Pulse charging curve control method and system for vehicle energy storage battery

Abstract

The present application belongs to the technical field of pulse charging curve control, and particularly relates to a pulse charging curve control method and system for vehicle energy storage batteries, a multi-dimensional perception network is constructed, temperature, voltage and electrochemical parameters are synchronously collected, and a three-domain data mapping relationship is established; a dynamic and electrically coupled optimization model is established, a temperature influence quantitative equation and deep learning are integrated, and adaptive learning of aging characteristics is completed; a thermal and electric collaborative closed-loop control algorithm is designed, a double safety threshold is used as a constraint, pulse parameters are optimized through model prediction, and a closed-loop execution is formed; a multi-modal safety boundary identification and early warning system is constructed, multi-feature extraction and sensing technology are integrated, safety risk early warning and parameter degradation control are completed; and a transition mechanism is optimized in the stage, a buffer period is set, a parameter gradual change strategy and impedance compensation are used, and voltage mutation in the switching between the pulse and constant current / constant voltage stages is inhibited.

Description

Technical Field

[0001] This invention belongs to the field of pulse charging curve control technology, and particularly relates to a pulse charging curve control method and system for automotive energy storage batteries. Background Technology

[0002] Traditional pulse charging technology for automotive energy storage batteries relies solely on single voltage and temperature sensors for parameter acquisition. This limited sensing dimension prevents the simultaneous acquisition of coordinated data across the thermal, electrical, and electrochemical domains, hindering the construction of multi-parameter correlation mapping relationships. Dynamic thermo-electric coupling models often employ fixed mechanistic parameters, failing to integrate temperature-influence quantification equations and deep learning algorithms. Consequently, they lack adaptive learning capabilities for battery aging characteristics and cannot adapt to the characteristic drift throughout the battery's entire lifecycle. Charging control is typically in open-loop or simplified closed-loop mode, constrained by fixed safety thresholds. It fails to achieve dynamic optimization of pulse parameters through model prediction, resulting in low precision in thermo-electric coordinated control and an inability to balance charging speed, energy efficiency, and battery lifespan.

[0003] The lack of a smooth transition mechanism when switching from pulse charging to constant current / constant voltage stages easily leads to voltage surges and oscillations, reducing charging stability and exacerbating damage to the internal battery structure. Safety risk warnings rely solely on single-mode sensor signals for judgment, without integrating multi-feature extraction and multi-dimensional boundary recognition technologies, resulting in low risk identification accuracy, high false alarm rates, and poor linkage between parameter degradation control and risk levels. Furthermore, the transmission, storage, and preprocessing of sensing data are not adapted to the resource limitations of the vehicle system, and the lack of online model updates and adaptive optimization mechanisms makes it difficult to simultaneously ensure safety protection, control response, and battery durability under complex conditions such as extreme temperatures and different aging stages. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a pulse charging curve control method for automotive energy storage batteries, the method comprising:

[0005] Construct a multi-dimensional sensing network to simultaneously collect temperature, voltage, and electrochemical parameters, and establish a three-domain data mapping relationship;

[0006] A dynamic, electrically coupled optimization model was established, incorporating temperature effect quantification equations and deep learning to achieve adaptive learning of aging characteristics.

[0007] Design a thermal and electrical coordinated closed-loop control algorithm, using dual safety thresholds as constraints, and optimize pulse parameters through model prediction to form a closed-loop execution;

[0008] Construct a multimodal security boundary identification and early warning system, integrating multi-feature extraction and sensing technologies to complete security risk early warning and parameter degradation control;

[0009] The phase transition mechanism is optimized by using buffer period settings, parameter gradual change strategies, and impedance compensation to suppress voltage spikes during the switching between pulse and constant current / constant voltage phases.

[0010] Furthermore, the present invention also provides a pulse charging curve control system for automotive energy storage batteries, comprising:

[0011] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described pulse charging curve control method for an automotive energy storage battery by executing the machine-executable instructions.

[0012] In another aspect, the present invention also provides a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the above-described pulse charging curve control method for automotive energy storage batteries.

[0013] Based on the above, by using multi-dimensional parameter collaborative sensing and thermal-electrical-aging coupled dynamic modeling, the limitations of traditional charging technology, such as single sensing dimension and poor model adaptability, are overcome. This enables precise control and smooth switching of charging parameters under all working conditions, effectively suppressing voltage oscillations and internal battery damage during stage switching. While improving charging speed, it also significantly improves battery charging consistency and cycle life.

[0014] This invention relies on a risk warning and closed-loop control strategy based on multi-feature fusion to achieve accurate identification and graded response to charging safety risks, significantly reducing false alarm rate and failure rate. At the same time, it optimizes the data processing and lightweight operation logic of the vehicle terminal, improves the system resource utilization efficiency, and enables the charging system to operate stably under extreme temperatures and different aging conditions, comprehensively taking into account charging efficiency, energy utilization rate and safety of use. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the execution flow of the pulse charging curve control method for vehicle energy storage batteries provided in an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of exemplary hardware and software components of the pulse charging curve control system for automotive energy storage batteries provided in an embodiment of the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1This is a flowchart illustrating a pulse charging curve control method for a vehicle energy storage battery according to an embodiment of the present invention. The pulse charging curve control method for the vehicle energy storage battery will be described in detail below.

[0018] Step S110: Construct a multi-dimensional sensing network, synchronously collect temperature, voltage and electrochemical parameters, and establish a three-domain data mapping relationship;

[0019] A multi-dimensional sensing network is constructed by integrating multiple types of sensing modules and data processing units. Temperature parameters are acquired through micro-nano-scale temperature sensing elements, voltage parameters are obtained through flexible voltage sensors, and electrochemical parameters are obtained through broadband scanning analysis using a high-frequency impedance spectroscopy module. The acquisition process of the three types of parameters is time-stamp aligned through a unified synchronous trigger signal to ensure data timing consistency. When constructing the three-domain data mapping relationship, the intrinsic relationship between the parameters in the thermal domain (temperature, temperature rise rate), electrical domain (voltage, current), and electrochemical domain (polarization resistance, double-layer capacitance) is first clarified based on the battery electrochemical mechanism. Then, sample data is collected through multi-condition experiments, and a random forest algorithm is used to train the mapping model. This model takes real-time data from the thermal and electrical domains as input and outputs the corresponding electrochemical parameters. The model supports online updates to ensure that the mapping relationship dynamically adapts to the battery state.

[0020] Step S111: Based on the structural characteristics of the battery cell, select micro-nano-level temperature sensing elements, flexible voltage sensors and high-frequency impedance spectrum modules. Through customized adaptation design, take into account the environmental adaptability, response performance and vehicle system compatibility of the devices to meet the requirements for accurate acquisition of multi-dimensional parameters.

[0021] When selecting micro-nano-scale temperature sensing elements, ceramic-encapsulated platinum resistance elements are preferred due to their resistance to electrolyte corrosion and high-temperature environmental adaptability, making them suitable for the internal working environment of batteries. The flexible voltage sensor uses a polyimide flexible substrate and integrates a low-power operational amplifier. Its response speed meets the acquisition requirements for pulse charging voltage fluctuations. Simultaneously, the contact interface between the sensor and the electrode is optimized, employing a conductive adhesive bonding design to reduce contact resistance. The high-frequency impedance spectroscopy module uses a wide-bandwidth integrated chip, with a frequency coverage suitable for electrochemical parameter extraction requirements. The module size is customized according to the battery cell installation space, and the power interface is matched to the vehicle's 12V power system. Electromagnetic compatibility design ensures no interference with other vehicle electronic devices. The overall component selection considers acquisition accuracy, environmental adaptability, and compatibility with the vehicle system.

[0022] Step S112: Deploy various sensors and modules according to the location of the battery cells, and take into account both installation reliability and structural integrity through minimally invasive implantation, bonding and fixing, interface connection and electromagnetic shielding design.

[0023] During the battery cell production stage, installation channels are pre-designed. The channel positions at the base of the tabs, on the surfaces of the positive and negative electrodes, and near the separator avoid the active areas of the electrodes, ensuring that the battery's charge and discharge performance is not affected. Temperature sensing elements are embedded into the channels using minimally invasive implantation tools and fixed with high-temperature resistant adhesives, ensuring a tight fit between the element and the electrodes and separator without damaging the battery's sealing structure. Flexible voltage sensors are directly attached to the connection points between the tabs and electrodes, and the lead terminals are fixed using laser welding. The welding power is adapted to the lead material and tab thickness to reduce signal transmission loss. The high-frequency impedance spectroscopy module connects to the positive and negative terminals of the battery cell via a quick-connect dedicated interface. The module housing is made of aluminum alloy and has a built-in electromagnetic shielding layer, which is grounded to effectively suppress electromagnetic interference from the vehicle environment.

[0024] Step S113: Build a multi-channel synchronous acquisition system with a high-performance control unit as the core, configure signal conditioning circuit, and ensure the synchronization and stability of the three-domain parameter acquisition through a unified synchronization mechanism and adapted sampling frequency;

[0025] A high-performance FPGA is selected as the core control unit, possessing multi-channel parallel processing capabilities, capable of simultaneously driving temperature, voltage sensors, and the impedance spectrum module. Independent signal conditioning circuits are configured for each of the three types of sensors. The filtering circuit uses a second-order low-pass filter, the amplification circuit uses an instrumentation amplifier, and the isolation circuit employs opto-isolation design to effectively eliminate environmental noise and electromagnetic interference. A synchronous triggering mechanism generates a 50kHz synchronization pulse signal through the FPGA's internal timer, controlling the three types of devices to start data acquisition at the same timestamp. The sampling frequency is set according to parameter characteristics: the voltage sampling frequency adapts to the pulse charging voltage fluctuation characteristics, the temperature sampling frequency balances response speed and data volume, and the impedance spectrum module completes wideband scanning at a preset period, ensuring the synchronization and stability of parameter acquisition across the three domains.

[0026] Step S114: Perform real-time preprocessing on the collected raw data, use digital filtering algorithms to remove high-frequency noise in the voltage signal and random fluctuations in the temperature signal, perform phase correction and amplitude calibration on the impedance spectrum scanning data, establish a sensor calibration system, perform multi-point temperature calibration on the temperature sensor through the calibration platform, perform amplitude calibration on the voltage sensor under different loads, perform standard impedance calibration on the impedance spectrum module, store calibration coefficients for real-time data correction, design data validity judgment logic, and remove abnormal data through threshold judgment and trend analysis;

[0027] In the data preprocessing stage, the voltage signal is filtered using a Kalman filter to remove high-frequency noise, the temperature signal is processed using a moving average algorithm to eliminate random fluctuations, and the impedance spectrum scan data is calibrated using the least squares method for phase correction and amplitude calibration to eliminate errors caused by the impedance of the test system itself. A dedicated calibration platform is built, including a high and low temperature chamber, a standard resistance chamber, and a precision DC power supply. The temperature sensor is calibrated at multiple points at different temperatures, and the calibration coefficients are recorded. The voltage sensor is calibrated under different load currents to ensure voltage acquisition accuracy. The impedance spectrum module is calibrated across the entire frequency range using standard impedance components. A data validity judgment logic is designed to identify abnormal data through parameter threshold range judgment and time-series trend consistency analysis. Data exceeding reasonable ranges or exhibiting abrupt trend changes are marked as invalid and discarded to ensure the reliability of the data input to the mapping model.

[0028] Step S115: Define the characteristic parameters of the thermal domain, electric domain, and electrochemical domain. Analyze the intrinsic correlation between these parameters based on the battery electrochemical mechanism. Collect sample data through multi-condition experiments, simultaneously recording the three types of parameter data under different temperature ranges, SOC ranges, and aging stages to construct a sample database. Use machine learning algorithms to build a mapping model, taking thermal and electric domain parameters as inputs and electrochemical parameters as outputs. Train and optimize the model using sample data. Design an online model update mechanism to monitor model prediction errors in real time. When the error exceeds a preset threshold, automatically call the latest collected valid data to fine-tune the model.

[0029] The core characteristic parameters of the thermal domain are defined as temperature and temperature rise rate; those of the electrical domain are voltage, current, and polarization voltage; and those of the electrochemical domain are polarization resistance, double-layer capacitance, and characteristic frequency. The correlation between these parameters is analyzed based on battery double-layer theory and charge transfer mechanisms. Multi-condition experiments cover different temperature ranges, SOC ranges, and aging stages, simultaneously recording data for all three types of parameters. The experimental process uses a charge / discharge cabinet to simulate pulse charging scenarios, and the data is stored via a data acquisition card, constructing a sample database covering the entire battery lifecycle. The mapping model is built using a BP neural network algorithm, with thermal and electrical domain parameters as the input layer and electrochemical parameters as the output layer. The model is trained using the sample database, optimizing the number of network layers and neurons. The online model update mechanism calculates the deviation between predicted and actual collected values ​​in real time. When the deviation exceeds a preset threshold, the latest valid data is automatically used to fine-tune the model weights, ensuring that the mapping relationship dynamically adapts to changes in battery state.

