A method and system for lithium-sulfur battery deactivated sulfur reactivation based on deep learning

Abstract

The application provides a lithium-sulfur battery inactivation sulfur reactivation method and system based on deep learning, and belongs to the technical field of lithium-sulfur batteries. The method comprises the following steps: S1, collecting voltage relaxation data and real-time voltage data of a lithium-sulfur battery and performing pretreatment to obtain pretreated voltage relaxation data and pretreated voltage-capacity differential data; S2, respectively constructing a gated recurrent unit with sulfur diffusion perception gate and a double convolution network with peak value enhancement, extracting sulfur relaxation time sequence features and sulfur reaction local features; then performing bimodal attention fusion to obtain fused sulfur features, and calculating the percentage of sulfidized lithium; S3, respectively extracting inactivation sulfur relaxation features and inactivation sulfur local features; and extracting inactivation sulfur comprehensive features, and calculating the inactivation sulfur proportion; S4, generating an electric pulse activation sequence; and S5, outputting a pulse signal to the electrode interface of the battery. The application can solve the problem that inactivated sulfidized lithium cannot be effectively oxidized after deep discharge of a high-sulfur-loading lithium-sulfur battery, thereby causing battery capacity attenuation.

Description

Technical Field

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

[0002] Lithium-sulfur batteries, as a high-energy-density rechargeable battery system with great development potential, have significantly higher theoretical specific capacity and specific energy than traditional lithium-ion batteries, and have broad application prospects in electric vehicles, energy storage systems, and other fields. However, when lithium-sulfur batteries are at a high sulfur loading, although they can effectively improve the energy density of the battery, they are prone to cathode performance degradation after the battery undergoes deep discharge. Specifically, the insulating lithium sulfide generated in the cathode becomes inactive due to the breakage of the electronic pathway. During the charging process, these inactive lithium sulfides cannot be effectively oxidized, leading to reversible capacity decay of the battery, which restricts the practical application of high-sulfur loading batteries.

[0003] Traditional solutions to the capacity decay problem in lithium-sulfur batteries mainly include: modifying cathode materials, such as using carbon-based composite materials and metal-organic framework materials to improve the dispersibility and conductivity of sulfur; adding functional additives to the electrolyte and developing highly stable electrolyte systems; and using novel separators and constructing three-dimensional electrode structures. These methods have improved the cycle performance of lithium-sulfur batteries to some extent, but their drawbacks are: the compatibility of some modified materials with electrolytes and separators needs to be verified separately, and new byproducts are easily introduced due to interfacial reactions; at the same time, traditional methods are mostly passive prevention of capacity decay and cannot actively repair the battery after the formation of deactivated lithium sulfide, making it difficult to adapt to the dynamic performance requirements of high-load lean electrolyte systems.

[0004] With the development of artificial intelligence technology, the application of deep learning in the field of lithium-sulfur batteries is gradually increasing. Existing research using deep learning technology is mostly focused on battery capacity prediction, lifespan estimation, fault diagnosis, and charging strategy optimization. For example, by building a neural network model and inputting conventional parameters such as battery charging and discharging voltage, current, and temperature, the performance status of the battery can be monitored and predicted. Alternatively, charging curves can be optimized based on historical charging and discharging data to improve charging efficiency and battery life. However, while existing deep learning applications have provided ideas for improving the performance of lithium-sulfur batteries, they are mostly focused on general optimization for conventional scenarios. They lack sufficient specificity and accuracy in addressing the problem of identifying and reactivating deactivated lithium sulfide in the cathode after deep discharge of lithium-sulfur batteries. It is difficult to achieve sensorless identification of deactivated sulfur, and it is also impossible to generate highly adaptable adaptive electrical pulse activation sequences based on the specific situation of deactivated sulfur to promote the electronic reconnection and electrochemical reactivation of lithium sulfide. Summary of the Invention

[0005] To address the problem of capacity decay in lithium-sulfur batteries with high sulfur loading that cannot effectively oxidize deactivated lithium sulfide after deep discharge, the present invention aims to provide a method and system for reactivating deactivated sulfur 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 reactivating deactivated sulfur in lithium-sulfur batteries includes the following steps:

[0008] S1: Collect voltage relaxation data and real-time voltage data of lithium-sulfur batteries and preprocess them to obtain preprocessed voltage relaxation data and preprocessed voltage-capacity differential data.

[0009] S2: Construct a gated recurrent unit with a sulfur diffusion sensing gate based on the preprocessed voltage relaxation data to extract sulfur relaxation time-series features; construct a dual convolutional network with peak enhancement based on the preprocessed voltage-capacity differential data to extract local sulfur reaction features; then perform dual-modal attention fusion to obtain fused sulfur features and calculate the lithium sulfide mass percentage.

[0010] The gated loop unit with sulfur diffusion sensing gate generates candidate hidden states and update gates through the gated loop unit. It calculates the sulfur diffusion sensing gate by combining the candidate hidden states processed by the hyperbolic tangent function with the mean operation result of the voltage relaxation data. Then, it performs Hadamard product operation by combining the sulfur diffusion sensing gate, update gate, and the hidden state at the previous time step to obtain the current hidden state. Finally, it extracts the sulfur relaxation time series features through global average pooling.

[0011] The dual convolutional network with peak enhancement obtains preliminary local reaction features through one-dimensional convolution, and then obtains value-enhanced features after absolute value processing by exponential function and first-order difference operator. It then obtains deep local reaction features through max pooling and one-dimensional convolution, and finally obtains local sulfur reaction features through flattening operation.

[0012] S3: Extract the deactivated sulfur relaxation characteristics based on the preprocessed voltage relaxation data; extract the comprehensive deactivated sulfur characteristics based on the preprocessed voltage-capacity differential data; finally, calculate the deactivated sulfur ratio.

[0013] S4: Based on the lithium sulfide mass percentage and the deactivated sulfur ratio, extract the pulse amplitude characteristics, pulse amplitude sequence, pulse frequency characteristics, and pulse frequency sequence, and generate the electrical pulse activation sequence;

[0014] S5: Based on the electrical pulse activation sequence, output pulse signals to the battery to control the charging process.

[0015] Furthermore, the specific process of step S1 is as follows:

[0016] S11: The voltage relaxation data of the lithium-sulfur battery is collected by the voltage acquisition device during the process of discharging to the relaxation capacity and then recharging to restore the rated voltage. The voltage relaxation data is preprocessed by the moving average method to remove random noise and obtain the preliminary preprocessed voltage relaxation data.

