be the node feature vector at time step t, where t is the time step. For splicing operations, This represents the value at time step t in the preprocessed current sequence. This represents the value at time step t in the preprocessed voltage sequence. This represents the value at time step t in the preprocessed temperature sequence. This represents the value at time step t in the transmittance sequence of the pretreated electrolyte. The initial node feature matrix, Let N be the node feature vector at the Nth time step, where N is the total number of time steps.

[0019] S22: Based on the initial node feature matrix, sulfur loading, and charging rate, construct a high-load adaptive gate and calculate the first-layer graph features. The calculation method is as follows:

[0020]

[0021]

[0022]

[0023] in, For message passing matrix, It is an adjacency matrix, in which each node is connected only to its directly adjacent preceding node and its directly adjacent succeeding node. For message passing weight matrix, For message passing bias vector, For high-load adaptability doors, For the Sigmoid function, To adapt the weight matrix to high load, For sulfur loading, This refers to the charging rate. For high load adaptive bias vectors, It is the dot product;

[0024] S23: Based on the first layer graph features, deep feature propagation is performed through a graph convolutional network to obtain enhanced graph features. The calculation method is as follows:

[0025]

[0026] in, For graph convolutional networks;

[0027] S24: Based on the enhanced image features and the preprocessed electrolyte transmittance sequence, a shuttle-sensitive attention mechanism is constructed to generate shuttle effect features. The calculation method is as follows:

[0028]

[0029]

[0030]

[0031]

[0032] in, As a characteristic of light transmittance, For Long Short-Term Memory (LSTM) networks, For the interactive energy matrix, For the interaction energy weight matrix, It is a multilayer perceptron. To navigate through sensitive attention weights, For the Softmax function, Characterized by the shuttle effect, This is the transpose symbol.

[0033] It should be further explained that the construction of the first layer of graph features aims to solve the problem of difficult dynamic response modeling of lithium-sulfur batteries under high sulfur loading and high rate charging conditions, which is caused by the aggravated polysulfide shuttle effect.

[0034] The characteristic of this operating condition is that the extremely high active material load and the demand for rapid ion migration exacerbate the nonlinear dissolution and diffusion of polysulfides, causing drastic changes in the internal state of the battery that are difficult to capture by traditional steady-state models. The challenge lies in how to accurately extract features strongly correlated with the shuttle effect in a highly dynamic and coupled physicochemical process.

[0035] In terms of technical logic, this invention introduces a high-load adaptive gate mechanism, using sulfur loading and charging rate as gating signals to modulate the message passing matrix. Specifically, sulfur loading directly affects the total amount of polysulfides generated, while the charging rate determines the electrochemical reaction rate and ion migration speed; both together determine the intensity and dynamic characteristics of the shuttle effect. The high-load adaptive gate maps these two macroscopic operating parameters to a gating value between zero and one using a sigmoid function. This gating value is then used to modulate the message passing matrix using a dot product, enabling the message passing process to adapt to the severity of the current operating conditions. The system adaptively enhances or suppresses the flow of feature information: when the sulfur load is high and the charging rate is high, the gate value approaches one, and the message passing matrix is ​​fully preserved, thus emphasizing the high-speed change relationship between features in adjacent time steps to capture the dramatic shuttle dynamics; conversely, under mild operating conditions, the gate value decreases to suppress high-frequency changes that may introduce noise and highlight steady-state trends. This modulation effect allows the first layer graph features to embed prior knowledge of operating conditions at the beginning of construction, providing a more accurate and representative initial spatiotemporal feature representation for subsequent deep graph convolutional networks, thus laying the foundation for accurate prediction of the shuttle rate.

[0036] In existing technologies, modeling the state of lithium-sulfur batteries typically relies on a single voltage or current signal and employs fixed empirical models or simplified electrochemical models. These methods struggle to effectively integrate multi-source heterogeneous sensor data (such as electrolyte transmittance) and cannot adapt to drastic dynamic changes under high load and high rate conditions, resulting in delayed and inaccurate perception of the shuttle effect. Compared to existing technologies, the advantage of this invention lies in introducing a high-load adaptive gating mechanism, using sulfur loading and charging rate as key gating signals to directly modulate the message passing process. This constructs a first-layer graph feature that more accurately reflects the changes in the battery's internal state under high dynamic conditions. As the cornerstone of subsequent deep feature extraction and fusion in the entire graph neural network, the improved characterization quality of this first-layer graph feature significantly enhances the subsequent ability to capture the polysulfide shuttle effect, providing crucial high-quality input for ultimately achieving accurate shuttle rate prediction.

[0037] This scenario involves multi-source data. The core challenge in fusion lies in the fact that macroscopic operating parameters (such as sulfur load and charging rate) and microscopic time-series graph node features (such as current, voltage, temperature, and transmittance sequences) belong to different scales, dimensions, and physical meanings. Traditional methods struggle to effectively couple these two types of heterogeneous information into a unified model without introducing bias or information loss. This invention constructs a lightweight gating network (i.e., a high-load adaptive gate), which is then directly applied to the message passing matrix calculated from microscopic time-series features via dot product. This achieves adaptive adjustment of the message passing intensity based on operating conditions, enabling the fusion of macroscopic operating conditions and microscopic features in the same graph calculation process. This avoids information redundancy or conflicts caused by the complex manual feature engineering and hard splicing in traditional methods, thus improving the model's performance under complex operating conditions.