[0030] Step S116: Employ a high-speed anti-interference transmission interface, a hierarchical storage strategy, and an optimized compression algorithm to achieve efficient transmission and reasonable storage of sensing data, adapting to the resource limitations of the vehicle system.

[0031] Data transmission utilizes the CANFD high-speed anti-interference bus interface, which features high transmission speed and strong electromagnetic interference resistance, making it suitable for automotive environments. In the hierarchical storage system, high-reliability NAND flash memory is used locally to cache recently acquired raw data, preprocessed results, and model output data, with the caching period set according to the vehicle's storage capacity. Historical data throughout the entire lifecycle is uploaded to a cloud-based distributed database via the vehicle-to-everything (V2X) 5G communication module, supporting subsequent model iteration and optimization. For data compression, numerical parameters employ the LZ77 lossless compression algorithm to preserve accuracy, while process data uses a tiered compression mode to reduce data volume within a preset accuracy threshold. A CRC data verification mechanism is configured to verify transmitted and stored data, ensuring data integrity. Through bandwidth priority allocation, hierarchical data storage, and compression optimization, the system adapts to the limited storage and bandwidth resources of the automotive system.

[0032] Step S120: Establish a dynamic, electrically coupled optimization model, incorporate temperature effect quantification equations and deep learning, and complete adaptive learning of aging characteristics;

[0033] Based on a pseudo-two-dimensional electrochemical mechanism model (P2D), this study elucidates the correlation between core parameters such as battery conductivity, ion diffusion coefficient, and electrode reaction rate constant and temperature. A temperature-dependent parameter quantification equation is constructed, transforming the influence of temperature on pulse charging response into model correction terms, thus forming a fundamental thermo-electric coupling mechanism model. The integrated deep learning network employs an LSTM architecture, which excels at capturing time-series data features and adapts to the dynamic changes during battery aging. Thermal and electrical parameter data at different aging stages are collected through accelerated aging experiments to construct a sample database. The LSTM network is then trained to establish a mapping relationship between aging characteristics and thermo-electric coupling parameters. This mapping relationship is embedded into the fundamental mechanism model, forming a fusion model. Adaptive learning of aging characteristics is achieved through online model updates, real-time monitoring of prediction bias, and dynamic adjustment of model parameters to ensure the model accurately describes the battery's thermo-electric coupling characteristics under different aging states.

[0034] Step S121: Customize the data transmission architecture, select a high-speed anti-interference bus interface adapted to the vehicle environment, combine the bus protocol to optimize point-to-point communication between the sensing module and the vehicle BMS, divide the priority level according to the real-time data requirements, allocate high bandwidth to ensure real-time transmission for core control parameters, and adopt batch transmission mode for non-critical information.

[0035] The data transmission architecture adopts a point-to-point communication mode, selecting the CANFD bus interface adapted to the automotive environment. This interface's transmission rate meets real-time control requirements and possesses good electromagnetic interference resistance. During bus protocol optimization, the data frame format is customized to simplify redundant fields and improve transmission efficiency. Based on real-time data requirements, three priority levels are established: Level 1 is for core control parameters such as temperature and voltage, allocated 50% of the bandwidth to ensure real-time transmission; Level 2 is for electrochemical parameters, allocated 30% of the bandwidth; Level 3 is for non-critical information such as historical data, using a batch transmission mode to time-division multiplex the remaining bandwidth and avoid bandwidth waste. Through priority allocation and transmission mode optimization, the real-time performance and reliability of core data transmission are ensured, while bandwidth utilization is improved.

[0036] Step S122: Build a hierarchical storage system. Use high-reliability flash memory to cache recently collected raw data, preprocessing results and model output data locally. Upload the historical data of the entire life cycle to the cloud distributed database through vehicle networking technology. Select appropriate compression algorithms for different types of data characteristics. Use lossless compression algorithms for numerical parameters and hierarchical compression mode for process data with high redundancy. Set up a data verification mechanism at the same time to ensure data integrity.

[0037] In the tiered storage system, local storage utilizes industrial-grade high-reliability flash memory, which is shock-resistant, operates over a wide temperature range, and is suitable for the complex in-vehicle environment. It caches recently collected raw data, preprocessed results, and model output data, meeting the data retrieval needs for real-time control and model updates. Cloud storage employs a distributed database architecture, uploading historical data throughout the entire lifecycle via 5G communication technology from the vehicle network. The database supports massive data storage and rapid retrieval, providing support for subsequent model iteration and optimization, and operational data analysis. Data compression uses algorithms adapted to different data types. Numerical parameters employ lossless compression algorithms to ensure data accuracy is not affected; highly redundant process data uses a tiered compression mode, setting different compression ratios according to data importance. A CRC data verification mechanism is synchronously implemented, performing verification before and after data transmission and storage to ensure data integrity.

[0038] Step S123: Enhance the anti-interference and fault tolerance capabilities of transmission and storage. The transmission link adopts shielded cable and differential transmission design, and the storage link introduces data backup and fault tolerance mechanisms.

[0039] In the anti-interference design of the transmission link, tin-plated copper wire shielded cables are used, with an outer layer of aluminum foil shielding. Both ends of the shielding layer are grounded, effectively suppressing electromagnetic radiation and conducted interference. Signal transmission employs a differential transmission circuit design, amplifying the useful signal and suppressing common-mode interference through a differential amplifier, thus improving the anti-interference capability of the transmitted signal. The storage fault tolerance mechanism adopts a local dual-backup strategy, storing critical data simultaneously in two independent flash memory chips. If one chip fails, data can be retrieved from the other. Cloud storage uses a geographically distributed backup mode, synchronously storing data to server nodes in different regions. A data error detection and repair mechanism is also introduced, periodically verifying the stored data and automatically recalling backup data to repair corrupted data, preventing data loss under extreme conditions.

[0040] Step S124: Based on the pseudo-two-dimensional electrochemical mechanism model, sort out the correlation between battery electrochemical parameters and temperature, construct a temperature-dependent parameter quantification equation, and transform the influence of temperature change on pulse charging response into model correction terms to form a basic thermo-electric coupling mechanism model.

[0041] Based on a pseudo-two-dimensional electrochemical mechanism model (P2D), this model encompasses the electrochemical processes of core structures such as electrodes, electrolytes, and separators, accurately reflecting the charge transfer and ion diffusion patterns within the battery. The correlation between core electrochemical parameters such as battery conductivity, ion diffusion coefficient, and electrode reaction rate constant and temperature is analyzed. Data collected through multi-temperature experiments is used to analyze the parameter trends with temperature, constructing temperature-dependent parameter quantification equations. The impact of temperature changes on pulse charging response is transformed into a computable correction term in the model. This correction term exhibits a non-linear relationship with the temperature change, and the correlation is obtained by fitting experimental data. After forming the basic thermo-electric coupling mechanism model, the model parameters are calibrated by comparing with battery charge-discharge experimental data at different temperatures, ensuring that the model accurately describes the influence of temperature on the battery's electrochemical characteristics.

[0042] Step S125: Analyze the intrinsic relationship between battery aging characteristics and thermal-electric coupling characteristics, define aging state evaluation indicators, collect thermal and electrical parameter data under different aging stages, temperature ranges and SOC ranges through accelerated aging experiments, and construct a multi-condition sample database covering the entire life cycle.

[0043] When analyzing battery aging characteristics, capacity decay rate and internal resistance growth rate are used as core evaluation indicators. The intrinsic relationship between these indicators and thermo-electric coupling characteristics is clarified: the higher the degree of aging, the greater the battery polarization resistance, the smaller the ion diffusion coefficient, and the more unstable the thermo-electric coupling characteristics. Accelerated aging experiments employ a combination of cyclic charge-discharge and high-temperature storage, setting different cycle numbers and storage durations to simulate the entire battery lifecycle aging process. The experiments cover different temperature ranges and SOC ranges, simultaneously collecting thermal (temperature, temperature rise rate) and electrical (voltage, current) parameter data. An electrochemical workstation is used to assist in testing the battery's internal electrochemical state. The collected data is categorized and labeled according to aging stage, temperature, and SOC, constructing a multi-condition sample database covering the entire lifecycle. The database includes data labels, collected operating condition information, and battery state parameters, providing data support for subsequent model training.

[0044] Step S126: Build a deep learning network model with LSTM network as the core, take thermal and electrical parameter data as input, output the prediction results of thermal-electric coupling characteristics under battery aging state, train and optimize the network through sample database, establish the mapping relationship between aging characteristics and thermal-electric coupling parameters, and embed the mapping relationship into the basic thermal-electric coupling mechanism model to form a dynamic thermal-electric coupling optimization model that integrates mechanism and data.

[0045] A deep learning model based on an LSTM network was built. This network excels at processing time-series data and can capture the dynamic changes in thermal-electric coupling characteristics during battery aging. The LSTM network was trained using a sample database, taking thermal and electrical parameters as input and outputting predictions of thermal-electric coupling characteristics under aging conditions, thus establishing a mapping relationship between aging characteristics and coupling parameters. This mapping relationship was embedded into a basic thermal-electric coupling mechanism model, and the parameter calling logic of the mechanism model was modified so that the model no longer relies on fixed parameters during runtime. Instead, it uses the prediction results of the LSTM model in real time via an interface as dynamic input parameters for the mechanism model, forming a dynamic thermal-electric coupling optimization model that integrates mechanism constraints and data-driven prediction. The fused model was jointly debugged, and the predicted data and actual test data were compared under different operating conditions to optimize model parameters and improve the model's accuracy and robustness.

[0046] Step S1261: Determine the dimensions of the thermal and electrical input parameters, select temperature and temperature rise rate as thermal parameters, and voltage, current and polarization voltage as electrical parameters, sort out the time series characteristics of the parameters, and perform sliding window segmentation on the original data according to a fixed time window to form time series data samples that meet the input requirements of the LSTM network. At the same time, set the output parameters as temperature-dependent polarization resistance, double layer capacitance, electrode reaction rate constant and other thermo-electric coupling characteristic indicators.

[0047] When determining the thermal domain input parameters, temperature and temperature rise rate are selected as core parameters based on the influence mechanism of temperature on the battery's electrochemical characteristics. These parameters directly reflect changes in the battery's thermal state. For the electrical domain parameters, voltage, current, and polarization voltage are selected. Polarization voltage characterizes the degree of resistance to charge transfer and ion diffusion within the battery and is strongly correlated with thermo-electric coupling characteristics. When analyzing the temporal characteristics of the parameters, the original data is segmented into sliding windows with a fixed time window length adapted to the pulse charging cycle and data acquisition frequency, ensuring that the segmented time-series data fully reflects the parameter change trends. The output parameters are set as temperature-dependent polarization resistance, double-layer capacitance, and electrode reaction rate constant. These parameters are the core input parameters of the basic P2D mechanism model, ensuring accurate matching between the LSTM model output and the mechanism model parameter requirements.

[0048] Step S1262: Design an LSTM network architecture to match the input layer dimension with the total number of thermal and electrical parameters, set multiple hidden layers, and adaptively adjust the number of neurons in each layer according to the sample data scale. The hidden layers use the ReLU activation function, the output layer uses the linear activation function, introduce the Dropout layer and L2 regularization mechanism, and configure the batch normalization layer.

[0049] The LSTM network input layer dimension is set according to the total number of thermal and electrical parameters to ensure that the input data can comprehensively reflect the battery's thermal-electrical state. Two to three hidden layers are used, with the number chosen based on a balance between data complexity and model generalization ability. The number of neurons in each layer is adaptively adjusted according to the sample data size to avoid overfitting or underfitting. The hidden layers use the ReLU activation function, which effectively alleviates the gradient vanishing problem and improves model training efficiency. The output layer uses a linear activation function to accommodate the continuous output requirements of coupled characteristic parameters. A Dropout layer is introduced to suppress overfitting, with the Dropout probability dynamically adjusted according to the model's generalization ability during training. A batch normalization layer is configured to normalize the input data of each layer, accelerating model training convergence and improving model stability.

[0050] Step S1263: Perform data preprocessing on the multi-condition sample database, use the Z-Score standardization method to normalize the input and output parameters, and divide the training set, validation set and test set according to a preset ratio to ensure the consistency of the dataset distribution.