[0017] S12: Collect real-time voltage data of the lithium-sulfur battery during the process of discharging to the relaxation capacity and then charging to a certain threshold SOC through current and voltage acquisition devices; calculate the cumulative charging capacity; and then perform a first-order difference operation on the real-time voltage data and the cumulative charging capacity during the charging process to obtain voltage-capacity differential data.

[0018] S13: For the pre-processed voltage relaxation data and voltage-capacity differential data, outliers are removed using the Raida criterion, and then the data range is standardized by the maximum and minimum values ​​to obtain the pre-processed voltage relaxation data and pre-processed voltage-capacity differential data.

[0019] Furthermore, in step S11, the relaxation capacity is any state of charge (SOC) value within a range, wherein the range is greater than or equal to 10% of the SOC and less than or equal to 20% of the SOC.

[0020] Furthermore, in step S12, a certain threshold SOC is preferably SOC = 60%.

[0021] Furthermore, the specific process of step S2 is as follows:

[0022] S21: Based on the preprocessed voltage relaxation data, a gated recurrent unit with a sulfur diffusion sensing gate is constructed to extract sulfur relaxation time-series features. The calculation method is as follows:

[0023]

[0024]

[0025]

[0026]

[0027] in, Let be the candidate hidden state at time t. Let be the update gate at time t. For gated loop unit, The voltage relaxation data at time t is the preprocessed data. It is a sulfur diffusion sensing gate. The sulfur diffusion sensing coefficient, It is the hyperbolic tangent function. For mean calculation, Let be the hidden state at time t. For Hadama accumulation, This represents the hidden state at time t-1. This represents the characteristics of sulfur relaxation time series. For global average pooling;

[0028] S22: Based on the preprocessed voltage-capacity differential data, a dual convolutional network with peak enhancement is constructed to extract local features of the sulfur reaction. The calculation method is as follows:

[0029]

[0030]

[0031]

[0032]

[0033] in, This is a preliminary local reaction characteristic. For one-dimensional convolution, For the preprocessed voltage-capacity differential data, Peak enhancement feature, It is an exponential function. The peak enhancement factor is... To take the absolute value, It is a first-order difference operator. This is a characteristic of deep, localized reactions. For max pooling, This is a local characteristic of sulfur reaction. For flattening operation;

[0034] S23: Dual-modal attention fusion is performed on the sulfur relaxation time-series characteristics and local sulfur reaction characteristics to obtain the fused sulfur characteristics. The calculation method is as follows:

[0035]

[0036]

[0037]

[0038]

[0039] in, It is a dual-modal interaction feature. Here is the sulfur relaxation weight matrix. This is the sulfur reaction weight matrix. For the two-modal interaction bias vector, This is the dual-modal attention weight vector. For the Softmax function, This is a two-modal interaction weight matrix. For the two-modal interaction bias vector, These are the sulfur relaxation attention weights and the sulfur reaction attention weights, respectively. For vector partitioning functions, Characterized by fused sulfur;

[0040] S24: Based on the characteristics of fused sulfur, the mass percentage of lithium sulfide is calculated using a fully connected network. The calculation method is as follows:

[0041]

[0042] in, This represents the percentage of lithium sulfide by mass. It is a fully connected network.

[0043] It should be further explained that the characteristics and difficulties of this scenario are that the voltage relaxation data of lithium-sulfur batteries contain dynamic time-series information on sulfur diffusion. However, this information is often affected by factors such as battery polarization and noise interference, exhibiting nonlinear and time-varying characteristics. Moreover, the correlation between the sulfur diffusion process and voltage relaxation is hidden. Traditional methods are unable to accurately extract the time-series features directly related to sulfur diffusion from complex data, which limits the accuracy of subsequent calculation of lithium sulfide mass percentage.

[0044] The gated recurrent unit with a sulfur diffusion sensing gate achieves targeted capture of sulfur diffusion-related temporal features. Specifically, the sulfur diffusion sensing gate combines the candidate hidden states processed by the hyperbolic tangent function with the mean calculation result of the voltage relaxation data. The candidate hidden states processed by the hyperbolic tangent function reflect the nonlinear temporal patterns learned by the gated recurrent unit, while the mean calculation result of the voltage relaxation data preserves the statistical characteristics of the data itself. The two are weighted and fused using sulfur diffusion sensing coefficients to highlight key information related to sulfur diffusion and suppress irrelevant noise. Simultaneously, the unit updates the gate with the previous hidden state and the current processing... The candidate hidden states are subjected to Hadamard product operation to dynamically update the hidden states, so that the hidden states can continuously accumulate and strengthen the time-series features related to sulfur diffusion. Finally, the sulfur relaxation time-series features extracted by global average pooling accurately correspond to the dynamic process of sulfur diffusion, providing a reliable basis for the subsequent calculation of the lithium sulfide mass percentage. Traditional gated recurrent units rely solely on their own gating mechanism to learn the time-series features indiscriminately when processing voltage relaxation data. They do not design a directional sensing mechanism for the specific process of sulfur diffusion, which makes the extracted time-series features prone to containing a large amount of redundant information unrelated to sulfur diffusion, or even being interfered with by noise, making it difficult to accurately reflect the dynamic characteristics of sulfur diffusion.

[0045] On the other hand, in the voltage-capacity differential data of lithium-sulfur batteries, the oxidation peak and reduction peak corresponding to the sulfur reaction are the core local features reflecting the sulfur reaction state. However, these peak signals are often masked by interference signals such as data noise and electrolyte decomposition. Moreover, the intensity and position of the peaks are easily changed with the battery cycle state. Traditional methods are difficult to accurately extract and enhance the peak features directly related to the sulfur reaction from the complex differential data, resulting in insufficient effectiveness of the subsequent local features of the sulfur reaction.