[0038] Furthermore, the specific process of step S3 is as follows:

[0039] S31: Based on the characteristics of the shuttle effect and the preprocessed voltage sequence, construct a voltage-sensitive gating mechanism and calculate the dynamic gain characteristics of the shuttle. The calculation method is as follows:

[0040]

[0041]

[0042]

[0043] in, It is a voltage change sequence. This is a first-order difference operation, where V is the preprocessed voltage sequence. It is a voltage-sensitive gate. For the Sigmoid function, For splicing operations, This is the voltage-sensitive weighting matrix. It is a mean function. It is the variance function. For voltage-sensitive bias vector, Characterized by the shuttle effect;

[0044] S32: Based on the shuttle dynamic gain characteristics and shuttle effect characteristics, the shuttle rate characteristics are extracted, and the polysulfide shuttle rate is predicted. The calculation method is as follows:

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] in, The first characteristic is the shuttle speed. It is a multilayer perceptron. The second characteristic of shuttle speed is The shuttle speed characteristic, As a characteristic of light transmittance, For rate mapping vectors, For the Softmax function, For rate mapping weight matrix, For rate mapping bias vector, For polysulfide shuttle rate, This is the transpose symbol.

[0051] It should be further explained that the introduction of shuttle dynamic gain characteristics aims to solve the problem of interference caused by large voltage fluctuations during high-rate charging on the prediction of polysulfide shuttle rate.

[0052] The characteristic of this scenario is that when lithium-sulfur batteries are charged at high rates, the electrode polarization intensifies, and the voltage signal changes rapidly and nonlinearly. This change is coupled with the voltage decay caused by the polysulfide shuttle itself, forming a complex dynamic system. The challenge lies in directly separating the dynamic component dominated by the shuttle effect from the shuttle effect characteristics that integrate multi-source information, and quantifying its contribution to rate prediction.

[0053] In terms of technical logic, this invention achieves this goal by constructing a voltage-sensitive gating mechanism; specifically, the first-order difference sequence of voltage directly reflects the instantaneous rate of change of the external terminal voltage of the battery, and its mean and variance characterize the average trend and fluctuation intensity of voltage change, respectively; the voltage-sensitive gate maps these two statistics to a gating value through the Sigmoid function, and the gating value then selectively modulates the shuttle effect characteristics;

[0054] The core logic of this design is as follows: when the voltage fluctuation is drastic (i.e., the variance is large), the gate value will automatically decrease, thereby suppressing the part of the shuttle effect feature that may be contaminated by voltage noise; when the voltage change is stable, the gate value will increase, allowing more feature information to pass through; the shuttle dynamic gain feature obtained after gate modulation is essentially a purified and enhanced representation that is insensitive to voltage dynamic changes, and it focuses more on reflecting the inherent law of the shuttle effect itself; this feature is then fused and weighted with the original shuttle effect feature and transmittance feature, so that the prediction result of the polysulfide shuttle rate can make full use of multi-source information and effectively resist the interference of external voltage disturbances, significantly improving the robustness and accuracy of the prediction model;

[0055] In terms of data fusion, this invention utilizes the statistical characteristics (mean and variance) of the voltage sequence itself as a gating signal to construct a lightweight built-in filtering mechanism, achieving an "adaptive filtering" effect. It eliminates the need for manual design of filter parameters and automatically learns the correlation between voltage fluctuations and feature reliability through a data-driven approach. Ultimately, it achieves dynamic noise reduction and enhancement at the feature level, providing a cleaner and more reliable input basis for subsequent rate prediction.

[0056] Furthermore, the specific process of step S4 is as follows:

[0057] S41: Calculate the amplitude of the first pulse current based on the polysulfide shuttle rate. The calculation method is as follows:

[0058]

[0059] in, Let be the amplitude of the first pulse current at time step t. It is the hyperbolic tangent function. This is the current pulse weighting matrix. This is the current pulse bias vector. This is the voltage regulation coefficient. This is the reference voltage for polysulfide deposition. This is a non-negative truncation operation. This is the temperature regulation coefficient;

[0060] S42: Based on the polysulfide shuttle rate and shuttle effect characteristics, determine whether a strong shuttle activity period has been entered, and determine whether to trigger a pulse charging mode switch based on the determination result. The determination logic is as follows:

[0061] when ,and If the pulse charging mode is activated, proceed to step S43; otherwise, do not activate the pulse charging mode and proceed to step S44.