[0051] The input and output parameters in the multi-condition sample database are processed using the Z-Score standardization method. The mean and standard deviation of the parameters are calculated to transform the raw data into standardized data, eliminating the influence of dimensional differences between different parameters. Training, validation, and test sets are divided according to the principle of data distribution consistency. Stratified sampling is used in the partitioning process to ensure that each dataset covers operating conditions with different temperatures, SOCs, and aging stages. The training set is used for iterative updates of network parameters, the validation set is used to monitor overfitting during training (hyperparameters are adjusted when the validation set loss function continuously increases), and the test set is used for final model performance evaluation, judging whether the model meets the requirements through indicators such as prediction error and accuracy. Standardization is completed before dataset partitioning to ensure the uniformity and accuracy of data processing.

[0052] Step S1264: Configure network training parameters, select Adam optimizer, set initial learning rate and dynamically adjust it using cosine annealing strategy, use mean squared error as loss function, set maximum number of training iterations and early stopping strategy, terminate training and save optimal network parameters when the validation set loss function does not decrease for a preset number of consecutive rounds.

[0053] The Adam optimizer was selected to configure the network training parameters. This optimizer combines momentum gradient descent and an adaptive learning rate strategy, enabling rapid convergence and reducing the likelihood of getting trapped in local optima. The initial learning rate was set based on model complexity and data size, and a cosine annealing strategy was used to dynamically adjust the learning rate. A higher learning rate was maintained in the early stages of training to accelerate convergence, while the learning rate was reduced in the later stages to optimize parameter accuracy. Mean squared error was used as the loss function, which effectively measures the continuous deviation between the predicted output and the actual coupling characteristic parameters. A maximum number of training iterations and an early stopping strategy were set. The maximum number of iterations was adapted to the data size and model complexity. The early stopping strategy was set to terminate training when the validation set loss function did not decrease for a preset number of consecutive iterations, avoiding overfitting and saving the current optimal network parameters for subsequent model deployment and application.

[0054] Step S1265: Start the network training process, input the preprocessed training set data into the LSTM network, calculate the predicted output through forward propagation, update the network weights and biases through back propagation, evaluate the model's generalization ability and adjust the hyperparameters using the validation set data, and verify the model's prediction accuracy using the test set data after training is completed.

[0055] When starting network training, preprocessed training data is input into the LSTM network in batches. The output of each layer is calculated through forward propagation to obtain the prediction results of the coupling characteristic parameters. Based on the deviation between the prediction results and the actual labels, the gradient is calculated using the backpropagation algorithm to update the network weights and biases, and the loss function is iteratively optimized. During training, after each iteration, the model's generalization ability is evaluated using validation set data. Hyperparameters such as the number of hidden layer neurons and the Dropout probability are adjusted based on the validation set loss to ensure the model maintains high-accuracy predictions even on unseen data. After training, test set data is input into the model, and performance metrics such as prediction error and accuracy are calculated. If the metrics do not meet the preset requirements, the feature selection process is reviewed to supplement features or the data preprocessing process is optimized, iteratively optimizing the model until the performance standards are met.

[0056] Step S1266: Extract the parameters of the trained LSTM network, construct a mapping relationship model between aging characteristics and thermo-electric coupling parameters. This model takes real-time collected thermal and electrical time-series data as input and outputs the predicted values ​​of coupling characteristic parameters under the current aging state. Define the interface specifications between this mapping relationship model and the basic pseudo-two-dimensional electrochemical mechanism model, and clarify the correspondence between the LSTM output parameters and the temperature-dependent parameters in the P2D model.

[0057] After training, core parameters such as weights, biases, and hidden layer states of the LSTM network are extracted to construct a mapping model between aging characteristics and thermo-electric coupling parameters. This model takes real-time acquired thermal and electrical time-series data as input and quickly outputs predicted values ​​of coupling characteristic parameters under the current aging state through forward propagation. The interface specifications between the mapping model and the basic P2D mechanism model are defined, clarifying the data format, transmission protocol, and parameter call timing of the interface, which supports real-time data interaction. The correspondence between LSTM output parameters and temperature-dependent parameters in the P2D model is established, creating a parameter mapping table to ensure that parameters such as polarization resistance and double-layer capacitance output by the LSTM can be directly used as input to the P2D model, achieving seamless integration between the two models.

[0058] Step S1267: Embed the mapping relationship model into the basic thermo-electric coupling mechanism model, modify the parameter calling logic of the P2D model, so that the output results of the LSTM mapping model can be called in real time through the interface during the model operation, and used as the dynamic input of the electrical conductivity, ion diffusion coefficient and reaction rate constant parameters in the P2D model, forming a dynamic thermo-electric coupling optimization model that integrates mechanism constraints and data-driven prediction.

[0059] When embedding the mapping relationship model into the basic thermo-electric coupling mechanism model, the parameter calling logic of the P2D model is modified, the original fixed parameter assignment statements are deleted, and an interface calling module is added. When the fused model runs, the interface calling module receives the predicted values ​​of coupling characteristic parameters output by the LSTM mapping model in real time through a preset interface specification, and uses them as dynamic inputs for key parameters such as conductivity, ion diffusion coefficient, and reaction rate constant in the P2D model. The operation flow of the fused model is as follows: multi-dimensional sensing network collects real-time data → inputs to the LSTM mapping model to obtain predicted parameters → the interface module transmits the data to the P2D model → the P2D model calculates and outputs the thermo-electric coupling characteristic results, forming a dynamic thermo-electric coupling optimization model with fused mechanism constraints and data-driven prediction, ensuring that the model can accurately describe the battery characteristics under different aging states.

[0060] Step S1268: Perform joint debugging on the fused model. Under different operating conditions such as temperature, SOC, and aging stage, compare the coupling characteristic parameters predicted by the model with the actual test data, correct the matching coefficient between the LSTM network output and the P2D model parameters, and adopt the model integration strategy to reduce the prediction variance.

[0061] During joint debugging, a multi-condition test platform was built to simulate the battery's operating environment under different temperatures, SOC, and aging stages. Pulse charging excitation was applied through a charging and discharging cabinet, and the actual thermo-electric coupling characteristic parameters of the battery were collected simultaneously. The prediction results of the fusion model were compared with the actual test data, and the prediction error of each parameter was calculated. Based on the error, the matching coefficient between the LSTM network output and the P2D model parameters was corrected to optimize the parameter correspondence. A model ensemble strategy was adopted to weight and fuse the prediction results of multiple trained LSTM models. The weights were allocated according to the prediction accuracy of each model to reduce the prediction variance of a single model. Through multiple rounds of debugging and optimization, the stability and robustness of the fusion model throughout its entire life cycle and under complex operating conditions were improved, ensuring that the model's prediction accuracy meets the requirements of vehicle control.

[0062] Step S127: Design an online model update mechanism to monitor the deviation between the model's predicted output and the actual collected data in real time, set a deviation threshold and update cycle, and automatically call the latest valid collected data to fine-tune the deep learning network parameters when the deviation exceeds the threshold or reaches the preset number of cycles, and synchronously update the coefficients of the temperature-dependent parameter quantization equation.

[0063] When designing the online model update mechanism, a deviation monitoring module is embedded in the fusion model to compare the model's predicted output with the actual data collected by the multi-dimensional perception network in real time, calculating the deviation value and deviation rate. A deviation threshold and update cycle are set; the deviation threshold is determined based on the model's prediction accuracy requirements, and the update cycle is adapted to the rate of battery state change. When the deviation value exceeds the threshold or reaches the preset number of cycles, the model update process is triggered. The system automatically selects the latest valid data (including normal and aging conditions) from the onboard storage unit, incrementally trains the LSTM network, and updates the network weights and bias parameters. Simultaneously, the temperature-dependent parametric quantization equation is refitted based on the new data, and the equation coefficients are updated to ensure the model can dynamically adapt to the drift of thermal-electric coupling characteristics during battery aging, maintaining the model's prediction accuracy.

[0064] Step S128: Optimize the computational complexity of the dynamic thermal-electric coupling optimization model through model simplification algorithm, and reduce redundant calculation steps while ensuring prediction accuracy.

[0065] A model reduction approach is employed to optimize computational complexity, simplifying the diffusion and charge transfer equations in the P2D model while retaining core influencing factors and eliminating secondary variables and higher-order terms, thereby reducing computational load while maintaining prediction accuracy. Redundant computational steps, such as repetitive parameter initialization and the calculation and storage of non-core intermediate variables, are identified and eliminated to optimize the model's execution flow. A pre-calculated parameter mapping table is used to store parameters such as the basic diffusion coefficient and polarization resistance range at different temperatures and SOCs locally. During model runtime, the mapping table data is directly called for interpolation calculations, shortening real-time computation time. A parallel computing architecture is adopted, distributing data preprocessing, model prediction, and parameter updates to different computing cores for parallel execution, ensuring that the control command response time matches the high-frequency characteristics of pulse charging and meets the real-time control requirements of vehicles.

[0066] Step S130: Design a thermal and electrical coordinated closed-loop control algorithm, using dual safety thresholds as constraints, and optimize pulse parameters through model prediction to form a closed-loop execution;

[0067] The thermo-electric co-operational closed-loop control algorithm uses thermal safety constraints and electrochemical safety constraints as dual core constraints, embedding a dynamic thermo-electric coupling optimization model into a model predictive control (MPC) framework. The algorithm first acquires real-time parameters through a multi-dimensional sensing network. The MPC framework then predicts battery state changes based on the coupling model, optimizing pulse amplitude, duty cycle, and frequency under dual threshold constraints. The optimized parameters are sent to the power execution unit for execution via a high-speed bus. Simultaneously, the sensing network collects feedback data in real time, updating the model state and constraint boundaries, forming a complete closed loop of "prediction-optimization-execution-feedback." This ensures that the charging process balances efficiency and safety, adapting to the dynamic control requirements of on-board pulse charging.

[0068] Step S131: Quantify and dynamically adapt the dual safety thresholds consisting of thermal safety constraints and electrochemical safety constraints. Combine the battery SOC range, aging degree and ambient temperature to define the upper limit of single cell temperature, upper limit of temperature rise rate, lower limit of negative electrode potential and upper limit of SEI film impedance change rate. Establish dynamic adjustment rules for the thresholds. Tighten the threshold boundaries in the high SOC stage, late aging stage or extreme temperature environment, and appropriately relax the threshold boundaries in the low SOC stage.

[0069] When quantifying the dual safety thresholds, the upper limit of the single-cell temperature and the upper limit of the temperature rise rate are defined as thermal safety constraints, while the lower limit of the negative electrode potential and the upper limit of the SEI film impedance change rate constitute electrochemical safety constraints, with reference to the battery's factory technical parameters and multi-condition experimental data. The dynamic adaptation rule is implemented by constructing a threshold adjustment function, taking SOC, aging degree, and ambient temperature as input variables: when the SOC stage is high (>80%), the number of aging cycles exceeds 1000, or the ambient temperature is below -10℃ / above 50℃, the threshold is tightened according to a preset ratio; when the SOC is below 20% and the ambient temperature is between 20-30℃, the threshold is appropriately relaxed to balance charging efficiency and safety redundancy, ensuring that the threshold is dynamically adjusted according to the real-time status of the battery.

[0070] Step S132: Determine the pulse parameter optimization variables and multi-objective function. The optimization variables include pulse amplitude, duty cycle and frequency. The objective function covers charging speed, energy efficiency and battery life. The priority of each objective is dynamically allocated by weighting coefficients, and the weight ratio of each objective is adaptively adjusted under different operating conditions.

[0071] The pulse parameter optimization variables are determined to be pulse amplitude, duty cycle, and frequency, all of which are taken within the physical limits allowed by the battery and power devices. The multi-objective function covers charging speed (maximizing average charging current), energy efficiency (minimizing charge / discharge losses), and battery life (minimizing polarization and SEI film damage), with priorities assigned through dynamic weighting coefficients. In fast charging scenarios, charging speed is given the highest weight; in everyday slow charging scenarios, battery life is prioritized, increasing its weight; and in low-temperature or aging battery conditions, energy efficiency is given a higher weight. The weighting coefficients are automatically adjusted by the operating condition identification module to ensure that the target priorities are accurately matched to actual usage requirements.

[0072] Step S133: Integrate the dynamic thermal and electrical coupling optimization model into the model predictive control framework, set the prediction time domain and control time domain, and predict the correlation between the battery state change trend and pulse parameters within a preset time period based on the real-time collected thermal, electrical and electrochemical parameters.