[0046] The dual convolutional network with peak enhancement first performs preliminary processing on the preprocessed voltage-capacity differential data using one-dimensional convolution to obtain preliminary local reaction features, which initially filter out some noise and extract local texture information from the data. Next, the rate of change of the preliminary local reaction features is calculated using a first-order difference operator. After taking the absolute value, the difference in the peak region is amplified by an exponential function, and then a Hadamard product operation is performed with the preliminary local reaction features to generate peak enhancement features. In this process, the first-order difference operator is used to locate the peak edges, and the exponential function further amplifies the signal difference between the peak and non-peak regions, thereby directionally enhancing the peak features corresponding to the sulfur reaction and suppressing interference signals in the non-peak regions. Subsequently, the peak enhancement features are downsampled by max pooling to retain key information in the peak region and reduce redundant data. Then, deep local reaction features are extracted by one-dimensional convolution to further optimize the representation ability of the peak features. Finally, the deep local reaction features are transformed into a vector form suitable for subsequent fusion through a flattening operation. The obtained local features of the sulfur reaction can accurately reflect the peak state of the sulfur reaction, providing reliable local feature support for subsequent dual-modal fusion.

[0047] Traditional single-convolutional or double-convolutional networks, when processing voltage-capacity differential data, only perform conventional feature extraction through convolution and pooling, without designing a dedicated enhancement mechanism for peak features. This results in low distinction between peak signals and interference signals in the extracted features, and even weakening of peak features, making it difficult to effectively reflect the sulfur reaction state. In contrast, this invention introduces a peak enhancement stage, specifically designing enhancement logic for the characteristics of sulfur reaction peaks. This actively amplifies peak signals and suppresses interference, resulting in a higher proportion and stronger discriminative power of peak information in the extracted local features of the sulfur reaction, outperforming the feature extraction effect of traditional convolutional networks.

[0048] Furthermore, the specific process of step S3 is as follows:

[0049] S31: Based on the preprocessed voltage relaxation data, extract the deactivated sulfur relaxation characteristics. The calculation method is as follows:

[0050]

[0051]

[0052] in, This represents the difference between the preprocessed voltage relaxation data at time t and time t-1. This is the preprocessed voltage relaxation data at time t-1. This is a characteristic of deactivated sulfur relaxation. Time weighting factor;

[0053] S32: Based on the preprocessed voltage-capacity differential data, extract the local features of deactivated sulfur. The calculation method is as follows:

[0054]

[0055]

[0056]

[0057] in, Here is the channel attention weight vector. For the Sigmoid function, This is the spatial attention weight vector. For batch normalization, This is a localized characteristic of deactivated sulfur;

[0058] S33: Based on the relaxation characteristics of deactivated sulfur, local characteristics of deactivated sulfur, and the mass percentage of lithium sulfide, the comprehensive characteristics of deactivated sulfur are extracted. The calculation method is as follows:

[0059]

[0060]

[0061] in, This is a characteristic of deactivated sulfur cross-linking. For splicing operations, For dimensional expansion operations, The overall characteristics of deactivated sulfur, For ReLU function, This is the comprehensive weight matrix for deactivated sulfur. This is the deactivated sulfur comprehensive bias vector;

[0062] S34: Calculate the proportion of deactivated sulfur based on the overall characteristics of deactivated sulfur. The calculation method is as follows:

[0063]

[0064] in, This represents the proportion of deactivated sulfur.

[0065] It should be further explained that the deactivated sulfur state of lithium-sulfur batteries needs to be judged by a combination of time-series and local features. The percentage of lithium sulfide mass directly affects the total scale of deactivated sulfur. The information is interconnected but also has dimensional differences. If only one type of feature is used, the synergistic relationship between the information is easily ignored, resulting in a one-sided characterization of the deactivated sulfur state. If features are simply spliced ​​together without processing the dimensionality and correlation, feature redundancy or information conflict will occur, making it difficult to form effective features that can accurately support the calculation of the deactivated sulfur ratio.

[0066] In step S3 of this invention, the deactivated sulfur relaxation features (reflecting the relaxation timing anomalies caused by deactivated sulfur), deactivated sulfur local features (reflecting the weakening of reaction peaks caused by deactivated sulfur), and the lithium sulfide mass percentage after dimension expansion (reflecting the total background of deactivated sulfur) are first integrated into a feature vector of a unified dimension through a splicing operation, ensuring that the three types of key information are included in the feature system. Next, the spliced ​​features are processed using one-dimensional convolution to generate deactivated sulfur cross features. Finally, the deactivated sulfur cross features are nonlinearly optimized using the ReLU function to filter out invalid negative feature responses and retain effective feature information. The resulting comprehensive deactivated sulfur feature not only covers information in the three dimensions of timing, local, and total, but also strengthens the synergy between information through cross-correlation, which can comprehensively and accurately reflect the true state of deactivated sulfur and provide high-quality feature input for the subsequent calculation of the deactivated sulfur ratio.

[0067] In step S31, this invention first calculates the difference between the preprocessed voltage relaxation data at time t and time t-1. This difference directly reflects the change in relaxation rate—deactivated sulfur slows down the relaxation rate, resulting in a smaller absolute value of the difference. This operation transforms the original relaxation data into a difference signal that better reflects the "rate anomaly." Secondly, by taking the absolute value of the difference, the direction of rate change is unified, meaning that regardless of whether the rate increases or decreases, its amplitude information is preserved, ensuring that the relaxation rate anomaly caused by deactivated sulfur is not canceled or weakened due to sign issues. Finally, the absolute value is multiplied by an exponential function of the time weighting factor, assigning differentiated weights to the relaxation differences at different times—because the relaxation inertia of deactivated sulfur is more significant in the later stages, the exponential function gradually strengthens the later weights as time increases, highlighting the more obvious deactivation-related difference signals in the later stages and suppressing early interference. After the input processed in this way enters the gated loop unit, it can more efficiently learn the unique relaxation timing pattern of deactivated sulfur, and the correlation between the extracted deactivated sulfur relaxation features and the deactivation state is significantly enhanced.

[0068] Furthermore, the specific process of step S4 is as follows:

[0069] S41: Based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, pulse amplitude features and pulse amplitude sequences are extracted. The calculation method is as follows:

[0070]

[0071]

[0072] in, The characteristics of pulse amplitude, For embedding layer, It is a pulse amplitude sequence. The reference amplitude;

[0073] S42: Based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, pulse frequency features and pulse frequency sequences are extracted. The calculation method is as follows:

[0074]

[0075]

[0076]

[0077] in, It is a simulated sequence of pulse frequencies. For repeated generation operations, To generate the number of times repeatedly, The pulse frequency characteristic For Long Short-Term Memory (LSTM) networks, It is a pulse frequency sequence. For the SoftPlus function, The reference frequency;

[0078] S43: Generate an electrical pulse activation sequence based on pulse amplitude characteristics, pulse frequency characteristics, pulse amplitude sequence, and pulse frequency sequence. The calculation method is as follows:

[0079]

[0080]

[0081]

[0082]

[0083] in, It is a duty cycle sequence. It is a pulse conduction time sequence. It is a pulse interval time series. It is an electrical pulse activation sequence. This is a sequence assembly function.