[0062] in, The shuttle rate threshold, The gradient norm of the shuttle effect characteristics over time. Gradient-sensitive threshold;

[0063] S43: In pulse charging mode, based on the shuttle rate and current voltage, pulse charging parameters are generated, including the amplitude of the second pulse current, the current on-time length of a single pulse cycle, the current off-time length of a single pulse cycle, the number of pulse cycles, and the charging cutoff voltage. The calculation methods are as follows:

[0064]

[0065]

[0066]

[0067]

[0068]

[0069] in, Let be the amplitude of the second pulse current at time step t. To perform the minimum value operation, For the upper limit of current safety, The current conduction time is the length of a single pulse cycle. The minimum pulse on-time for a single pulse cycle. The maximum pulse on-time for a single pulse cycle. This is the conduction time adjustment coefficient. This is the voltage difference normalization factor. The current-off time length for a single pulse cycle. The minimum interval time of a single pulse cycle. The maximum interval time of a single pulse cycle. To perform an exponential operation on the natural constant, The attenuation coefficient is the intermittent time. This is the temperature-adjustable gain coefficient. As a reference temperature threshold, The number of pulse cycles. This is the charging cutoff voltage;

[0070] S44: In non-pulse charging mode, maintain the current constant current charging state and generate conventional charging parameters; the conventional charging parameters include charging current value and charging cut-off voltage; the charging current value is determined using a constant current charging setting method based on battery capacity, and the charging cut-off voltage is calculated in the same way as the charging cut-off voltage in pulse charging mode.

[0071] Furthermore, the shuttle rate threshold and gradient sensitive threshold Configure according to your needs.

[0072] It should be further explained that the dynamic generation of pulse charging parameters aims to solve the problem of irreversible loss of active material and rapid capacity decay caused by polysulfide shuttle effect under high sulfur loading and high rate charging conditions. The characteristic of this scenario is that the shuttle process is highly dynamic and nonlinear. The traditional constant current charging mode cannot compensate for the loss of active sulfur caused by polysulfide dissolution and diffusion in real time. The difficulty lies in how to dynamically generate a complex pulse charging strategy that can effectively suppress shuttle while ensuring charging efficiency and safety based on the real-time prediction of shuttle rate.

[0073] In terms of technical logic, this invention achieves precise pulse control by constructing a dynamic compensation decision function and mode switching conditions;

[0074] Specifically, the dynamic compensation decision function takes the predicted polysulfide shuttle rate as its core input and maps it to a compensation coefficient between -1 and +1 using a hyperbolic tangent function. This coefficient determines the direction and intensity of the base current compensation required. The function also incorporates a non-negative cutoff term for voltage deviation and a temperature-dependent sigmoid adjustment term, which together constitute the amplitude of the first pulse current. The underlying logic of this design is that the shuttle rate processed by the hyperbolic tangent function directly reflects the intensity and direction of the compensation demand (positive or negative compensation); the voltage deviation term ensures that the instantaneous high current is used to promote polysulfide reduction and deposition rather than exacerbate shuttle operation; and the temperature adjustment term avoids the thermal risks associated with overcompensation at extreme temperatures. This nonlinear combination of the three factors allows the generated pulse current amplitude to adapt to shuttle intensity, electrochemical environment, and thermal state. This mechanism achieves dynamic compensation and is designed based on the inherent characteristics of lithium-sulfur battery reactions: the electrochemical reduction and deposition reaction of polysulfides is strongly dependent on the potential environment, and its reaction rate is highest when it approaches the thermodynamic equilibrium potential (i.e., the polysulfide deposition reference voltage). When the real-time voltage approaches the polysulfide deposition reference voltage, this term generates a non-negative excitation signal. At this time, a pulse of instantaneously enhanced reduction current is applied, the electrochemical environment is ready, and energy will be efficiently used to drive polysulfide molecules near the interface to rapidly gain electrons, generate solid Li2S, and deposit it on the cathode framework. The kinetic rate of this process is much higher than the rate at which polysulfides diffuse from the bulk electrolyte to the electrode interface. Conversely, if the voltage is much higher than the polysulfide deposition reference voltage, applying a large reduction current is more likely to generate soluble polysulfides or trigger other side reactions, which may exacerbate the subsequent shuttle risk.

[0075] Subsequently, by judging whether the shuttle rate exceeds the threshold and whether its change gradient is drastic, the pulse mode switching is intelligently triggered to ensure that the pulse strategy is only applied during periods of strong shuttle activity. In pulse mode, various parameters (such as turn-on time and turn-off time) are non-linearly related to real-time shuttle rate, voltage, and temperature. For example, the turn-on time adaptively extends as the shuttle rate and voltage deviation increase to provide a longer deposition time; the turn-off time shortens as the shuttle rate increases, but is simultaneously regulated by temperature to avoid heat accumulation and ensure that the pulse strategy always matches the actual chemical state inside the battery.

[0076] Existing technologies generally employ fixed pulse patterns or triggering mechanisms based on simple voltage thresholds. Their parameters (such as pulse amplitude and duty cycle) are usually preset constants, which cannot adapt to different sulfur loading, different rates, and changes in the internal state during battery aging, resulting in poor compensation effects or even side effects.

[0077] Compared with existing technologies, the advantage of this invention is that it transforms the pulse charging strategy from a static, open-loop control method into a dynamic, adaptive optimization process based on real-time prediction of the battery's internal chemical state, which significantly improves the compensation effect and battery cycle life under high dynamic conditions.