[0073] When integrating the dynamic thermo-electric coupling optimization model into the MPC framework, the prediction time domain is set to the next 5-10 control cycles, and the control time domain is set to one control cycle, based on the pulse charging cycle and battery response characteristics. Using real-time thermal (temperature, temperature rise rate), electrical (voltage, current), and electrochemical (polarization resistance, double-layer capacitance) parameters collected by a multi-dimensional sensing network, the coupling model predicts battery temperature changes, polarization voltage accumulation, and safety margin within a preset time period. The correlation between these state changes and pulse parameter adjustments is established, providing an accurate predictive basis for MPC optimization calculations.

[0074] Step S134: Construct a constraint condition matrix, transform the dual safety thresholds into MPC inequality constraints, and incorporate pulse parameter physical constraints to form a multi-dimensional constraint set. The pulse parameter physical constraints include pulse amplitude not exceeding the maximum allowable current of the battery, frequency matching the switching capability of power devices, and duty cycle within a reasonable range.

[0075] When constructing the constraint matrix, the dual thresholds of thermal safety and electrochemical safety are transformed into inequality constraints recognizable by the MPC framework, clearly defining the boundary range of each constraint. Simultaneously, physical constraints on pulse parameters are incorporated: the pulse amplitude does not exceed the battery's maximum allowable charging current, the frequency matches the switching capability of the onboard DC / DC converter (e.g., 5-20kHz), and the duty cycle is limited to a reasonable range of 10%-90%. By integrating multi-dimensional constraints in matrix form, a unified constraint set is formed. During MPC optimization, constraint verification ensures that the optimal solution does not exceed safety limits and hardware performance constraints, guaranteeing the feasibility of control commands.

[0076] Step S135: Design a fast rolling optimization solution mechanism, use a quadratic programming solver to solve the objective function of MPC in real time, output the current optimal pulse parameter combination in each control cycle, execute the parameter instructions for the first control step, and restart the optimization calculation in the next control cycle based on the latest feedback battery state data, so as to realize the rolling iteration of prediction, solution, execution and feedback;

[0077] The fast rolling optimization solution mechanism employs an embedded adaptive quadratic programming solver, whose computational efficiency is well-suited to the real-time control requirements of vehicles. Within each control cycle, the solver performs real-time solving of the multi-objective function based on the current battery state data and constraint matrix, outputting the current optimal pulse parameter combination. Only the parameter instructions for the first control step are executed to avoid the impact of future prediction uncertainties on control performance. At the start of the next control cycle, the optimization problem is re-initialized and solved based on the latest battery state data fed back from the multi-dimensional sensing network, achieving a rolling iteration of "prediction-solution-execution-feedback" to ensure real-time matching of parameter optimization with battery state.

[0078] Step S136: Establish a closed-loop execution and state feedback link, send the optimized pulse parameters to the vehicle power execution unit through the high-speed bus, drive it to output the corresponding pulse charging waveform, and at the same time collect the battery temperature, voltage, current and electrochemical parameters in real time during the charging process through the multi-dimensional sensing network, and feed them back to the MPC framework and the thermo-electric coupling model to update the model state and constraint boundary.

[0079] When establishing a closed-loop execution and state feedback link, the optimized pulse parameters are transmitted to the on-board power execution unit (DC / DC converter) via the CANFD high-speed bus. This unit generates corresponding pulse charging waveforms based on the parameter commands, precisely controlling the charging current and voltage. Simultaneously, a multi-dimensional sensing network collects battery temperature, voltage, current, and electrochemical parameters in real time during the charging process at a preset sampling frequency, and synchronously feeds them back to the MPC framework and the thermo-electric coupling model via the same bus. The MPC framework updates the constraint boundaries based on the feedback data, and the coupling model corrects its own state parameters, ensuring the real-time performance and accuracy of the entire closed-loop link and achieving dynamic adaptive control.

[0080] Step S137: Formulate a threshold-triggered emergency adjustment strategy. When the feedback data shows that the battery status has reached a preset proportion of the safety threshold, start the pre-adjustment mechanism and slightly correct the pulse parameters. If the battery status is detected to have exceeded the safety threshold, immediately trigger the emergency adjustment command to adjust the pulse amplitude or frequency until the battery status returns to the safe range.

[0081] When formulating a threshold-triggered emergency adjustment strategy, a 90% threshold is preset as the warning line. When feedback data shows that the battery status has reached the warning line, a pre-adjustment mechanism is activated, slightly reducing the pulse amplitude by 5%-10% or fine-tuning the duty cycle to suppress risk accumulation in advance. If the battery status is detected to exceed the safety threshold, an emergency adjustment command is immediately triggered, significantly reducing the pulse amplitude by 20%-30% or increasing the frequency to the upper limit, while limiting the peak charging current. During the adjustment process, the sensing network continuously collects battery status data and dynamically corrects the parameter adjustment range until the battery status returns to the safe range, preventing the safety risk from escalating.

[0082] Step S138: Optimize the real-time performance of the algorithm. Reduce the MPC operation time by reducing the model order, pre-calculating the optimization interval, and simplifying the solution complexity. Use a parallel computing architecture to allocate data processing and optimization tasks to ensure that the control command response time matches the high-frequency characteristics of pulse charging.

[0083] To optimize the real-time performance of the algorithm, a model reduction method is used to simplify the dynamic thermo-electric coupling model, eliminating minor variables and higher-order terms while retaining core influencing factors and reducing computational redundancy. By pre-calculating the optimization parameter ranges under different operating conditions, the optimal solution is searched only within the preset range during the solution process, shortening the solution time. The computational flow of the quadratic programming solver is simplified to reduce complexity. An FPGA parallel computing architecture is adopted, distributing tasks such as data preprocessing, model prediction, and optimization to different computing cores for parallel execution, ensuring that the control command response time is controlled at the millisecond level, matching the high-frequency characteristics of pulse charging, and meeting the requirements of on-board real-time control.

[0084] Step S140: Construct a multimodal security boundary identification and early warning system, integrating multi-feature extraction and sensing technologies to complete security risk early warning and parameter degradation control;

[0085] The constructed multimodal safety boundary identification and early warning system integrates voltage sensing, high-frequency impedance spectroscopy, ultrasonic sensing, and temperature sensing modules to achieve accurate identification of safety risks through multi-source data fusion. The system first synchronously collects multimodal data and extracts core features. A dynamic safety boundary identification model determines the risk level, and then outputs differentiated early warning signals based on the level, triggering parameter degradation control. Simultaneously, a closed-loop verification and adaptive optimization mechanism is established to monitor system performance in real time and dynamically update model parameters, ensuring that the system can efficiently complete risk early warning and safety control throughout the battery's entire life cycle and under complex operating conditions.

[0086] Step S141: Integrate voltage sensing, high-frequency impedance spectrum, ultrasonic sensing and temperature sensing modules to establish a multi-source data synchronous acquisition mechanism. Through a unified synchronous trigger signal, collect voltage fluctuation data, broadband impedance spectrum data, ultrasonic propagation signal and temperature data in real time during battery charging to ensure the consistency of timestamps for different modal data.

[0087] When integrating multiple sensor modules, the hardware interfaces of voltage sensors, high-frequency impedance spectroscopy modules, ultrasonic sensors, and temperature sensors are uniformly adapted and connected to the core control unit. A multi-source data synchronous acquisition mechanism is established, using a timer within the control unit to generate a unified synchronous trigger signal, ensuring that the four types of sensors start data acquisition at the same timestamp, achieving timestamp consistency for voltage fluctuation data, broadband impedance spectroscopy data, ultrasonic propagation signals, and temperature data. During acquisition, hardware anti-shake and signal synchronization calibration technologies are used to further reduce data timing deviations, providing a high-quality data foundation for subsequent multi-dimensional feature fusion and risk identification.

[0088] Step S142: Extract fluctuation features from voltage data, analyze the relevant parameters related to lithium deposition and SEI film damage from impedance spectrum data, extract features characterizing internal stress accumulation and SEI film integrity from ultrasonic signals, normalize and remove redundancy of the extracted features, and retain features strongly related to safety risks.

[0089] In the feature extraction stage, fluctuation features such as peak-to-peak value, kurtosis, and skewness are extracted from voltage data. Parameters related to lithium deposition and SEI film damage, such as the real part of low-frequency impedance and characteristic frequency, are analyzed from impedance spectroscopy data. Features characterizing electrode stress and SEI film integrity, such as propagation velocity variation and attenuation coefficient, are extracted from ultrasonic signals. All extracted features are normalized using the Z-Score standardization method to eliminate dimensional differences. Features with strong correlation to safety risks are screened through mutual information entropy analysis, and redundant features are eliminated. The retained core features focus on four major risk dimensions: voltage anomaly, lithium deposition, SEI film damage, and electrode stress accumulation, ensuring the relevance and effectiveness of the features.

[0090] Step S143: Construct a dynamic safety boundary identification model, classify safety risks into four identification dimensions: voltage anomaly, lithium deposition, SEI film damage, and electrode stress accumulation. Based on the battery electrochemical mechanism and multi-condition experimental data, establish the mapping relationship between each feature and safety risk. Combine SOC, temperature, and aging degree to dynamically adjust the boundary threshold. Use machine learning algorithms to train the boundary recognizer and optimize the model parameters.

[0091] When constructing the dynamic safety boundary identification model, four identification dimensions are defined according to the type of safety risk. The correlation logic between features and risks is clarified based on the battery electrochemical mechanism, and the correlation is quantified by combining multi-condition experimental data. A dynamic adjustment function is established using SOC, temperature, and aging degree as threshold adjustment factors to achieve boundary threshold adaptation to battery state. The XGBoost algorithm is selected to build the boundary recognizer. The labeled sample data is divided into training, validation, and test sets, and the model is trained after optimizing hyperparameters. A multi-feature weighted fusion and cross-validation mechanism is incorporated, and an online model update module is designed to correct model parameters in real time, ensuring that the model accurately identifies safety boundaries in all usage scenarios.

[0092] Step S1431: For the four safety risk dimensions of voltage anomaly, lithium deposition, SEI film damage, and electrode stress accumulation, determine the strongly correlated characteristic indicators to form a characteristic system for each risk dimension;

[0093] The voltage anomaly dimension focuses on features such as voltage peak-to-peak value, instantaneous fluctuation amplitude, and duration of exceeding threshold, directly reflecting voltage stability. The lithium deposition dimension focuses on indicators such as the increment of the real part of the low-frequency impedance in the impedance spectrum, the characteristic frequency shift, and the estimated value of the negative electrode potential, which are related to interface changes caused by lithium deposition. The SEI film damage dimension selects parameters such as the characteristic frequency change rate, the phase angle shift of the impedance spectrum, and the self-discharge rate increment to characterize interface integrity. The electrode stress accumulation dimension determines features such as the change in ultrasonic propagation velocity, the signal attenuation coefficient, and stress-related impedance parameters, matching the stress state of the electrode structure to ensure that each risk dimension has strongly correlated features to support identification accuracy.

[0094] Step S1432: Simulate multiple charging scenarios under different SOC ranges, ambient temperatures, and aging stages, and simultaneously collect multimodal data such as voltage, impedance spectrum, ultrasound, and temperature. Obtain internal battery state information through professional testing methods, label the collected data with safety risk levels, and construct a labeled sample dataset covering all usage scenarios.

[0095] Multi-condition experiments cover different SOC ranges (divided in 10% increments), ambient temperatures (-20℃, 0℃, 25℃, 45℃, 60℃), and aging stages (new batteries, 500 / 1000 / 1500 cycles), simulating typical scenarios such as normal charging, overcharging, and low-temperature fast charging. Pulse charging excitation is applied through a charge / discharge cabinet, simultaneously collecting multi-modal data including voltage, impedance spectrum, ultrasound, and temperature. The negative electrode potential is tested using an electrochemical workstation, and the SEI film thickness and lithium dendrite growth are analyzed. Ultrasonic imaging is used to observe electrode stress distribution, and the collected data are labeled with a safety risk level of 0-4. All labeled data are integrated to construct a highly accurate sample dataset covering all usage scenarios, providing ample support for model training.