[0084] It should be further explained that the reactivation of deactivated sulfur in lithium-sulfur batteries depends on a suitable electrical pulse sequence. The amplitude, frequency, and on / off time of the pulses must be precisely matched with the mass percentage of lithium sulfide (total background) and the proportion of deactivated sulfur (stubbornness). If the amplitude is too low, it will not be able to break the insulating layer of deactivated sulfur, and if it is too high, it will easily cause side reactions. An improper frequency will lead to a mismatch in reaction kinetics, resulting in low repair efficiency or aggravated electrode polarization.

[0085] In terms of pulse amplitude design, this invention transforms the lithium sulfide mass percentage and the deactivated sulfur ratio into high-dimensional features through an embedding layer. The joint correlation between the two is extracted through one-dimensional convolution. For example, if the lithium sulfide mass percentage is high and the deactivated sulfur ratio is high, a stronger amplitude is required. Then, by combining the baseline amplitude and the ReLU function, the amplitude is ensured to be non-negative and adapted to the insulation layer breakage threshold of deactivated sulfur. The generated pulse amplitude sequence can dynamically match the stubbornness of deactivated sulfur.

[0086] Secondly, in terms of pulse frequency design, the lithium sulfide mass percentage and deactivated sulfur ratio are converted into a time sequence through repeated generation operations. The sequence is then input into an LSTM to capture the cumulative effect of the two over time. If the content remains high, the frequency needs to be gradually reduced to match the reaction. The reference frequency and the SoftPlus function are then combined to ensure that the frequency is positive and relatively smooth.

[0087] Finally, in the sequence assembly stage, a duty cycle sequence is generated by a fully connected layer and a sigmoid function. This sequence is then combined with the reciprocal of the frequency to obtain the on-time and interval time. High amplitude is paired with a low duty cycle to avoid polarization, while high frequency is paired with a high duty cycle to improve efficiency. Finally, the amplitude, frequency, on-time, and interval time are integrated through an assembly function to generate an electrical pulse activation sequence that achieves multi-parameter synergy and precisely acts on the deactivated sulfur reactivation process.

[0088] The present invention also discloses a deep learning-based lithium-sulfur battery deactivated sulfur reactivation system, including a battery data acquisition module, a lithium sulfide mass percentage calculation module, a deactivated sulfur ratio calculation module, an electrical pulse activation sequence generation module, and a pulse control module;

[0089] The battery data acquisition module collects voltage relaxation data and real-time voltage data of lithium-sulfur batteries, and performs preprocessing to obtain preprocessed voltage relaxation data and preprocessed voltage-capacity differential data.

[0090] The lithium sulfide mass percentage calculation module constructs a gated recurrent unit with a sulfur diffusion sensing gate and a dual convolutional network with peak enhancement based on the preprocessed voltage relaxation data and the preprocessed voltage-capacity differential data, respectively, to extract sulfur relaxation time-series features and sulfur reaction local features; then, it performs dual-modal attention fusion to obtain fused sulfur features and calculates the lithium sulfide mass percentage.

[0091] The deactivated sulfur ratio calculation module extracts the deactivated sulfur relaxation features and local features based on the preprocessed voltage relaxation data and preprocessed voltage-capacity differential data, respectively; and extracts the comprehensive features of deactivated sulfur to calculate the deactivated sulfur ratio.

[0092] The electrical pulse activation sequence generation module extracts pulse amplitude features, pulse amplitude sequences, pulse frequency features, and pulse frequency sequences based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, and generates an electrical pulse activation sequence.

[0093] The pulse control module outputs pulse signals to the battery electrode interface according to the electrical pulse activation sequence.

[0094] Furthermore, the specific working process of the pulse control module is as follows:

[0095] The battery management system reads the real-time status data of the lithium-sulfur battery, including the current open-circuit voltage, cell temperature and historical charge and discharge records, and confirms that the battery has no over-temperature or under-voltage alarms; then the battery management system sends a "permit to apply" command to the pulse control module.

[0096] After receiving the instruction, the pulse control module outputs a pulse signal to the battery electrode interface according to the electrical pulse activation sequence.

[0097] After the electrical pulse activation sequence is completed, the pulse control module sends a "pulse complete" signal to the battery management system. Upon receiving the signal, the battery management system switches to the normal charging mode and controls the charging process.

[0098] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0099] (1) To address the problems of inaccurate identification of deactivated lithium sulfide and poor adaptability of reactivation after deep discharge of lithium-sulfur batteries, this invention first utilizes a gated recurrent unit with a sulfur diffusion sensing gate and a dual convolutional network with peak enhancement to capture the temporal and local features related to the sulfur reaction, and combines dual-modal attention fusion to improve the accuracy of lithium sulfide mass percentage calculation; then, by directionally extracting the relaxation, local and comprehensive features of deactivated sulfur, the deactivated sulfur ratio is identified, avoiding the defect of traditional general optimization that is difficult to focus on the state of deactivated sulfur; finally, based on the lithium sulfide mass percentage and the deactivated sulfur ratio, an appropriate electrical pulse activation sequence is generated to match the electron reconnection requirements of deactivated sulfur, effectively promoting the electrochemical reactivation of deactivated lithium sulfide, avoiding the problem of low activation efficiency of traditional fixed pulse parameters, and preventing battery damage caused by improper parameters, thereby improving the performance stability and capacity recovery effect of lithium-sulfur batteries.

[0100] (2) To address the issues of nonlinear and time-varying voltage relaxation data and hidden sulfur diffusion correlation in lithium-sulfur batteries, this invention designs a gated recurrent unit with a sulfur diffusion sensing gate, which overcomes the limitations of indiscriminate learning in traditional gated recurrent units. The candidate hidden state and the mean of voltage relaxation data are processed by the hyperbolic tangent function and then weighted and fused by the sulfur diffusion sensing coefficient to form a sulfur diffusion sensing gate, highlighting key information about sulfur diffusion and suppressing noise. Then, the current hidden state is dynamically updated by combining the update gate with the hidden state of the previous time step, accumulating and strengthening the sulfur diffusion time sequence features. Finally, accurate sulfur relaxation time sequence features are extracted through global average pooling, providing reliable support for the calculation of lithium sulfide mass percentage and improving the accuracy of feature extraction.