[0078] The present invention also discloses a deep learning-based lithium-sulfur battery shuttle effect suppression system, including a data acquisition and preprocessing unit, a shuttle effect feature extraction unit, a shuttle rate prediction unit, a charging parameter generation unit, and a charging control unit.

[0079] The data acquisition and preprocessing unit acquires real-time charging current, voltage, temperature data and electrolyte transmittance data, and performs preprocessing to obtain preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence and preprocessed electrolyte transmittance sequence.

[0080] The shuttle effect feature extraction unit calculates the initial node feature matrix, first layer graph features, enhancement graph features, and shuttle effect features sequentially based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence, and preprocessed electrolyte transmittance sequence.

[0081] The shuttle rate prediction unit calculates the shuttle dynamic gain characteristics, extracts the shuttle rate characteristics, and predicts the polysulfide shuttle rate based on the shuttle effect characteristics and the preprocessed voltage sequence.

[0082] The charging parameter generation unit constructs a dynamic compensation decision function based on the polysulfide shuttle rate and calculates the amplitude of the first pulse current; then it determines whether it has entered a strong shuttle active period and triggers the pulse charging mode switching condition; in pulse charging mode, it generates pulse charging parameters; in non-pulse charging mode, it generates regular charging parameters.

[0083] The charging control unit encodes the pulse charging parameters and regular charging parameters into data packets and sends them to the charging device.

[0084] Furthermore, the specific working process of the charging control unit is as follows:

[0085] The charging control unit encodes pulse charging parameters or regular charging parameters into data packets according to the communication protocol standard between the battery management system and the charging equipment; it then sends the data packets to the control module of the charging equipment through the communication interface; the control module of the charging equipment parses the received data packets and configures the internal power output parameters; the charging equipment executes the charging operation according to the internal power output parameters and returns a status confirmation signal to the battery management system.

[0086] Furthermore, the communication protocol includes the CAN bus protocol, the I2C protocol, or the Modbus protocol.

[0087] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0088] (1) This invention provides a deep learning-based method for suppressing the shuttle effect in lithium-sulfur batteries, which effectively solves the capacity decay problem caused by polysulfide shuttle under high sulfur loading and high rate conditions. This invention integrates sulfur loading and charging rate into graph feature extraction through a high load adaptation gate to construct a shuttle effect feature that accurately characterizes the internal state of the battery. It uses a voltage-sensitive gating mechanism to generate a voltage-resistant dynamic gain feature of the shuttle, thereby accurately predicting the polysulfide shuttle rate. Based on this rate, the amplitude of the first pulse current is generated, and the pulse charging mode is intelligently triggered. Its pulse parameter generation mechanism integrates the voltage deviation term and the temperature regulation term to ensure that the pulse current is applied at the optimal deposition voltage window, promotes polysulfide immobilization, and ultimately achieves source suppression of the shuttle effect and a significant improvement in battery life.

[0089] (2) This invention innovatively proposes a high-load adaptive gating mechanism, which effectively solves the problem of fusing macroscopic operating parameters and microscopic time-series features. This mechanism uses sulfur loading and charging rate as gating signals, generates adaptive modulation coefficients through the Sigmoid function, and directly controls the strength of the message passing matrix in a dot product manner. This enables the model to dynamically adjust the feature extraction strategy according to the severity of the operating conditions: under high load and high rate conditions, the feature flow is strengthened to capture violent dynamics, and under mild operating conditions, noise is suppressed to highlight steady-state trends. This achieves seamless coupling of heterogeneous data and provides high-quality first-layer graph features that accurately reflect the internal state of the battery for subsequent networks, significantly improving the accuracy of polysulfide shuttle rate prediction.

[0090] (3) This invention proposes a voltage-sensitive gating mechanism, which effectively solves the problem of interference of drastic voltage fluctuations on the prediction of polysulfide shuttle rate under high-rate charging. The mechanism extracts the mean and variance of the first-order difference sequence of voltage as the gating signal, generates adaptive modulation coefficients through the Sigmoid function, and dynamically purifies the shuttle effect features. When the voltage fluctuation is drastic, it automatically suppresses noise interference and enhances feature transmission when the voltage is stable, thereby extracting the enhanced characterization that focuses on reflecting the intrinsic law of the shuttle effect. It realizes adaptive filtering at the feature level, and can remove voltage fluctuation noise without manual parameter design, which significantly improves the accuracy and robustness of polysulfide shuttle rate prediction.

[0091] (4) This invention innovatively constructs an adaptive pulse charging strategy based on real-time chemical state prediction. The strategy takes the polysulfide shuttle rate as the core input and generates the first pulse current amplitude by fusing the voltage deviation term and temperature regulation term through a dynamic compensation decision function. The voltage deviation term ensures that the pulse current is applied only in the optimal deposition voltage window, effectively promoting polysulfide solidification, while the temperature regulation term ensures thermal safety. Through intelligent mode switching and nonlinear parameter mapping, the dynamic matching of parameters such as on-time and off-time with the internal state of the battery is realized, which changes the crude mode of traditional fixed parameter pulse charging and realizes the source suppression of the shuttle effect and the significant improvement of battery life. Attached Figure Description

[0092] Figure 1 This is a schematic diagram of the algorithm flow for first-layer graph feature extraction provided by the present invention.