[0096] Step S1433: Based on electrochemical mechanism analysis, clarify the correlation logic between features and safety risks, then use statistical modeling methods to quantify the correlation between features and risk levels, screen features, establish a quantitative mapping model between features and risk levels through appropriate modeling methods, and clarify the risk probability distribution corresponding to different feature combinations;

[0097] First, the correlation between features and safety risks is clarified through electrochemical mechanism analysis, such as lithium deposition leading to an increase in the real part of low-frequency impedance and SEI film damage causing a shift in characteristic frequency. Then, based on a labeled sample dataset, the Pearson correlation coefficient is used to quantify the correlation between features and risk levels, and core features with strong correlations are selected. A quantitative mapping model between features and risk levels is established using a multiple regression method to clarify the risk probability distribution corresponding to different feature combinations. For example, when both the increment of the real part of low-frequency impedance and the characteristic frequency shift simultaneously meet preset conditions, the probability of lithium deposition risk is output, ensuring that the mapping relationship is both mechanistically supported and data reliable.

[0098] Step S1434: Using SOC, temperature, and aging degree as threshold adjustment factors, establish a dynamic adjustment function for each characteristic threshold based on multi-condition experimental data to achieve dynamic adaptation of the threshold to the battery state.

[0099] Using SOC, temperature, and aging level as threshold adjustment factors, a dynamic adjustment function for each core characteristic threshold is constructed based on multi-condition experimental data and a nonlinear fitting method. At high SOC (>80%), the output coefficient of the adjustment function is greater than 1, tightening the threshold; at low SOC (<20%), the coefficient is less than 1, relaxing the threshold; the higher the aging level (the more cycles), the closer the coefficient is to 1.5; the further the temperature deviates from the optimal range of 25℃, the closer the coefficient is to 1.5. Through this function, the threshold can be automatically adjusted according to the real-time state of the battery, such as tightening the lithium deposition risk threshold under low-temperature conditions and relaxing the voltage anomaly threshold during low-SOC fast charging, achieving dynamic threshold adaptation.

[0100] Step S1435: Combine the feature dimension and data scale to select an appropriate machine learning algorithm to build a boundary recognizer, divide the labeled feature data into training set, validation set and test set, standardize the features, and use appropriate methods to optimize the algorithm hyperparameters;

[0101] Combining feature dimensions (multi-dimensional tabular features) and data scale, the XGBoost algorithm was selected to build the boundary recognizer. This algorithm is suitable for multi-feature fusion and has strong generalization ability. The labeled core feature data was divided into training, validation, and test sets in a 7:2:1 ratio. The Z-Score method was used to standardize the features to eliminate the influence of dimensions. The grid search method was used to optimize the algorithm's hyperparameters, including tree depth, learning rate, and regularization coefficient. By traversing the preset hyperparameter combinations, the combination with the highest recognition accuracy on the validation set was selected as the optimal hyperparameters, ensuring that the recognizer architecture and parameters are adapted to the data characteristics.

[0102] Step S1436: Use the risk level corresponding to the dynamic threshold as the model output label, train the model using the training set, monitor the model performance using the validation set and adopt an early stopping strategy, evaluate the model recognition performance using the test set, and if the preset performance requirements are not met, backtrack to the feature selection stage for iterative optimization.

[0103] Using the risk level (0-4) corresponding to a dynamic threshold as the model output label, the training set data is input into the XGBoost recognizer for training. Forward propagation is used to calculate the predicted risk level, and backpropagation updates the model parameters. The model performance is monitored in real time using a validation set, employing an early stopping strategy: training is terminated when the validation set accuracy shows no improvement for 10 consecutive rounds to avoid overfitting. After training, the model's recognition performance is evaluated using test set data, calculating metrics such as accuracy, recall, and false positive rate. If the recognition accuracy for a certain risk dimension falls below 90%, the feature selection process is reviewed, adding new features or optimizing feature weights, iteratively optimizing the model until the preset performance requirements are met.

[0104] Step S1437: Use a weighted fusion strategy to integrate multiple features of the same risk dimension. The weights are assigned based on the correlation coefficient between the feature and the risk. When a single feature triggers the warning threshold, cross-feature cross-validation is initiated to establish feature conflict adjudication rules.

[0105] A weighted fusion strategy is employed to integrate multiple features within the same risk dimension. Weights are allocated based on the correlation coefficient between the feature and the risk; higher correlation results in higher weights. When a single feature triggers an early warning threshold, cross-feature cross-validation is initiated. For example, a lithium deposition early warning requires both exceeding the limit for the real part increment of low-frequency impedance and meeting the characteristic frequency shift standard, with both having a confidence level ≥85%, to avoid misjudgments caused by accidental fluctuations of a single feature. A feature conflict resolution rule is established: when different features indicate inconsistent risk levels, the feature with a stronger correlation to the electrochemical mechanism takes precedence. For instance, in SEI film damage identification, impedance spectrum phase angle shift characteristics are prioritized to ensure the accuracy of risk identification.

[0106] Step S1438: Record the deviation between the model recognition result and the actual state of the battery in real time, count the recognition error of each risk dimension, set the error threshold, and when the error reaches the threshold, automatically select the latest collected valid samples to perform incremental training on the model, and update the feature weights, mapping relationship parameters and the coefficients of the dynamic threshold adjustment function.

[0107] The system records the deviation between the model's recognition results and the actual battery condition (such as subsequent disassembly verification and charging anomaly feedback) in real time, and statistically analyzes the recognition error and false alarm rate for each risk dimension. An error threshold is set (e.g., a false alarm rate exceeding 2%). When the error reaches the threshold, the system automatically selects the latest valid samples (including normal and risk conditions) from the onboard storage unit to incrementally train the model, updating the feature weights and mapping parameters. Simultaneously, the coefficients of the dynamic threshold adjustment function are refitted based on the new sample data to ensure that the model parameters continuously adapt to the battery's lifecycle characteristics, maintaining long-term recognition accuracy.

[0108] Step S144: Design a multi-level risk assessment system, divide risk levels and clarify the corresponding feature threshold ranges and judgment conditions for each level. When a single modal feature reaches the warning threshold, multi-modal feature cross-validation is initiated. Set continuous monitoring cycle judgment rules. Warning is triggered only when the risk conditions are met within a continuous preset cycle. Differentiated warning signals are output according to the risk level.

[0109] A multi-level risk assessment system is designed, classifying safety risks into levels 0-4, and clearly defining the corresponding feature threshold ranges and judgment conditions for each level: Level 0 is no risk, all features are within the safe range; Levels 1-2 are low risk, a single-dimensional feature is close to the threshold; Levels 3-4 are medium-high risk, multiple-dimensional features exceed the threshold. When a single modal feature reaches the warning threshold, multi-modal feature cross-validation is initiated to avoid false alarms. A continuous monitoring cycle (e.g., 3 collection cycles) is set, and a warning is triggered only when the risk conditions are met within the continuous cycle. Differentiated warning signals are output according to the risk level, including on-board instrument panel prompts, BMS pop-ups, and background alarm log records, ensuring that users and the system are promptly aware of the risk status.

[0110] Step S145: Formulate a parameter downgrade control strategy that links risk levels, establish a correspondence between warning levels and pulse parameter adjustment ranges, make small adjustments to pulse parameters when the risk level is low, make large adjustments to pulse parameters and limit charging current peaks when the risk level is medium to high, provide real-time feedback of battery status data during the downgrade adjustment process, and dynamically correct the parameter adjustment range.

[0111] A risk level-linked parameter degradation control strategy was developed, establishing a correspondence between warning levels and pulse parameter adjustment amplitudes: For low-risk levels (levels 1-2), the pulse amplitude was slightly reduced by 5%-10% or the duty cycle was finely adjusted (±5%) to reduce charging efficiency loss while ensuring safety; for medium-to-high-risk levels (levels 3-4), the pulse amplitude was significantly reduced by 20%-30%, the pulse frequency was increased to the upper limit, or the pulse duration was shortened, while limiting the peak charging current to below the safety threshold. During the degradation adjustment process, a multi-dimensional sensing network provided real-time feedback on battery status data, dynamically correcting the parameter adjustment amplitude to ensure the battery status quickly returns to the safe range, balancing safety control and user experience.

[0112] Step S146: Establish a closed-loop verification and adaptive optimization mechanism for the system. Conduct tests under different operating conditions such as temperature, SOC range, and aging stage to verify the accuracy of safety boundary identification, early warning response speed, and effectiveness of degradation control. Design an online model update module to monitor identification error and false alarm rate in real time. When the error exceeds the preset threshold, automatically call the latest collected operating condition data to fine-tune the boundary model and feature weights.

[0113] A closed-loop verification and adaptive optimization mechanism was established for the system. Test conditions covered different temperatures, SOC ranges, and aging stages. A charging / discharging cabinet simulated real-world usage scenarios to verify the accuracy of safety boundary identification, early warning response speed, and the effectiveness of degradation control. An online model update module was designed to monitor identification errors and false alarm rates in real time. When the error exceeds a preset threshold (e.g., identification accuracy below 95% or false alarm rate above 2%), the system automatically retrieves the latest collected operating condition data to fine-tune the boundary model and feature weights. Through multiple rounds of verification and optimization, the system's robustness and adaptability under complex operating conditions and throughout the battery's entire lifecycle are continuously improved, ensuring stable and reliable early warning and control effects.

[0114] Step S150: Optimize the stage transition mechanism by setting a buffer period, using a parameter gradual change strategy and impedance compensation to suppress voltage spikes during the switching between pulse and constant current / constant voltage stages.

[0115] The transition mechanism between the pulse and constant current / constant voltage stages is optimized through three key technologies: buffer period setting, scenario-specific parameter gradation strategy, and virtual impedance compensation. These technologies work together to suppress voltage surges during stage switching. First, the switching timing is predicted based on the real-time battery status, and a buffer period is initiated, collecting the necessary state data for the transition. Then, adaptive parameter gradation schemes are designed for different switching scenarios to avoid parameter abrupt changes. Finally, polarization voltage estimation and impedance compensation are used to offset voltage fluctuations caused by polarization accumulation. These three technologies form a closed-loop synergy, ensuring a smooth voltage transition without significant oscillations during stage switching, thus meeting the stability and safety requirements of on-board charging systems.

[0116] Step S151: Establish a stage switching prediction and buffer period triggering mechanism. Based on the real-time SOC, voltage, and polarization state of the battery, determine the switching trigger threshold. When the parameters reach the preset warning range before the threshold, the buffer period is automatically started. During the buffer period, the current pulse charging mode is maintained, and relevant data such as voltage fluctuation peak, polarization voltage, and temperature change rate are collected synchronously.

[0117] A phased switching prediction and buffer period triggering mechanism is established. Based on real-time battery SOC, terminal voltage, and polarization state data, and combined with multi-condition experiments, the switching trigger threshold is determined. When a parameter is detected to enter a preset warning range before the threshold, a buffer period is automatically initiated (the duration is dynamically adjusted according to the operating condition, typically 5-10 seconds). During the buffer period, the current pulse charging mode remains unchanged to avoid premature adjustments that could cause state fluctuations. Simultaneously, key data such as voltage fluctuation peaks, polarization voltage, and temperature change rate are collected at a high-frequency sampling rate. These data serve as the core inputs for subsequent parameter gradual change strategy formulation and impedance compensation calculation, providing data support for a smooth transition.

[0118] Step S152: Design a scenario-specific parameter gradient strategy. Develop adaptation schemes for the two switching scenarios: pulse to constant current and pulse to constant voltage. When switching from pulse to constant current, linearly reduce the pulse amplitude at a preset rate and simultaneously gradually change the duty cycle to the corresponding target value. During the gradient process, match the target current value of the constant current stage in real time. When switching from pulse to constant voltage, adjust the pulse amplitude and duty cycle according to an exponential law to make the average voltage gradually approach the target voltage of the constant voltage stage. The gradient rate is dynamically adjusted based on the polarization accumulation collected during the buffer period.

[0119] The design incorporates a scenario-specific parameter gradient strategy, adapting to two core switching scenarios: pulse-to-constant-current and pulse-to-constant-voltage. When switching from pulse to constant-current, the pulse amplitude is reduced at a preset linear rate, while the duty cycle is gradually increased to 100%. During this transition, the average output current is compared in real-time with the target current value for the constant-current stage, and the gradient rate is dynamically adjusted to ensure a smooth transition to constant-current mode. When switching from pulse to constant-voltage, the pulse amplitude and duty cycle are adjusted exponentially, causing the average output voltage to gradually approach the target voltage for the constant-voltage stage. The gradient rate is dynamically adjusted based on the accumulated polarization during the buffer period; the greater the accumulated polarization, the slower the gradient rate, preventing oscillations caused by voltage surges.

[0120] Step S153: Construct a polarization voltage estimation model. Based on the current and voltage time series data and high-frequency impedance spectrum parameters collected during the buffer period, and combined with the battery electrochemical mechanism, calculate the sum of concentration polarization and electrochemical polarization voltage at the end of the pulse phase in real time.