[0101] (3) In view of the problem that the peak value of sulfur reaction is easily masked by interference in the voltage-capacity differential data of lithium-sulfur batteries and is difficult to accurately extract by traditional convolutional networks, this invention proposes a dual convolutional network with peak enhancement. First, one-dimensional convolution is used to filter noise and extract local texture to obtain preliminary local reaction features. Then, the peak edge is located by the first-order difference operator, and the difference between the peak and non-peak regions is amplified by the exponential function. The peak enhancement features are generated by Hadamard product to directionally enhance the peak value of sulfur reaction and suppress interference. Finally, redundancy is reduced by max pooling and deep features are extracted by one-dimensional convolution. After flattening, local features reflecting the state of the peak value of sulfur reaction are obtained, which provides reliable support for dual-modal fusion and solves the defects of low peak discrimination and insufficient feature effectiveness of traditional convolutional networks.

[0102] (4) To address the problem that the deactivated sulfur state of lithium-sulfur batteries requires comprehensive judgment of multi-dimensional features and that traditional methods are prone to one-sided characterization or feature redundancy, this invention constructs a comprehensive feature extraction logic for deactivated sulfur. First, the deactivated sulfur relaxation features, deactivated sulfur local features, and lithium sulfide mass percentage after dimension expansion are concatenated into a unified dimension vector, incorporating three types of key information. Then, cross-features are generated by mining information cross-correlation through one-dimensional convolution, and invalid responses are filtered by ReLU function optimization. Finally, a comprehensive feature covering time series, local, and total information with strong synergy is obtained, which accurately reflects the deactivated sulfur state. In this invention, the data is transformed into a signal that highlights the deactivated-related differences by calculating the voltage relaxation data difference, taking the absolute value, and multiplying it with the time weight exponential function. The relaxation features extracted after inputting into the gated recurrent unit have a stronger correlation with the deactivated state, providing high-quality time series support for the comprehensive feature.

[0103] (5) To address the problem that reactivation of deactivated sulfur in lithium-sulfur batteries requires precise matching of pulse parameters and that traditional fixed parameters are prone to activation failure or damage to the battery, this invention designs a dynamically adapted electrical pulse activation sequence generation logic. In terms of amplitude design, the mass percentage of lithium sulfide and the proportion of deactivated sulfur are jointly correlated through an embedding layer and one-dimensional convolution extraction. Combined with the reference amplitude and the ReLU function, a non-negative amplitude sequence adapted to the stubbornness of deactivated sulfur is generated. In terms of frequency design, the two types of parameters are converted into time series through repeated generation operations. The input is LSTM to capture the cumulative effect. Combined with the reference frequency and the SoftPlus function, a smooth positive frequency sequence is obtained. During sequence assembly, a duty cycle sequence is generated using a fully connected layer and the Sigmoid function. The conduction and interval times are obtained by combining the reciprocal of the frequency. This achieves multi-parameter synergy of high amplitude with low duty cycle to prevent polarization and high frequency with high duty cycle to improve efficiency. Finally, a pulse sequence that is precisely adapted to the reactivation of deactivated sulfur is generated. Detailed Implementation

[0104] 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.

[0105] A deep learning-based method for reactivating deactivated sulfur in lithium-sulfur batteries includes the following steps:

[0106] S1: Collect voltage relaxation data and real-time voltage data from lithium-sulfur batteries and preprocess them to obtain preprocessed voltage relaxation data and preprocessed voltage-capacity differential data, including:

[0107] S11: Using a voltage acquisition device, collect voltage relaxation data of the lithium-sulfur battery during the process of discharging to the relaxation capacity and then recharging until the voltage recovers to the rated voltage. Use a moving average method to remove random noise to obtain pre-processed voltage relaxation data. The relaxation capacity is any SOC value within a range greater than or equal to 10% of the SOC and less than or equal to 20% of the SOC.

[0108] S12: Collect real-time voltage data of the lithium-sulfur battery during the process of discharging to the relaxation capacity and then charging to SOC=60% using a current and voltage acquisition device; calculate the cumulative charging capacity; and then perform a first-order difference operation on the real-time voltage data and cumulative charging capacity during the charging process to obtain voltage-capacity differential data.

[0109] The cumulative charging capacity is the sum of the amount of electricity received by the battery during all time periods from the "start time" of the charging process (after the discharge ends) to the "current charging time," representing the "total amount of electricity already charged" by the battery during this charging phase.

[0110] S13: For the pre-processed voltage relaxation data and voltage-capacity differential data, outliers are removed using the Raida criterion, and then the data range is standardized by the maximum and minimum values ​​to obtain the pre-processed voltage relaxation data and pre-processed voltage-capacity differential data.

[0111] S2: Based on the preprocessed voltage relaxation data and the preprocessed voltage-capacity differential data, a gated recurrent unit with a sulfur diffusion sensing gate and a dual convolutional network with peak enhancement are constructed to extract sulfur relaxation time-series features and sulfur reaction local features. Then, dual-modal attention fusion is performed to obtain fused sulfur features, and the lithium sulfide mass percentage is calculated, including:

[0112] S21: Based on the preprocessed voltage relaxation data, a gated recurrent unit with a sulfur diffusion sensing gate is constructed to extract sulfur relaxation time-series features. The calculation method is as follows:

[0113]

[0114]

[0115]

[0116]

[0117] in, Let be the candidate hidden state at time t. Let be the update gate at time t. For gated loop unit, The voltage relaxation data at time t is the preprocessed data. It is a sulfur diffusion sensing gate. The sulfur diffusion sensing coefficient, It is the hyperbolic tangent function. For mean calculation, Let be the hidden state at time t, where t is the time index. For Hadama accumulation, This represents the hidden state at time t-1. This represents the characteristics of sulfur relaxation time series. For global average pooling;

[0118] S22: Based on the preprocessed voltage-capacity differential data, a dual convolutional network with peak enhancement is constructed to extract local features of the sulfur reaction. The calculation method is as follows:

[0119]

[0120]

[0121]

[0122]

[0123] in, This is a preliminary local reaction characteristic. For one-dimensional convolution, For the preprocessed voltage-capacity differential data, Peak enhancement feature, It is an exponential function. The peak enhancement factor is... To take the absolute value, It is a first-order difference operator. This is a characteristic of deep, localized reactions. For max pooling, This is a local characteristic of sulfur reaction. For flattening operation;