[0093] Figure 2 A schematic diagram of the algorithm flow for shuttle dynamic gain feature extraction provided by the present invention.

[0094] Figure 3 This is a schematic diagram of the lithium-sulfur battery shuttle effect suppression monitoring system provided by the present invention. Detailed Implementation

[0095] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

[0096] A deep learning-based method for suppressing the shuttle effect in lithium-sulfur batteries includes the following steps:

[0097] S1: Collect real-time charging current, voltage, temperature data, and electrolyte transmittance data, and perform preprocessing to obtain preprocessed current, voltage, temperature, and electrolyte transmittance sequences, including:

[0098] S11: Collect real-time charging current, voltage, temperature data and electrolyte transmittance data at equal time intervals, and arrange them in chronological order to obtain current sequence, voltage sequence, temperature sequence and electrolyte transmittance sequence.

[0099] S12: Perform maximum and minimum value normalization on the current sequence, voltage sequence, temperature sequence, and electrolyte transmittance sequence respectively to obtain the normalized current sequence, voltage sequence, temperature sequence, and electrolyte transmittance sequence.

[0100] S13: The normalized current sequence, voltage sequence, temperature sequence, and electrolyte transmittance sequence are smoothed by the moving average method to obtain the pre-processed current sequence, pre-processed voltage sequence, pre-processed temperature sequence, and pre-processed electrolyte transmittance sequence.

[0101] S2: Based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence, and preprocessed electrolyte transmittance sequence, calculate the initial node feature matrix, first-layer graph features, enhancement graph features, and shuttle effect features sequentially, including:

[0102] S21: Based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence, and preprocessed electrolyte transmittance sequence, calculate the initial node feature matrix. The calculation method is as follows:

[0103]

[0104]

[0105] in, Let be the node feature vector at time step t, where t is the time step. For splicing operations, This represents the value at time step t in the preprocessed current sequence. This represents the value at time step t in the preprocessed voltage sequence. This represents the value at time step t in the preprocessed temperature sequence. This represents the value at time step t in the transmittance sequence of the pretreated electrolyte. The initial node feature matrix, These are the node feature vectors for the 1st, 2nd, ..., Nth time steps, respectively, where N is the total number of time steps;

[0106] S22: Based on the initial node feature matrix, sulfur loading, and charging rate, construct a high-load adaptive gate and calculate the first-layer graph features, such as... Figure 1 As shown, the calculation method is as follows:

[0107]

[0108]

[0109]

[0110] in, For message passing matrix, It is an adjacency matrix, in which each node is connected only to its directly adjacent preceding node and its directly adjacent succeeding node. For message passing weight matrix, For message passing bias vector, For high-load adaptability doors, For the Sigmoid function, To adapt the weight matrix to high load, For sulfur loading, This refers to the charging rate. For high load adaptive bias vectors, Features of the first layer graph It is the dot product;

[0111] S23: Based on the features of the first layer graph, deep feature propagation is performed through a graph convolutional network to obtain enhanced graph features. The calculation method is as follows:

[0112]

[0113] in, To enhance graph features, For graph convolutional networks;

[0114] S24: Based on the enhanced image features and the preprocessed electrolyte transmittance sequence, a shuttle-sensitive attention mechanism is constructed to generate shuttle effect features. The calculation method is as follows:

[0115]

[0116]

[0117]

[0118]

[0119] in, As a characteristic of light transmittance, For Long Short-Term Memory (LSTM) networks, For the interactive energy matrix, It is a multilayer perceptron. To navigate through sensitive attention weights, For the Softmax function, Characterized by the shuttle effect, This is the transpose symbol.

[0120] S3: Based on the shuttle effect characteristics and the preprocessed voltage sequence, calculate the shuttle dynamic gain characteristics, extract the shuttle rate characteristics, and predict the polysulfide shuttle rate, including:

[0121] S31: Based on the characteristics of the shuttle effect and the preprocessed voltage sequence, a voltage-sensitive gating mechanism is constructed, and the dynamic gain characteristics of the shuttle are calculated, such as... Figure 2 As shown, the calculation method is as follows:

[0122]

[0123]

[0124]

[0125] in, It is a voltage change sequence. This is a first-order difference operation, where V is the preprocessed voltage sequence. It is a voltage-sensitive gate. For the Sigmoid function, For splicing operations, This is the voltage-sensitive weighting matrix. It is a mean function. It is the variance function. For voltage-sensitive bias vector, For shuttle dynamic gain characteristics, Characterized by the shuttle effect;

[0126] S32: Based on the shuttle dynamic gain characteristics and shuttle effect characteristics, the shuttle rate characteristics are extracted, and the polysulfide shuttle rate is predicted. The calculation method is as follows:

[0127]

[0128]

[0129]

[0130]

[0131]

[0132] in, The first characteristic is the shuttle speed. It is a multilayer perceptron. The second characteristic of shuttle speed is The shuttle speed characteristic, As a characteristic of light transmittance, For rate mapping vectors, For the Softmax function, For rate mapping weight matrix, For rate mapping bias vector, For polysulfide shuttle rate, This is the transpose symbol.