[0121] A polarization voltage estimation model is constructed based on multi-source data collected during the buffer period, combined with the battery electrochemical mechanism to achieve accurate estimation. First, high-frequency impedance spectroscopy data is fitted using an equivalent circuit model to separate key electrochemical parameters. Then, calculation sub-models for electrochemical polarization and concentration polarization are built separately to quantify the two types of polarization voltages. A dynamic correction mechanism for temperature and SOC is incorporated, and the model coefficients are optimized through online fitting. Finally, the two polarization components are integrated to obtain the total polarization voltage, and a validity verification rule is established. Simultaneously, model order reduction and pre-calculation strategies are employed to optimize computational efficiency, ensuring that multiple rounds of iterative estimation are completed within the buffer period, meeting the real-time compensation requirements during stage switching.

[0122] Step S1531: Organize the input parameter system for polarization voltage estimation, clarify that the input data includes the current time series, voltage time series, and high-frequency impedance spectrum raw data during the buffer period, synchronously associate the battery nominal structural parameters and real-time state parameters, perform timestamp alignment processing on all input data, use digital filtering algorithms to remove high-frequency noise in the current and voltage signals, and smooth the impedance spectrum data to eliminate measurement fluctuations;

[0123] The input parameter system for polarization voltage estimation is outlined. Core input data includes the current time series, voltage time series, and raw high-frequency impedance spectrum data within the buffer period, synchronously correlated with nominal battery structural parameters (such as electrode thickness and active material particle size) and real-time state parameters (SOC, temperature). All input data undergoes timestamp alignment to ensure timing consistency. A Kalman filter algorithm is used to remove high-frequency noise from the current and voltage signals, and a moving average method is used to smooth the impedance spectrum data, eliminating measurement fluctuations. The processed input data possesses both completeness and reliability, laying the foundation for accurate subsequent polarization voltage estimation.

[0124] Step S1532: Fit the broadband impedance spectrum data using an equivalent circuit model to separate the relevant parameters of ohmic resistance, charge transfer polarization resistance and diffusion impedance. Dynamically update the polarization resistance value through the real-time impedance spectrum fitting results and establish the correlation between characteristic frequency and ion diffusion coefficient.

[0125] An equivalent circuit model (such as the Randle model) was used to fit and analyze the broadband impedance spectrum data. Ohmic resistance, charge transfer polarization resistance, and diffusion impedance-related parameters (such as characteristic frequency and diffusion resistance) were separated from the original impedance spectrum data. Based on real-time acquired impedance spectrum data, the polarization resistance value was dynamically updated to avoid estimation bias caused by using fixed parameters and improve model adaptability. Through statistical analysis and experimental verification, a quantitative correlation between the characteristic frequency and the ion diffusion coefficient was established. This correlation directly provides core fundamental parameters for subsequent concentration polarization voltage calculation, ensuring the accuracy of concentration polarization estimation.

[0126] Step S1533: Construct a sub-model for calculating electrochemical polarization voltage. Based on the charge transfer mechanism, using the real-time updated polarization resistance as a parameter, and combining the instantaneous current value during the buffer period, calculate the basic value of electrochemical polarization voltage using the Ohm's law derived formula. Introduce a current change rate correction term and dynamically adjust the polarization resistance weighting coefficient to compensate for the influence of current mutation on electrochemical polarization.

[0127] A sub-model for calculating electrochemical polarization voltage is constructed. Based on the battery charge transfer mechanism, the model uses the real-time updated charge transfer polarization resistance as the core parameter and combines it with the instantaneous current value during the buffer period. The basic value of the electrochemical polarization voltage is calculated through the derived relationship of Ohm's law. Considering that the current may fluctuate during pulse charging, a current change rate correction term is introduced. When the current fluctuation exceeds a preset range (e.g., ±10%), the weighting coefficient of the polarization resistance is dynamically adjusted to compensate for the impact of sudden current changes on the electrochemical polarization state. This ensures that the calculation results can accurately match the actual charge transfer process of the battery, improving the estimation accuracy of the electrochemical polarization voltage.

[0128] Step S1534: Build a concentration polarization voltage calculation sub-model. Based on Fick's first law and the simplified diffusion equation, take the ion diffusion coefficient derived from the impedance spectrum as the core, and combine the nominal structural parameters of the battery to establish a quantitative relationship between concentration polarization and current, diffusion coefficient and real-time state parameters. Use the pulse peak-valley difference in the voltage time series data during the buffer period to extract the concentration polarization related features and reverse correct the diffusion coefficient.

[0129] A sub-model for calculating concentration polarization voltage was constructed. Based on Fick's first law and a simplified diffusion equation, the ion diffusion coefficient derived from impedance spectroscopy data was used as the core parameter. Combined with nominal battery structural parameters (such as electrode thickness and active material particle size), a quantitative relationship between concentration polarization voltage, current, diffusion coefficient, and state of charge (SOC) was established. The peak-to-valley difference in voltage time-series data during the buffer period was used to extract characteristic information related to concentration polarization. By comparing the theoretical diffusion voltage with the concentration polarization component in actual voltage fluctuations, the ion diffusion coefficient was corrected in reverse, effectively reducing the estimation error caused by model simplification and ensuring the reliability of the concentration polarization voltage calculation results.

[0130] Step S1535: Incorporate a multi-parameter dynamic correction mechanism, using temperature and SOC as correction factors, establish temperature dependence functions and SOC adaptation functions for polarization resistance and diffusion coefficient, dynamically adjust parameters based on real-time temperature and SOC data collected during the buffer period, and use the least squares method to fit the voltage and current time series data during the buffer period online to continuously optimize the model coefficients.

[0131] A multi-parameter dynamic correction mechanism is incorporated, using temperature and SOC as core correction factors. Through fitting multi-condition experimental data, temperature dependence functions and SOC adaptation functions for polarization resistance and ion diffusion coefficient are established. Based on real-time temperature and SOC data collected during the buffer period, the core parameters in electrochemical polarization and concentration polarization calculations are dynamically adjusted to adapt the estimation model to different battery states. The least squares method is used to perform online fitting of voltage-current time-series data during the buffer period, continuously optimizing the model coefficients and tracking the dynamic changes in battery polarization state in real time, further improving the accuracy and adaptability of polarization voltage estimation.

[0132] Step S1536: Integrate the calculation results of electrochemical polarization voltage and concentration polarization voltage to obtain the total polarization voltage at the end of the pulse phase. Establish a rule for verifying the validity of the estimation results. By comparing the difference between the total polarization voltage and the measured voltage, ohmic voltage drop, and open circuit voltage in the voltage time series data, the estimation deviation is judged. If the deviation exceeds the preset range, the model parameter recalibration process is automatically started.

[0133] The calculated results of electrochemical polarization voltage and concentration polarization voltage are integrated and directly superimposed to obtain the total polarization voltage at the end of the pulse phase. A validity verification rule for the estimation result is established. The difference between "measured voltage - ohmic voltage drop - open-circuit voltage" is calculated and compared with the estimated total polarization voltage to determine the estimation deviation. If the deviation exceeds a preset range (e.g., ±5%), the model parameter recalibration process is automatically initiated. This re-optimizes the calculation logic of core parameters such as polarization resistance and diffusion coefficient, corrects the model coefficients, and ensures the accuracy of the total polarization voltage estimation result, providing a reliable basis for subsequent impedance compensation.

[0134] Step S1537: The diffusion equation and impedance fitting process are simplified by using the model order reduction method, while retaining the influencing factors. The real-time calculation time is shortened by using the pre-calculated parameter mapping table, ensuring that multiple rounds of iterative estimation are completed within the buffer period, thus adapting to the real-time requirements of the vehicle scenario.

[0135] To adapt to the real-time requirements of automotive scenarios and optimize model computational efficiency, a model order reduction method is adopted to simplify the diffusion equation and impedance spectrum fitting process, eliminate secondary influencing factors, and retain only core parameters and key calculation steps, reducing redundant computation. Through multi-condition experiments, the basic diffusion coefficient and polarization resistance range at different temperatures and SOCs are pre-calculated, and a parameter mapping table is established. During model runtime, the data in the mapping table is directly called for interpolation calculations, significantly shortening real-time computation time. This ensures that the model can complete multiple rounds of iterative estimation within the buffer period, meeting the real-time requirements of impedance compensation during stage switching and adapting to the response speed of the automotive control system.

[0136] Step S154: Implement the virtual impedance compensation algorithm, dynamically generate a compensation voltage signal based on the polarization voltage estimation result, and superimpose it onto the output terminal through the on-board power execution unit at the moment of stage switching. During the compensation process, a feedback adjustment mechanism is adopted to monitor the voltage response after switching in real time. If the voltage fluctuation exceeds the preset range, the compensation amount is automatically corrected.

[0137] A virtual impedance compensation algorithm is implemented, dynamically generating a corresponding compensation voltage signal based on the total polarization voltage output from the polarization voltage estimation model. At the instant of switching between the pulse and constant current / constant voltage stages, the compensation voltage signal is superimposed on the output via the onboard power execution unit (DC / DC converter) to counteract the impact of sudden polarization voltage changes on the switching process. A feedback adjustment mechanism is employed during compensation, monitoring the voltage response data after switching at a high-frequency sampling rate. If the detected voltage fluctuation exceeds a preset allowable range, the amplitude and duration of the compensation voltage signal are automatically corrected to ensure that the voltage stabilizes quickly after switching and to suppress oscillations.

[0138] Step S155: Establish a real-time monitoring and dynamic correction mechanism for the transient process. During the parameter gradual change and impedance compensation execution, collect voltage and current data at a high-frequency sampling rate, calculate the voltage change amplitude and fluctuation frequency. If the change amplitude exceeds the allowable range, immediately adjust the gradual change rate and optimize the compensation algorithm parameters to form a closed-loop control of monitoring, judgment and correction.

[0139] A real-time monitoring and dynamic correction mechanism for the transient process is established. During parameter gradual change and impedance compensation, voltage and current data are collected at a high-frequency sampling rate of no less than 1 kHz. The voltage surge amplitude and fluctuation frequency are calculated in real time, and an allowable voltage surge threshold is set (e.g., ≤50mV). If the detected surge amplitude exceeds this threshold, the parameter gradual change rate is immediately adjusted, such as reducing the pulse amplitude change rate or extending the gradual change time. At the same time, the core parameters of the impedance compensation algorithm are optimized. This forms a closed-loop control system of real-time monitoring, threshold judgment, and parameter correction, continuously suppressing voltage oscillations during the transient phase and ensuring smooth phase switching.

[0140] Step S156: Formulate the switching completion judgment rule, set the voltage stability threshold and duration condition. When the voltage fluctuation amplitude after switching is lower than the preset value and remains stable for a specified time, the transition process is judged to be completed, the parameter gradual change and impedance compensation are automatically terminated, and the conventional control strategy of constant current or constant voltage stage is switched to. If the stability condition is not met, the transition time is extended and the parameters are corrected again.

[0141] Establish rules for determining the completion of the switchover, setting a voltage stability threshold (e.g., fluctuation amplitude ≤ 20mV) and a duration condition (e.g., 2 seconds). After parameter gradation and impedance compensation are executed, monitor the voltage status in real time. When the voltage fluctuation amplitude is lower than the stability threshold and the stable time reaches the specified duration, the transition process is determined to be complete, and the parameter gradation and impedance compensation strategy is automatically terminated, switching to the conventional charging control strategy in the constant current or constant voltage stage. If the stability condition is not met, the transition time is automatically extended, and the gradation rate and compensation amount are adjusted again until the voltage reaches a stable state, ensuring that the stage switchover is completely completed.

[0142] Step S157: Conduct full-condition adaptation optimization. Under different operating conditions, such as temperature, SOC range, and aging stage, test the impact of buffer period duration, gradual change rate, and compensation algorithm parameters on switching stability, record relevant data, establish the mapping relationship between operating conditions and optimization parameters, and form an adaptive parameter library.

[0143] A full-condition adaptation and optimization experiment was conducted, covering different ambient temperatures (-20℃, 0℃, 25℃, 45℃, 60℃), SOC range (0%-100%), and aging stages (new battery, 500 cycles, 1000 cycles, 1500 cycles). Under each condition, the impact of different buffer periods, parameter gradient rates, and compensation algorithm parameters on switching stability was tested, and key data such as voltage surge amplitude and stabilization time were recorded. Based on the test data, a mapping relationship between operating conditions and optimized parameters was established, forming an adaptive parameter library. When actual charging conditions change, the system can call the adapted optimized parameters from the parameter library to ensure effective suppression of voltage surges even in complex scenarios.