[0124] S23: Dual-modal attention fusion is performed on the sulfur relaxation time-series characteristics and local sulfur reaction characteristics to obtain the fused sulfur characteristics. The calculation method is as follows:

[0125]

[0126]

[0127]

[0128]

[0129] in, It is a dual-modal interaction feature. Here is the sulfur relaxation weight matrix. This is the sulfur reaction weight matrix. For the two-modal interaction bias vector, This is the dual-modal attention weight vector. For the Softmax function, This is a two-modal interaction weight matrix. These are the sulfur relaxation attention weights and the sulfur reaction attention weights, respectively. For vector partitioning functions, To incorporate sulfur characteristics, This is a two-modal interaction bias vector;

[0130] S24: Based on the characteristics of fused sulfur, the mass percentage of lithium sulfide is calculated using a fully connected network. The calculation method is as follows:

[0131]

[0132] in, This represents the percentage of lithium sulfide by mass. It is a fully connected network.

[0133] S3: Based on the preprocessed voltage relaxation data and preprocessed voltage-capacity differential data, extract the deactivated sulfur relaxation features and local features of deactivated sulfur, respectively; and extract the comprehensive features of deactivated sulfur, calculate the proportion of deactivated sulfur, including:

[0134] S31: Based on the preprocessed voltage relaxation data, extract the deactivated sulfur relaxation characteristics. The calculation method is as follows:

[0135]

[0136]

[0137] in, This represents the difference between the preprocessed voltage relaxation data at time t and time t-1. This is the preprocessed voltage relaxation data at time t-1. This is a characteristic of deactivated sulfur relaxation. Time weighting factor;

[0138] S32: Based on the preprocessed voltage-capacity differential data, extract the local features of deactivated sulfur. The calculation method is as follows:

[0139]

[0140]

[0141]

[0142] in, Here is the channel attention weight vector. For the Sigmoid function, This is the spatial attention weight vector. For batch normalization, This is a localized characteristic of deactivated sulfur;

[0143] S33: Based on the relaxation characteristics of deactivated sulfur, local characteristics of deactivated sulfur, and the mass percentage of lithium sulfide, the comprehensive characteristics of deactivated sulfur are extracted. The calculation method is as follows:

[0144]

[0145]

[0146] in, This is a characteristic of deactivated sulfur cross-linking. For splicing operations, For dimensional expansion operations, The overall characteristics of deactivated sulfur, For ReLU function, This is the comprehensive weight matrix for deactivated sulfur. This is the deactivated sulfur comprehensive bias vector;

[0147] S34: Calculate the proportion of deactivated sulfur based on the overall characteristics of deactivated sulfur. The calculation method is as follows:

[0148]

[0149] in, This represents the proportion of deactivated sulfur.

[0150] S4: Based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, extract pulse amplitude features, pulse amplitude sequences, pulse frequency features, and pulse frequency sequences, and generate an electrical pulse activation sequence, including:

[0151] S41: Based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, pulse amplitude features and pulse amplitude sequences are extracted. The calculation method is as follows:

[0152]

[0153]

[0154] in, The characteristics of pulse amplitude, For embedding layer, It is a pulse amplitude sequence. The baseline amplitude is set according to requirements;

[0155] S42: Based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, pulse frequency features and pulse frequency sequences are extracted. The calculation method is as follows:

[0156]

[0157]

[0158]

[0159] in, It is a simulated sequence of pulse frequencies. For repeated generation operations, To generate the number of times repeatedly, The pulse frequency characteristic For Long Short-Term Memory (LSTM) networks, It is a pulse frequency sequence. For the SoftPlus function, The reference frequency;

[0160] S43: Generate an electrical pulse activation sequence based on pulse amplitude characteristics, pulse frequency characteristics, pulse amplitude sequence, and pulse frequency sequence. The calculation method is as follows:

[0161]

[0162]

[0163]

[0164]

[0165] in, It is a duty cycle sequence. It is a pulse conduction time sequence. It is a pulse interval time series. It is an electrical pulse activation sequence. For sequence assembly functions;

[0166] S5: Output pulse signals to the battery according to the electrical pulse activation sequence.

[0167] Example 1

[0168] A deep learning-based lithium-sulfur battery deactivated sulfur reactivation system includes:

[0169] Battery data acquisition module: Acquires voltage relaxation data and real-time voltage data of lithium-sulfur batteries and performs preprocessing to obtain preprocessed voltage relaxation data and preprocessed voltage-capacity differential data.

[0170] The lithium sulfide mass percentage calculation module: Based on the preprocessed voltage relaxation data and the preprocessed voltage-capacity differential data, a gated recurrent unit with a sulfur diffusion sensing gate and a dual convolutional network with peak enhancement are constructed to extract sulfur relaxation time-series features and sulfur reaction local features; then, dual-modal attention fusion is performed to obtain fused sulfur features and calculate the lithium sulfide mass percentage.

[0171] Deactivated sulfur ratio calculation module: Based on the preprocessed voltage relaxation data and the preprocessed voltage-capacity differential data, extract the deactivated sulfur relaxation features and local features of deactivated sulfur respectively; and extract the comprehensive features of deactivated sulfur to calculate the deactivated sulfur ratio.

[0172] Electrical pulse activation sequence generation module: Based on the lithium sulfide mass percentage and the deactivated sulfur ratio, extract pulse amplitude features, pulse amplitude sequence, pulse frequency features, and pulse frequency sequence, and generate an electrical pulse activation sequence;

[0173] Pulse control module: Based on the electrical pulse activation sequence, it outputs pulse signals to the battery electrode interface. The specific process is as follows:

[0174] The battery management system reads the real-time status data of the lithium-sulfur battery, including the current open-circuit voltage, cell temperature and historical charge and discharge records, and confirms that the battery has no over-temperature or under-voltage alarms; then the battery management system sends a "permit to apply" command to the pulse control module.

[0175] After receiving the instruction, the pulse control module outputs a pulse signal to the battery electrode interface according to the electrical pulse activation sequence.

[0176] After the electrical pulse activation sequence is completed, the pulse control module sends a "pulse complete" signal to the battery management system. Upon receiving the signal, the battery management system switches to the normal charging mode and controls the charging process.