[0133] S4: Based on the polysulfide shuttle rate, construct a dynamic compensation decision function to calculate the amplitude of the first pulse current; then determine whether the strong shuttle activity period has been entered and trigger the pulse charging mode switching condition; in pulse charging mode, generate pulse charging parameters; in non-pulse charging mode, generate regular charging parameters, including:

[0134] S41: Based on the polysulfide shuttle rate, a dynamic compensation decision function is constructed to calculate the amplitude of the first pulse current. The calculation method is as follows:

[0135]

[0136] in, Let be the amplitude of the first pulse current at time step t. It is the hyperbolic tangent function. This is the current pulse weighting matrix. This is the current pulse bias vector. This is the voltage regulation coefficient. This is the reference voltage for polysulfide deposition. This is a non-negative truncation operation. This is the temperature regulation coefficient;

[0137] S42: Based on the polysulfide shuttle rate and shuttle effect characteristics, determine whether a strong shuttle activity period has been entered and trigger the pulse charging mode switching condition. The determination logic is as follows:

[0138] when ,and At this time, the pulse charging mode is activated;

[0139] in, The shuttle rate threshold, The gradient norm of the shuttle effect characteristics over time. Gradient-sensitive threshold;

[0140] S43: In pulse charging mode, based on the shuttle rate and current voltage, pulse charging parameters are generated, including the amplitude of the second pulse current, the current on-time length of a single pulse cycle, the current off-time length of a single pulse cycle, the number of pulse cycles, and the charging cutoff voltage. The calculation methods are as follows:

[0141]

[0142]

[0143]

[0144]

[0145]

[0146] in, Let be the amplitude of the second pulse current at time step t. To perform the minimum value operation, For the upper limit of current safety, The current conduction time is the length of a single pulse cycle. The minimum pulse on-time for a single pulse cycle. The maximum pulse on-time for a single pulse cycle. This is the conduction time adjustment coefficient. This is the voltage difference normalization factor. The current-off time length for a single pulse cycle. The minimum interval time of a single pulse cycle. The maximum interval time of a single pulse cycle. To perform an exponential operation on the natural constant, The attenuation coefficient is the intermittent time. This is the temperature-adjustable gain coefficient. As a reference temperature threshold, The number of pulse cycles. This is the charging cutoff voltage;

[0147] S44: In non-pulse charging mode, maintain the current constant current charging state and generate conventional charging parameters; the conventional charging parameters include charging current value and charging cut-off voltage; the charging current value is determined using a constant current charging setting method based on battery capacity, and the charging cut-off voltage is calculated in the same way as the charging cut-off voltage in pulse charging mode.

[0148] For example, for a single cell in a lithium-sulfur battery, the specific parameters set in this embodiment are as follows: sulfur loading. Charging rate The sampling interval is 1 second; the reference voltage for polysulfide deposition is... Current safety limit Reference temperature threshold shuttle rate threshold Gradient-sensitive threshold ;

[0149] The polysulfide shuttle rate r is a unit of... The time-series data; in the initial stage of charging (t=0−300s), the r value is low, typically in... arrive The fluctuations between these values ​​indicate a weak shuttle effect; during the middle of the charging process, the r value gradually increases, possibly from... Rise to This reflects the intensified dissolution and diffusion of polysulfides; in the later stages of charging, if well controlled, the r value can be reduced to... the following;

[0150] Based on the real-time predicted value of r, the system dynamically generates pulse parameters:

[0151] when ,Voltage (higher than) ),temperature At that time, the system determined that the strong shuttle activity period had not been reached and maintained normal charging (constant current 2C, about 4.0A).

[0152] when Rise to ,and , At that time, the pulse mode is triggered:

[0153] Calculated , and then calculate , , ;

[0154] when Further rise to , Down to (near ), hour:

[0155] Calculated , , , .

[0156] S5: Encodes the pulse charging parameters and regular charging parameters into a data packet and sends it to the charging device, including:

[0157] S51: According to the communication protocol standard between the battery management system and the charging equipment, the pulse charging parameters and the regular charging parameters are encoded into data packets; the communication protocol includes CAN bus protocol, I2C protocol or Modbus protocol;

[0158] S52: Control module that sends data packets to the charging device via the communication interface;

[0159] S53: The control module of the charging device parses the received data packet and configures the internal power output parameters;

[0160] S54: The charging device performs the charging operation according to the internal power output parameters and returns a status confirmation signal to the battery management system.

[0161] Example 2

[0162] A deep learning-based system for suppressing the shuttle effect in lithium-sulfur batteries includes:

[0163] Data acquisition and preprocessing unit: Acquires real-time charging current, voltage, temperature data and electrolyte transmittance data, and performs preprocessing to obtain preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence and preprocessed electrolyte transmittance sequence;

[0164] Shuttle effect feature extraction unit: Based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence and preprocessed electrolyte transmittance sequence, calculate the initial node feature matrix, first layer graph features, enhancement graph features and shuttle effect features in sequence;

[0165] Shuttle rate prediction unit: Based on the shuttle effect characteristics and the preprocessed voltage sequence, calculates the shuttle dynamic gain characteristics, extracts the shuttle rate characteristics, and predicts the polysulfide shuttle rate, such as... Figure 3 As shown;

[0166] Charging parameter generation unit: Based on the polysulfide shuttle rate, construct a dynamic compensation decision function to calculate the amplitude of the first pulse current; then determine whether it has entered the strong shuttle active period and trigger the pulse charging mode switching condition; in pulse charging mode, generate pulse charging parameters; in non-pulse charging mode, generate regular charging parameters.