[0144] Based on the same inventive concept, please refer to Figure 2 This paper shows a schematic block diagram of a pulse charging curve control system 100 for a vehicle energy storage battery, which is provided in an embodiment of this application for executing the pulse charging curve control method of the above-described vehicle energy storage battery. The pulse charging curve control system 100 for the vehicle energy storage battery may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0145] In this embodiment, both the machine-readable storage medium 120 and the processor 130 are located within the pulse charging curve control system 100 of the vehicle energy storage battery and are separately configured. However, it should be understood that the machine-readable storage medium 120 may also be independent of the pulse charging curve control system 100 of the vehicle energy storage battery and may be accessed by the processor 130 via a bus interface. Alternatively, the machine-readable storage medium 120 may also be integrated into the processor 130 and may communicate with external systems via the communication unit 110.

[0146] The processor 130 is the control center of the pulse charging curve control system 100 for the vehicle energy storage battery. It connects to various parts of the pulse charging curve control system 100 via various interfaces and lines. By running or executing software programs and / or modules stored in the machine-readable storage medium 120, and by calling data stored in the machine-readable storage medium 120, it performs various functions and processes data of the pulse charging curve control system 100, thereby providing overall monitoring of the vehicle energy storage battery pulse charging curve control system 100. Optionally, the processor 130 may include one or more processing cores; for example, the processor 130 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor. The machine-readable storage medium 120 is used to store machine-executable instructions for executing the scheme of this application, and the processor 130 is used to execute the machine-executable instructions stored in the machine-readable storage medium 120 to implement the pulse charging curve control method for vehicle energy storage batteries provided in the aforementioned method embodiments.

[0147] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

[0148] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for controlling the pulse charging curve of an automotive energy storage battery, characterized in that: Includes the following steps: Construct a multi-dimensional sensing network to simultaneously collect temperature, voltage, and electrochemical parameters, and establish a three-domain data mapping relationship; A dynamic, electrically coupled optimization model was established, incorporating temperature effect quantification equations and deep learning to achieve adaptive learning of aging characteristics. Design a thermal and electrical coordinated closed-loop control algorithm, using dual safety thresholds as constraints, and optimize pulse parameters through model prediction to form a closed-loop execution; Construct a multimodal security boundary identification and early warning system, integrating multi-feature extraction and sensing technologies to complete security risk early warning and parameter degradation control; The phase transition mechanism is optimized by using buffer period settings, parameter gradual change strategies, and impedance compensation to suppress voltage abrupt changes during the switching between pulse and constant current / constant voltage phases. The optimized transition mechanism, through buffer period setting, parameter gradual change strategy, and impedance compensation, suppresses voltage surges during the switching between pulse and constant current / constant voltage phases, including: Establish a phase switching prediction and buffer period triggering mechanism. Based on the real-time SOC, voltage, and polarization state of the battery, determine the switching trigger threshold. When the parameters reach the preset warning range before the threshold, the buffer period is automatically started. During the buffer period, the current pulse charging mode is maintained, and relevant data such as voltage fluctuation peak, polarization voltage, and temperature change rate are collected synchronously. The design employs a scenario-specific parameter gradient strategy, with adaptation schemes tailored to the two switching scenarios: pulse to constant current and pulse to constant voltage. When switching from pulse to constant current, the pulse amplitude is linearly reduced at a preset rate, and the duty cycle is simultaneously gradiented to the corresponding target value. During the gradient process, the target current value of the constant current stage is matched in real time. When switching from pulse to constant voltage, the pulse amplitude and duty cycle are adjusted exponentially to gradually bring the average voltage closer to the target voltage of the constant voltage stage. The gradient rate is dynamically adjusted based on the polarization accumulation collected during the buffer period. A polarization voltage estimation model was constructed. Based on the current and voltage time-series data and high-frequency impedance spectrum parameters collected during the buffer period, and combined with the battery electrochemical mechanism, the sum of concentration polarization and electrochemical polarization voltage at the end of the pulse phase was calculated in real time. A virtual impedance compensation algorithm is implemented, which dynamically generates a compensation voltage signal based on the polarization voltage estimation result. At the moment of stage switching, the signal is superimposed on the output terminal through the on-board power execution unit. During the compensation process, a feedback adjustment mechanism is adopted to monitor the voltage response after switching in real time. If the voltage fluctuation exceeds the preset range, the compensation amount is automatically corrected. Establish a real-time monitoring and dynamic correction mechanism for the transient process. During parameter gradual change and impedance compensation, collect voltage and current data at a high-frequency sampling rate, calculate the voltage change amplitude and fluctuation frequency, and immediately adjust the gradual change rate and optimize the compensation algorithm parameters if the change amplitude exceeds the allowable range, forming a closed-loop control of monitoring, judgment and correction. Establish switching completion judgment rules, set voltage stability threshold and duration conditions. When the voltage fluctuation amplitude after switching is lower than the preset value and remains stable for a specified time, the transition process is judged to be complete, the parameter gradual change and impedance compensation are automatically terminated, and the conventional control strategy of constant current or constant voltage stage is switched to. If the stability conditions are not met, the transition time is extended and the parameters are corrected again. Conduct full-condition adaptation optimization. Under different operating conditions, such as temperature, SOC range, and aging stage, test the impact of buffer period duration, gradual change rate, and compensation algorithm parameters on switching stability, record relevant data, establish the mapping relationship between operating conditions and optimization parameters, and form an adaptive parameter library.

2. The pulse charging curve control method for a vehicle energy storage battery according to claim 1, characterized in that: The construction of a multi-dimensional sensing network, which simultaneously collects temperature, voltage, and electrochemical parameters, and establishes a three-domain data mapping relationship, includes: Based on the structural characteristics of the battery cell, micro-nano-level temperature sensing elements, flexible voltage sensors and high-frequency impedance spectroscopy modules are selected. Through customized adaptation design, the device's environmental adaptability, response performance and vehicle system compatibility are taken into account to meet the requirements for accurate acquisition of multi-dimensional parameters. Various sensors and modules are deployed according to the location of each battery cell. Through minimally invasive implantation, bonding and fixation, interface connection and electromagnetic shielding design, both installation reliability and structural integrity are taken into account. A multi-channel synchronous acquisition system with a high-performance control unit as its core was built, and a signal conditioning circuit was configured. Through a unified synchronization mechanism and an adapted sampling frequency, the synchronization and stability of the three-domain parameter acquisition were ensured. The raw data is preprocessed in real time. Digital filtering algorithms are used to remove high-frequency noise from the voltage signal and random fluctuations from the temperature signal. Phase correction and amplitude calibration are performed on the impedance spectrum scanning data. A sensor calibration system is established. The temperature sensor is calibrated at multiple points through the calibration platform. The voltage sensor is calibrated under different loads. The impedance spectrum module is calibrated with standard impedance components. Calibration coefficients are stored for real-time data correction. Data validity judgment logic is designed. Abnormal data is removed through threshold judgment and trend analysis. Characteristic parameters of the thermal, electrical, and electrochemical domains are defined. Based on the battery electrochemical mechanism, the intrinsic correlation between these parameters is analyzed. Sample data is collected through multi-condition experiments, and the three types of parameter data are recorded simultaneously under different temperature ranges, SOC ranges, and aging stages to construct a sample database. A mapping model is built using machine learning algorithms, with thermal and electrical domain parameters as inputs and electrochemical parameters as outputs. The model is trained and optimized using sample data. An online model update mechanism is designed to monitor the model prediction error in real time. When the error exceeds a preset threshold, the latest collected valid data is automatically used to fine-tune the model. By employing a high-speed anti-interference transmission interface, a hierarchical storage strategy, and an optimized compression algorithm, the system achieves efficient transmission and reasonable storage of sensing data, adapting to the resource limitations of the vehicle system.

3. The pulse charging curve control method for a vehicle energy storage battery according to claim 1, characterized in that: The establishment of a dynamic, electrically coupled optimization model, incorporating a temperature influence quantification equation and deep learning, to achieve adaptive learning of aging characteristics includes: A customized data transmission architecture is designed, a high-speed anti-interference bus interface adapted to the vehicle environment is selected, and point-to-point communication between the sensing module and the vehicle BMS is achieved by combining bus protocol optimization. Priority levels are divided according to the real-time data requirements, high bandwidth is allocated to core control parameters to ensure real-time transmission, and batch transmission mode is adopted for non-critical information. A hierarchical storage system is built. The local storage uses high-reliability flash memory to cache recently collected raw data, preprocessed results and model output data. The historical data of the entire life cycle is uploaded to the cloud distributed database through vehicle networking technology. Appropriate compression algorithms are selected for different types of data characteristics. Lossless compression algorithms are used for numerical parameters and hierarchical compression mode is used for process data with high redundancy. Data verification mechanisms are set up simultaneously to ensure data integrity. To enhance the anti-interference and fault tolerance capabilities of transmission and storage, shielded cables and differential transmission design are adopted for the transmission link, and data backup and fault tolerance mechanisms are introduced in the storage stage. Based on the pseudo-two-dimensional electrochemical mechanism model, the correlation between battery electrochemical parameters and temperature is sorted out, a temperature-dependent parameter quantification equation is constructed, and the influence of temperature change on pulse charging response is transformed into model correction terms, forming a basic thermo-electric coupling mechanism model. The intrinsic relationship between battery aging characteristics and thermal-electric coupling properties was analyzed, aging state evaluation index was defined, and thermal and electrical parameter data under different aging stages, temperature ranges and SOC ranges were collected through accelerated aging experiments to construct a multi-condition sample database covering the entire life cycle. A deep learning network model with LSTM network as the core is built. The thermal and electrical parameters are used as inputs, and the output is the prediction result of thermal-electric coupling characteristics under battery aging. The network is trained and optimized through sample database to establish the mapping relationship between aging characteristics and thermal-electric coupling parameters. This mapping relationship is then embedded into the basic thermal-electric coupling mechanism model to form a dynamic thermal-electric coupling optimization model that integrates mechanism and data. The design of the online model update mechanism monitors the deviation between the model's predicted output and the actual collected data in real time, sets the deviation threshold and update cycle, and automatically calls the latest valid collected data to fine-tune the parameters of the deep learning network when the deviation exceeds the threshold or reaches the preset number of cycles, and synchronously updates the coefficients of the temperature-dependent parameter quantization equation. The computational complexity of the dynamic thermo-electric coupling optimization model is optimized by simplifying the model algorithm, and redundant calculation steps are reduced while ensuring prediction accuracy.

4. The pulse charging curve control method for a vehicle energy storage battery according to claim 3, characterized in that: The process involves constructing a deep learning network model with an LSTM network as its core. Taking thermal and electrical parameter data as input, the model outputs predictions of the thermal-electric coupling characteristics under battery aging conditions. The network is trained and optimized using a sample database to establish a mapping relationship between aging characteristics and thermal-electric coupling parameters. This mapping relationship is then embedded into a basic thermal-electric coupling mechanism model, forming a dynamic thermal-electric coupling optimization model that integrates mechanism and data-driven approaches. This includes: The dimensions of the thermal and electrical input parameters are determined. Temperature and temperature rise rate are selected as thermal parameters, and voltage, current and polarization voltage are selected as electrical parameters. The temporal characteristics of the parameters are sorted out, and the original data is divided into sliding window segments according to a fixed time window to form a time-series data sample that meets the input requirements of the LSTM network. At the same time, the output parameters are set as thermal-electric coupling characteristic indicators such as temperature-dependent polarization resistance, double layer capacitance and electrode reaction rate constant. Design an LSTM network architecture to match the input layer dimension with the total number of thermal and electrical parameters. Set up multiple hidden layers, with the number of neurons in each layer adaptively adjusted according to the sample data scale. The hidden layers use the ReLU activation function, and the output layer uses the linear activation function. Introduce Dropout layer and L2 regularization mechanism, and configure batch normalization layer. Data preprocessing is performed on the multi-condition sample database. The input and output parameters are normalized using the Z-Score standardization method. The training set, validation set, and test set are divided according to a preset ratio to ensure the consistency of the dataset distribution. Configure network training parameters, select the Adam optimizer, set the initial learning rate and dynamically adjust it using a cosine annealing strategy, use mean squared error as the loss function, set the maximum number of training iterations and an early stopping strategy, and terminate training and save the optimal network parameters when the validation set loss function does not decrease for a preset number of consecutive rounds. Start the network training process, input the preprocessed training set data into the LSTM network, calculate the predicted output through forward propagation, update the network weights and biases through back propagation, evaluate the model's generalization ability and adjust the hyperparameters using the validation set data, and verify the model's prediction accuracy using the test set data after training is completed. Extract the parameters of the trained LSTM network and construct a mapping model between aging characteristics and thermo-electric coupling parameters. This model takes real-time collected thermal and electrical time-series data as input and outputs the predicted values ​​of coupling characteristic parameters under the current aging state. Define the interface specification between this mapping model and the basic pseudo-two-dimensional electrochemical mechanism model, and clarify the correspondence between the LSTM output parameters and the temperature-dependent parameters in the P2D model. The mapping relationship model is embedded into the basic thermo-electric coupling mechanism model, and the parameter calling logic of the P2D model is modified so that the output results of the LSTM mapping model can be called in real time through the interface during the model runtime. This output is used as the dynamic input of the conductivity, ion diffusion coefficient, and reaction rate constant parameters in the P2D model, forming a dynamic thermo-electric coupling optimization model that integrates mechanism constraints and data-driven prediction. The fused model was jointly debugged. Under different operating conditions such as temperature, SOC, and aging stage, the coupling characteristic parameters predicted by the model were compared with the actual test data. The matching coefficient between the LSTM network output and the P2D model parameters was corrected, and the prediction variance was reduced by adopting the model integration strategy.