[0177] Scenario 1: Low lithium sulfide mass percentage (L=20%) + low deactivated sulfur ratio (rate=10%). In this case, deactivated sulfur is easily activated, and the duty cycle fluctuates slightly to balance efficiency.

[0178] Pulse amplitude sequence (Unit: V);

[0179] Pulse frequency sequence (Unit: Hz);

[0180] Duty cycle sequence The amount was slightly reduced as deactivated sulfur decreased to avoid excess energy.

[0181] On-time series (Unit: ms);

[0182] Interval Time Series (Unit: ms);

[0183] Scenario 2: Medium lithium sulfide mass percentage (L=50%) + medium deactivated sulfur ratio (rate=30%). In this case, the activation difficulty of deactivated sulfur is moderate, and the duty cycle is gradually adjusted.

[0184] Pulse amplitude sequence [1.2, 1.19, 1.18, 1.17, 1.15] (unit: V);

[0185] Pulse frequency sequence [15,14.8,14.6,14.4,14] (unit: Hz);

[0186] Duty cycle sequence [0.42, 0.41, 0.4, 0.39, 0.38], each step is reduced by 0.01 to match the gradual decrease in deactivated sulfur;

[0187] On-time series [28,27.7,27.4,27.1,27.1] (unit: ms);

[0188] Interval Time Series [38.7,39.3,40,40.7,43.8] (unit: ms);

[0189] Scenario 3: High lithium sulfide mass percentage (L=80%) + high deactivated sulfur ratio (rate=60%). In this case, the deactivated sulfur is stubborn, the duty cycle is stable, and side reactions are prevented.

[0190] Pulse amplitude sequence [1.5, 1.48, 1.45, 1.43, 1.4] (unit: V);

[0191] Pulse frequency sequence [5,4.8,4.6,4.4,4.2] (unit: Hz);

[0192] Duty cycle sequence [0.3,0.3,0.3,0.3,0.3], constant throughout, ensuring stable energy output of high-amplitude pulses and avoiding electrolyte decomposition caused by duty cycle fluctuations;

[0193] On-time series [60, 62.5, 65.2, 68.2, 71.4] (unit: ms);

[0194] Interval Time Series [140,145.8,151.5,158.5,166.3] (unit: ms).

[0195] 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.

[0196] 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.

Claims

1. A deep learning-based method for reactivating deactivated sulfur in lithium-sulfur batteries, characterized in that, Includes the following steps: S1: Collect voltage relaxation data and real-time voltage data of lithium-sulfur batteries and preprocess them to obtain preprocessed voltage relaxation data and preprocessed voltage-capacity differential data. S2: Construct a gated recurrent unit with a sulfur diffusion sensing gate based on the preprocessed voltage relaxation data to extract sulfur relaxation time-series features; construct a double convolutional network with peak enhancement based on the preprocessed voltage-capacity differential data to extract local sulfur reaction features. Then, dual-modal attention fusion is performed to obtain the fused sulfur features, and the mass percentage of lithium sulfide is calculated. The gated loop unit with sulfur diffusion sensing gate generates candidate hidden states and update gates through the gated loop unit. It calculates the sulfur diffusion sensing gate by combining the candidate hidden states processed by the hyperbolic tangent function with the mean operation result of the voltage relaxation data. Then, it performs Hadamard product operation by combining the sulfur diffusion sensing gate, update gate, and the hidden state at the previous time step to obtain the current hidden state. Finally, it extracts the sulfur relaxation time series features through global average pooling. The dual convolutional network with peak enhancement obtains preliminary local reaction features through one-dimensional convolution, and then obtains value-enhanced features after absolute value processing by exponential function and first-order difference operator. It then obtains deep local reaction features through max pooling and one-dimensional convolution, and finally obtains local sulfur reaction features through flattening operation. S3: Extract the deactivated sulfur relaxation characteristics based on the preprocessed voltage relaxation data; extract the comprehensive deactivated sulfur characteristics based on the preprocessed voltage-capacity differential data; finally, calculate the deactivated sulfur ratio. S4: Based on the lithium sulfide mass percentage and the deactivated sulfur ratio, extract the pulse amplitude characteristics, pulse amplitude sequence, pulse frequency characteristics, and pulse frequency sequence, and generate the electrical pulse activation sequence; S5: Based on the electrical pulse activation sequence, output pulse signals to the battery to control the charging process.

2. The method for reactivating deactivated sulfur in lithium-sulfur batteries based on deep learning as described in claim 1, characterized in that, The specific process of step S1 is as follows: S11: The voltage relaxation data of the lithium-sulfur battery is collected by the voltage acquisition device during the process of discharging to the relaxation capacity and then recharging to restore the rated voltage. The voltage relaxation data is preprocessed by the moving average method to remove random noise and obtain the preliminary preprocessed voltage relaxation data. S12: Collect real-time voltage data of lithium-sulfur batteries during the process of discharging to the relaxation capacity and then charging to a certain threshold SOC through current and voltage acquisition devices. Calculate the cumulative charging capacity, and then perform a first-order difference operation on the real-time voltage data and cumulative charging capacity during the charging process to obtain voltage-capacity differential data. S13: For the pre-processed voltage relaxation data and voltage-capacity differential data, outliers are removed using the Raida criterion, and then the data range is standardized by the maximum and minimum values ​​to obtain the pre-processed voltage relaxation data and pre-processed voltage-capacity differential data.

3. The deep learning-based method for reactivating deactivated sulfur in lithium-sulfur batteries as described in claim 2, characterized in that, In step S11, the relaxation capacity is any state of charge (SOC) value within the range of values, which is greater than or equal to 10% of the SOC and less than or equal to 20% of the SOC.

4. The method for reactivating deactivated sulfur in lithium-sulfur batteries based on deep learning as described in claim 2, characterized in that, In step S12, a certain threshold SOC is 0.6SOC.