[0167] Charging control unit: Encodes pulse charging parameters and regular charging parameters into data packets and sends them to the charging device.

[0168] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0169] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0170] The above description is merely a specific embodiment of the present invention. Any feature disclosed in this specification may be replaced by other equivalent or similar features unless otherwise specified. All disclosed features, or steps in all methods or processes, may be combined in any way except for mutually exclusive features and / or steps.

Claims

1. A method for suppressing the shuttle effect in lithium-sulfur batteries based on deep learning, characterized in that, Includes the following steps: S1: Collect real-time charging current, voltage, temperature data and electrolyte transmittance data, and perform preprocessing to obtain preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence and preprocessed electrolyte transmittance sequence. S2: Based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence, and preprocessed electrolyte transmittance sequence, the initial node feature matrix, first layer graph features, enhanced graph features, and shuttle effect features are calculated sequentially; the first layer graph features are constructed based on the initial node feature matrix, sulfur loading, and charging rate, and a high load adaptation gate is constructed, which is then combined with the adjacency matrix to modulate the message passing matrix. S3: Based on the characteristics of the shuttle effect and the preprocessed voltage sequence, calculate the shuttle dynamic gain characteristics, extract the shuttle rate characteristics, and predict the polysulfide shuttle rate. S4: Based on the polysulfide shuttle rate, construct a dynamic compensation decision function and calculate the amplitude of the first pulse current; at the same time, based on the polysulfide shuttle rate and shuttle effect characteristics, determine whether the strong shuttle active period has been entered. If the strong shuttle active period has been entered, trigger the pulse charging mode and generate pulse charging parameters in the pulse charging mode. Otherwise, maintain the current constant current charging state and generate normal charging parameters; The calculation process for the amplitude of the first pulse current is as follows: combining the non-negative cutoff term of the voltage deviation with the temperature-related Sigmoid adjustment term, the polysulfide shuttle rate is nonlinearly mapped through the hyperbolic tangent function, and finally summed to obtain the value. S5: Encode the pulse charging parameters or regular charging parameters into a data packet and send it to the charging device.

2. The method for suppressing the shuttle effect in lithium-sulfur batteries based on deep learning as described in claim 1, characterized in that, The specific process of step S2 is as follows: S21: Based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence, and preprocessed electrolyte transmittance sequence, calculate the initial node feature matrix. The calculation method is as follows: ， in, Let be the node feature vector at time step t, where t is the time step. For splicing operations, This represents the value at time step t in the preprocessed current sequence. This represents the value at time step t in the preprocessed voltage sequence. This represents the value at time step t in the preprocessed temperature sequence. This represents the value at time step t in the transmittance sequence of the pretreated electrolyte. The initial node feature matrix, Let N be the node feature vector at the Nth time step, where N is the total number of time steps. S22: Based on the initial node feature matrix, sulfur loading, and charging rate, construct a high-load adaptive gate and calculate the first-layer graph features. The calculation method is as follows: ， in, For message passing matrix, It is an adjacency matrix, in which each node is connected only to its directly adjacent preceding node and its directly adjacent succeeding node. For message passing weight matrix, For message passing bias vector, For high-load adaptability doors, For the Sigmoid function, To adapt the weight matrix to high load, For sulfur loading, This refers to the charging rate. For high load adaptive bias vectors, It is the dot product; S23: Based on the first layer graph features, deep feature propagation is performed through a graph convolutional network to obtain enhanced graph features. The calculation method is as follows: ， in, For graph convolutional networks; S24: Based on the enhanced image features and the preprocessed electrolyte transmittance sequence, a shuttle-sensitive attention mechanism is constructed to generate shuttle effect features. The calculation method is as follows: ， in, As a characteristic of light transmittance, For Long Short-Term Memory (LSTM) networks, For the interactive energy matrix, For the interaction energy weight matrix, It is a multilayer perceptron. To navigate through sensitive attention weights, For the Softmax function, Characterized by the shuttle effect, This is the transpose symbol.

3. The method for suppressing the shuttle effect in lithium-sulfur batteries based on deep learning as described in claim 2, characterized in that, The specific process of step S3 is as follows: S31: Based on the characteristics of the shuttle effect and the preprocessed voltage sequence, construct a voltage-sensitive gating mechanism and calculate the dynamic gain characteristics of the shuttle. The calculation method is as follows: ， in, It is a voltage change sequence. This is a first-order difference operation, where V is the preprocessed voltage sequence. It is a voltage-sensitive gate. For the Sigmoid function, For splicing operations, This is the voltage-sensitive weighting matrix. It is a mean function. It is the variance function. For voltage-sensitive bias vector, Characterized by the shuttle effect; S32: Based on the shuttle dynamic gain characteristics and shuttle effect characteristics, the shuttle rate characteristics are extracted, and the polysulfide shuttle rate is predicted. The calculation method is as follows: ， in, The first characteristic is the shuttle speed. It is a multilayer perceptron. The second characteristic of shuttle speed is The shuttle speed characteristic, As a characteristic of light transmittance, For rate mapping vectors, For the Softmax function, For rate mapping weight matrix, For rate mapping bias vector, For polysulfide shuttle rate, This is the transpose symbol.