5. The pulse charging curve control method for a vehicle energy storage battery according to claim 1, characterized in that: The aforementioned thermal and electrical coordinated closed-loop control algorithm, constrained by dual safety thresholds, optimizes pulse parameters through model prediction and forms a closed-loop execution, including: The dual safety thresholds consisting of quantified thermal safety constraints and electrochemical safety constraints are dynamically adapted. Combining the battery SOC range, aging degree and ambient temperature, the upper limit of single cell temperature, upper limit of temperature rise rate, lower limit of negative electrode potential and upper limit of SEI film impedance change rate are defined. Threshold dynamic adjustment rules are established, and the threshold boundaries are tightened in the high SOC stage, late aging stage or extreme temperature environment, and the threshold boundaries are appropriately relaxed in the low SOC stage. The pulse parameter optimization variables and multi-objective functions are determined. The optimization variables include pulse amplitude, duty cycle and frequency. The objective functions cover charging speed, energy efficiency and battery life. The priority of each objective is dynamically allocated by weighting coefficients and the weight ratio of each objective is adaptively adjusted under different operating conditions. The dynamic thermal-electric coupling optimization model is integrated into the model predictive control framework. The prediction time domain and control time domain are set. Based on the real-time collected thermal, electrical and electrochemical parameters, the correlation between the battery state change trend and pulse parameters within a preset time period is predicted. A constraint matrix is ​​constructed to transform the dual safety thresholds into inequality constraints of MPC, while incorporating pulse parameter physical constraints to form a multi-dimensional constraint set. The pulse parameter physical constraints include pulse amplitude not exceeding the maximum allowable current of the battery, frequency matching the switching capability of power devices, and duty cycle being within a reasonable range. A fast rolling optimization solution mechanism is designed, which uses a quadratic programming solver to solve the objective function of MPC in real time. In each control cycle, the current optimal pulse parameter combination is output. Only the parameter instructions for the first control step are executed. In the next control cycle, the optimization calculation is restarted based on the latest feedback battery state data, realizing the rolling iteration of prediction, solution, execution and feedback. A closed-loop execution and state feedback link is established, and the optimized pulse parameters are sent to the vehicle power execution unit through a high-speed bus to drive it to output the corresponding pulse charging waveform. At the same time, the battery temperature, voltage, current and electrochemical parameters during the charging process are collected in real time through a multi-dimensional sensing network and fed back to the MPC framework and the thermo-electric coupling model to update the model state and constraint boundary. A threshold-triggered emergency adjustment strategy is formulated. When the feedback data shows that the battery status has reached a preset proportion of the safety threshold, the pre-adjustment mechanism is activated and the pulse parameters are slightly corrected. If the battery status is detected to have exceeded the safety threshold, an emergency adjustment command is immediately triggered to adjust the pulse amplitude or frequency until the battery status returns to the safe range. To optimize the real-time performance of the algorithm, the computation time of MPC is reduced by model order reduction, pre-calculation of optimization intervals, and simplification of solution complexity. A parallel computing architecture is adopted to allocate data processing and optimization tasks, ensuring that the response time of control commands matches the high-frequency characteristics of pulse charging.

6. The pulse charging curve control method for a vehicle energy storage battery according to claim 1, characterized in that: The aforementioned construction of a multimodal security boundary identification and early warning system integrates multi-feature extraction and sensing technologies to achieve security risk early warning and parameter degradation control, including: By integrating voltage sensing, high-frequency impedance spectrum, ultrasonic sensing and temperature sensing modules, a multi-source data synchronous acquisition mechanism is established. Through a unified synchronous trigger signal, voltage fluctuation data, broadband impedance spectrum data, ultrasonic propagation signal and temperature data during battery charging are acquired in real time to ensure the consistency of timestamps for different modal data. Fluctuation features were extracted from voltage data, parameters related to lithium deposition and SEI film damage were analyzed from impedance spectroscopy data, and features characterizing internal stress accumulation and SEI film integrity were extracted from ultrasonic signals. The extracted features were normalized and redundantly removed, and features strongly correlated with safety risks were retained. A dynamic safety boundary identification model was constructed, which was divided into four identification dimensions according to the type of safety risk: voltage anomaly, lithium deposition, SEI film damage, and electrode stress accumulation. Based on the battery electrochemical mechanism and multi-condition experimental data, the mapping relationship between each feature and safety risk was established. The boundary threshold was dynamically adjusted in combination with SOC, temperature, and aging degree. The boundary recognizer was trained and the model parameters were optimized using machine learning algorithms. Design a multi-level risk assessment system, divide risk levels and clarify the corresponding characteristic threshold ranges and judgment conditions for each level. When a single modal feature reaches the warning threshold, multi-modal feature cross-validation is initiated. Set continuous monitoring cycle judgment rules, trigger warnings only when risk conditions are met within a continuous preset cycle, and output differentiated warning signals according to the risk level. Develop a parameter downgrade control strategy that links risk levels, establish a correspondence between warning levels and pulse parameter adjustment ranges, make small adjustments to pulse parameters for low-risk levels, make large adjustments to pulse parameters and limit charging current peak values ​​for medium- and high-risk levels, provide real-time feedback of battery status data during the downgrade adjustment process, and dynamically correct the parameter adjustment range. A closed-loop verification and adaptive optimization mechanism was established for the system. Tests were conducted under different operating conditions, including different temperatures, SOC ranges, and aging stages, to verify the accuracy of safety boundary identification, early warning response speed, and the effectiveness of degradation control. An online model update module was designed to monitor identification errors and false alarm rates in real time. When the error exceeds a preset threshold, the latest collected operating condition data is automatically used to fine-tune the boundary model and feature weights.

7. The pulse charging curve control method for a vehicle energy storage battery according to claim 6, characterized in that: The proposed dynamic safety boundary identification model categorizes safety risks into four dimensions: voltage anomaly, lithium deposition, SEI film damage, and electrode stress accumulation. Based on battery electrochemical mechanisms and multi-condition experimental data, it establishes a mapping relationship between each feature and safety risk. Boundary thresholds are dynamically adjusted in conjunction with SOC, temperature, and aging degree. Machine learning algorithms are used to train the boundary recognizer and optimize model parameters, including: For the four safety risk dimensions of voltage anomaly, lithium deposition, SEI film damage, and electrode stress accumulation, strongly correlated characteristic indicators are identified to form a characteristic system for each risk dimension. Under different SOC ranges, ambient temperatures, and aging stages, various charging scenarios are simulated, and multimodal data such as voltage, impedance spectrum, ultrasound, and temperature are collected simultaneously. The internal state information of the battery is obtained through professional testing methods, and the collected data is labeled with safety risk levels to build a labeled sample dataset covering all usage scenarios. Based on the analysis of electrochemical mechanisms, the correlation between features and safety risks is clarified. Then, statistical modeling methods are used to quantify the correlation between features and risk levels, features are screened, and a quantitative mapping model between features and risk levels is established through appropriate modeling methods to clarify the risk probability distribution corresponding to different feature combinations. Using SOC, temperature, and aging degree as threshold adjustment factors, a dynamic adjustment function for each characteristic threshold is established based on multi-condition experimental data to achieve dynamic adaptation of the threshold to the battery state. By combining feature dimensions and data scale, an appropriate machine learning algorithm is selected to build a boundary recognizer. The labeled feature data is divided into training set, validation set and test set. The features are standardized and the algorithm hyperparameters are optimized using appropriate methods. The risk level corresponding to the dynamic threshold is used as the model output label. The model is trained using the training set, the model performance is monitored using the validation set and an early stopping strategy is adopted, and the model recognition performance is evaluated using the test set. If the preset performance requirements are not met, the feature selection process is backtracked for iterative optimization. A weighted fusion strategy is adopted to integrate multiple features of the same risk dimension. The weights are assigned based on the correlation coefficient between the feature and the risk. When a single feature triggers the warning threshold, cross-feature cross-validation is initiated to establish feature conflict adjudication rules. The system records the deviation between the model's recognition results and the actual state of the battery in real time, calculates the recognition error of each risk dimension, sets an error threshold, and automatically selects the latest collected valid samples to incrementally train the model when the error reaches the threshold, updating the feature weights, mapping parameters, and coefficients of the dynamic threshold adjustment function.

8. The pulse charging curve control method for a vehicle energy storage battery according to claim 1, characterized in that: The constructed polarization voltage estimation model, based on current and voltage time-series data and high-frequency impedance spectrum parameters collected during the buffer period, combined with the battery electrochemical mechanism, calculates in real time the sum of concentration polarization and electrochemical polarization voltage at the end of the pulse phase, including: The input parameter system for polarization voltage estimation is sorted out, and the input data includes the current time series, voltage time series, and high-frequency impedance spectrum raw data during the buffer period. The nominal structural parameters and real-time status parameters of the battery are synchronously correlated. All input data are timestamped and aligned. Digital filtering algorithms are used to remove high-frequency noise in the current and voltage signals. The impedance spectrum data is smoothed to eliminate measurement fluctuations. An equivalent circuit model was used to fit the broadband impedance spectrum data, and the relevant parameters of ohmic resistance, charge transfer polarization resistance and diffusion impedance were separated. The polarization resistance value was dynamically updated by the real-time impedance spectrum fitting results, and the correlation between characteristic frequency and ion diffusion coefficient was established. A sub-model for calculating electrochemical polarization voltage is constructed. Based on the charge transfer mechanism, the polarization resistance is updated in real time as a parameter. Combined with the instantaneous current value during the buffer period, the basic value of electrochemical polarization voltage is calculated through the Ohm's law derived formula. A current change rate correction term is introduced, and the polarization resistance weighting coefficient is dynamically adjusted to compensate for the influence of current mutation on electrochemical polarization. A concentration polarization voltage calculation sub-model was built. Based on Fick's first law and the simplified diffusion equation, the ion diffusion coefficient derived from impedance spectroscopy was used as the core. Combined with the nominal structural parameters of the battery, a quantitative relationship between concentration polarization and current, diffusion coefficient and real-time state parameters was established. The pulse peak-valley difference in the voltage time series data during the buffer period was used to extract the concentration polarization related features and to correct the diffusion coefficient in reverse. A multi-parameter dynamic correction mechanism is incorporated, with temperature and SOC as correction factors. Temperature dependence functions and SOC adaptation functions of polarization resistance and diffusion coefficient are established. Parameters are dynamically adjusted based on real-time temperature and SOC data collected during the buffer period. The least squares method is used to fit the voltage and current time series data during the buffer period online and continuously optimize the model coefficients. By integrating the calculation results of electrochemical polarization voltage and concentration polarization voltage, the total polarization voltage at the end of the pulse phase is obtained. A validity verification rule for the estimation result is established. The estimation deviation is judged by comparing the difference between the total polarization voltage and the measured voltage, ohmic voltage drop and open circuit voltage in the voltage time series data. If the deviation exceeds the preset range, the model parameter recalibration process is automatically started. A model reduction method is adopted to simplify the diffusion equation and impedance fitting process, retain the influencing factors, and shorten the real-time calculation time by pre-calculating the parameter mapping table, ensuring that multiple rounds of iterative estimation are completed within the buffer period, thus adapting to the real-time requirements of vehicle scenarios.

9. A pulse charging curve control system for an automotive energy storage battery, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the pulse charging curve control method for an automotive energy storage battery according to any one of claims 1 to 8 by executing the machine-executable instructions.

Images