5. The method for reactivating deactivated sulfur in lithium-sulfur batteries based on deep learning as described in claim 2, characterized in that, The specific process of step S2 is as follows: S21: Based on the preprocessed voltage relaxation data, a gated recurrent unit with a sulfur diffusion sensing gate is constructed to extract sulfur relaxation time-series features. The calculation method is as follows: ， in, Let be the candidate hidden state at time t. Let be the update gate at time t. For gated loop unit, The voltage relaxation data at time t is the preprocessed data. It is a sulfur diffusion sensing gate. The sulfur diffusion sensing coefficient, It is the hyperbolic tangent function. For mean calculation, Let be the hidden state at time t. For Hadama accumulation, This represents the hidden state at time t-1. This represents the characteristics of sulfur relaxation time series. For global average pooling; S22: Based on the preprocessed voltage-capacity differential data, a dual convolutional network with peak enhancement is constructed to extract local features of the sulfur reaction. The calculation method is as follows: ， in, This is a preliminary local reaction characteristic. For one-dimensional convolution, For the preprocessed voltage-capacity differential data, Peak enhancement feature, It is an exponential function. The peak enhancement factor is... To take the absolute value, It is a first-order difference operator. This is a characteristic of deep, localized reactions. For max pooling, This is a local characteristic of sulfur reaction. For flattening operation; S23: Dual-modal attention fusion is performed on the sulfur relaxation time-series characteristics and local sulfur reaction characteristics to obtain the fused sulfur characteristics. The calculation method is as follows: ， in, It is a dual-modal interaction feature. Here is the sulfur relaxation weight matrix. This is the sulfur reaction weight matrix. For the two-modal interaction bias vector, This is the dual-modal attention weight vector. For the Softmax function, This is a two-modal interaction weight matrix. For the two-modal interaction bias vector, These are the sulfur relaxation attention weights and the sulfur reaction attention weights, respectively. For vector partitioning functions, Characterized by fused sulfur; S24: Based on the characteristics of fused sulfur, the mass percentage of lithium sulfide is calculated using a fully connected network. The calculation method is as follows: ， in, This represents the percentage of lithium sulfide by mass. It is a fully connected network.

6. The deep learning-based method for reactivating deactivated sulfur in lithium-sulfur batteries as described in claim 5, characterized in that, The specific process of step S3 is as follows: S31: Based on the preprocessed voltage relaxation data, extract the deactivated sulfur relaxation characteristics. The calculation method is as follows: ， in, This represents the difference between the preprocessed voltage relaxation data at time t and time t-1. This is the preprocessed voltage relaxation data at time t-1. This is a characteristic of deactivated sulfur relaxation. Time weighting factor; S32: Based on the preprocessed voltage-capacity differential data, extract the local features of deactivated sulfur. The calculation method is as follows: ， in, Here is the channel attention weight vector. For the Sigmoid function, This is the spatial attention weight vector. For batch normalization, This is a localized characteristic of deactivated sulfur; S33: Based on the relaxation characteristics of deactivated sulfur, local characteristics of deactivated sulfur, and the mass percentage of lithium sulfide, the comprehensive characteristics of deactivated sulfur are extracted. The calculation method is as follows: ， in, This is a characteristic of deactivated sulfur cross-linking. For splicing operations, For dimensional expansion operations, The overall characteristics of deactivated sulfur, For ReLU function, This is the comprehensive weight matrix for deactivated sulfur. This is the deactivated sulfur comprehensive bias vector; S34: Calculate the proportion of deactivated sulfur based on the overall characteristics of deactivated sulfur. The calculation method is as follows: ， in, This represents the proportion of deactivated sulfur.

7. The deep learning-based method for reactivating deactivated sulfur in lithium-sulfur batteries as described in claim 6, characterized in that, The specific process of step S4 is as follows: S41: Based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, pulse amplitude features and pulse amplitude sequences are extracted. The calculation method is as follows: ， in, The characteristics of pulse amplitude, For embedding layer, It is a pulse amplitude sequence. The reference amplitude; S42: Based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, pulse frequency features and pulse frequency sequences are extracted. The calculation method is as follows: ， in, It is a simulated sequence of pulse frequencies. For repeated generation operations, To generate the number of times repeatedly, The pulse frequency characteristic For Long Short-Term Memory (LSTM) networks, It is a pulse frequency sequence. For the SoftPlus function, The reference frequency; S43: Generate an electrical pulse activation sequence based on pulse amplitude characteristics, pulse frequency characteristics, pulse amplitude sequence, and pulse frequency sequence. The calculation method is as follows: ， in, It is a duty cycle sequence. It is a pulse conduction time sequence. It is a pulse interval time series. It is an electrical pulse activation sequence. This is a sequence assembly function.

8. A deep learning-based lithium-sulfur battery deactivated sulfur reactivation system, characterized in that, It includes a battery data acquisition module, a lithium sulfide mass percentage calculation module, a deactivated sulfur ratio calculation module, an electrical pulse activation sequence generation module, and a pulse control module; The battery data acquisition module collects voltage relaxation data and real-time voltage data of lithium-sulfur batteries, and performs preprocessing to obtain preprocessed voltage relaxation data and preprocessed voltage-capacity differential data. The lithium sulfide mass percentage calculation module constructs a gated recurrent unit with a sulfur diffusion sensing gate and a dual convolutional network with peak enhancement based on the preprocessed voltage relaxation data and the preprocessed voltage-capacity differential data, respectively, to extract sulfur relaxation time-series features and local sulfur reaction features. Then, dual-modal attention fusion is performed to obtain the fused sulfur features, and the mass percentage of lithium sulfide is calculated. The deactivated sulfur ratio calculation module extracts the deactivated sulfur relaxation features and local features based on the preprocessed voltage relaxation data and preprocessed voltage-capacity differential data, respectively. The comprehensive characteristics of deactivated sulfur were extracted, and the proportion of deactivated sulfur was calculated. The electrical pulse activation sequence generation module extracts pulse amplitude features, pulse amplitude sequences, pulse frequency features, and pulse frequency sequences based on the lithium sulfide mass percentage and the proportion of deactivated sulfur, and generates an electrical pulse activation sequence. The pulse control module outputs pulse signals to the battery electrode interface according to the electrical pulse activation sequence.

9. The deep learning-based lithium-sulfur battery deactivated sulfur reactivation system as described in claim 8, characterized in that, The specific working process of the pulse control module is as follows: The battery management system reads the real-time status data of the lithium-sulfur battery, including the current open-circuit voltage, cell temperature, and historical charge and discharge records, and confirms that the battery has no over-temperature or under-voltage alarms; then the battery management system sends an application permission command to the pulse control module. After receiving the instruction, the pulse control module outputs a pulse signal to the battery electrode interface according to the electrical pulse activation sequence. After the electrical pulse activation sequence is completed, the pulse control module sends a pulse completion signal to the battery management system. After receiving the signal, the battery management system switches to the normal charging mode and controls the charging process.

Images