4. The method for suppressing the shuttle effect in lithium-sulfur batteries based on deep learning as described in claim 3, characterized in that, The specific process of step S4 is as follows: S41: Calculate the amplitude of the first pulse current based on the polysulfide shuttle rate. The calculation method is as follows: ， in, Let be the amplitude of the first pulse current at time step t. It is the hyperbolic tangent function. This is the current pulse weighting matrix. This is the current pulse bias vector. This is the voltage regulation coefficient. This is the reference voltage for polysulfide deposition. This is a non-negative truncation operation. This is the temperature regulation coefficient; S42: Based on the polysulfide shuttle rate and shuttle effect characteristics, determine whether a strong shuttle activity period has been entered, and determine whether to trigger a pulse charging mode switch based on the determination result. The determination logic is as follows: when ,and If the pulse charging mode is activated, proceed to step S43; otherwise, do not activate the pulse charging mode and proceed to step S44. in, The shuttle rate threshold, The gradient norm of the shuttle effect characteristics over time. Gradient-sensitive threshold; S43: In pulse charging mode, based on the shuttle rate and current voltage, pulse charging parameters are generated, including the amplitude of the second pulse current, the current on-time length of a single pulse cycle, the current off-time length of a single pulse cycle, the number of pulse cycles, and the charging cutoff voltage. The calculation methods are as follows: ， in, Let be the amplitude of the second pulse current at time step t. To perform the minimum value operation, To ensure the safe upper limit of current, The current conduction time is the length of a single pulse cycle. The minimum pulse on-time for a single pulse cycle. The maximum pulse on-time for a single pulse cycle. This is the conduction time adjustment coefficient. This is the voltage difference normalization factor. The current-off time length for a single pulse cycle. The minimum interval time of a single pulse cycle. The maximum interval time of a single pulse cycle. For the exponential operation of taking the natural constant, The attenuation coefficient is the intermittent time. This is the temperature-adjustable gain coefficient. As a reference temperature threshold, The number of pulse cycles. This is the charging cutoff voltage; S44: In non-pulse charging mode, maintain the current constant current charging state and generate conventional charging parameters; the conventional charging parameters include charging current value and charging cut-off voltage; the charging current value is determined using a constant current charging setting method based on battery capacity, and the charging cut-off voltage is calculated in the same way as the charging cut-off voltage in pulse charging mode.

5. The method for suppressing the shuttle effect in lithium-sulfur batteries based on deep learning as described in claim 4, characterized in that, shuttle rate threshold and gradient sensitive threshold Configure according to your needs.

6. A deep learning-based system for suppressing the shuttle effect in lithium-sulfur batteries, characterized in that, It includes a data acquisition and preprocessing unit, a shuttle effect feature extraction unit, a shuttle rate prediction unit, a charging parameter generation unit, and a charging control unit; The data acquisition and preprocessing unit acquires real-time charging current, voltage, temperature data and electrolyte transmittance data, and performs preprocessing to obtain preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence and preprocessed electrolyte transmittance sequence. The shuttle effect feature extraction unit calculates the initial node feature matrix, first layer graph features, enhancement graph features, and shuttle effect features sequentially based on the preprocessed current sequence, preprocessed voltage sequence, preprocessed temperature sequence, and preprocessed electrolyte transmittance sequence. The shuttle rate prediction unit calculates the shuttle dynamic gain characteristics, extracts the shuttle rate characteristics, and predicts the polysulfide shuttle rate based on the shuttle effect characteristics and the preprocessed voltage sequence. The charging parameter generation unit constructs a dynamic compensation decision function based on the polysulfide shuttle rate and calculates the amplitude of the first pulse current; then it determines whether it has entered a strong shuttle active period and triggers the pulse charging mode switching condition; in pulse charging mode, it generates pulse charging parameters; in non-pulse charging mode, it generates regular charging parameters. The charging control unit encodes the pulse charging parameters and regular charging parameters into data packets and sends them to the charging device.

7. The lithium-sulfur battery shuttle effect suppression system as described in claim 6, characterized in that, The specific working process of the charging control unit is as follows: The charging control unit encodes pulse charging parameters or regular charging parameters into data packets according to the communication protocol standard between the battery management system and the charging equipment; it then sends the data packets to the control module of the charging equipment through the communication interface; the control module of the charging equipment parses the received data packets and configures the internal power output parameters; the charging equipment executes the charging operation according to the internal power output parameters and returns a status confirmation signal to the battery management system.

8. The lithium-sulfur battery shuttle effect suppression system as described in claim 7, characterized in that, The communication protocol includes CAN bus protocol, I2C protocol or Modbus protocol.

Images