A sodium-ion battery charging optimization method and system

Abstract

The application discloses a sodium ion battery charging optimization method and system, and relates to the technical field of sodium ion battery energy storage detection. + The application comprises the following steps: collecting charging data of the sodium ion battery at a current sampling point in real time, obtaining an estimated value of the potential of the negative electrode of the current sampling point relative to Na / Na based on the charging data of the current sampling point, determining that there is a sodium precipitation risk when the estimated value of the potential is less than a risk threshold, performing charging optimization in the constant current stage and the constant voltage stage of the sodium ion battery when it is determined that there is a sodium precipitation risk, continuously reducing the charging current according to a closed-loop feedback algorithm in the constant current stage until the estimated value of the potential is not lower than the risk threshold, and reducing the constant voltage target or shortening the constant voltage duration according to the closed-loop feedback algorithm in the constant voltage stage until the estimated value of the potential is not lower than the risk threshold. The application can accurately describe the complex polarization behavior under a large rate without obtaining a large amount of labeled data with a reference electrode, and realizes the dynamic balance between safety and charging speed.

Description

Technical Field

[0001] This invention relates to the field of sodium-ion battery energy storage technology, and in particular to a method and system for optimizing sodium-ion battery charging. Background Technology

[0002] Sodium-ion batteries are considered a strong complement to lithium-ion batteries due to their abundant resources and low cost. However, their negative electrode is prone to sodium deposition (sodium precipitation) under fast charging or high-voltage conditions, which can induce capacity decay and safety risks. Therefore, it is necessary to optimize the charging process under sodium precipitation conditions.

[0003] Current charging management generally adopts the constant current-constant voltage (CC-CV) strategy used in the lithium battery field, avoiding sodium deposition by setting a fixed cutoff voltage or charging rate. However, the electrochemical window and polarization characteristics of the sodium system differ significantly from those of the lithium system. Simply using a fixed threshold often fails to balance fast charging efficiency and safety, and the original safety margin may be rapidly weakened after temperature fluctuations and cycle aging. Domestic and international academic communities have attempted to add a reference electrode to a single cell under laboratory conditions to directly measure the relative sodium content of the negative electrode. + The potential of the Na reference electrode is used to accurately determine the sodium deposition boundary: the negative electrode potential is close to 0V. At this point, the thermodynamic conditions for sodium deposition are met. If the charging current continues to be applied, sodium ions will be reduced and deposited as metallic sodium on the negative electrode surface at the electrolyte interface. Among these conditions, " "Sodium / sodium ion reference electrode" is a reference calibration method for electrochemical potential. However, this method is difficult and costly to implement in mass-produced cells and modules, and the additional leads can damage the integrity of the battery packaging, so it has not yet been commercialized. The mainstream battery management system (BMS) in the industry relies more on terminal voltage and coulomb integration to estimate the state of charge (SOC). It can only passively limit current or stop charging when the full cell voltage is close to the set limit, and cannot detect the dynamic changes of the negative electrode potential in advance, resulting in the risk of sudden sodium deposition during fast charging. With the development of artificial intelligence technology, some studies have begun to use data-driven models to estimate the battery SOC or state of health (SOH) to improve the accuracy and intelligence of battery management. However, SOC and SOH estimates mainly reflect the overall capacity change and degradation trend of the battery, and cannot directly reveal the electrochemical state of the single electrode, especially the transient potential change of the negative electrode during charging.

[0004] With the development of artificial intelligence technology, some studies have begun to use data-driven models to predict State of Charge (SOC) or State of Harshness (SOH), such as pure black-box models and pure physical models. However, pure black-box models require a large amount of labeled data with reference electrodes, which is extremely costly to obtain; while pure physical models have good interpretability, they are difficult to accurately describe the complex polarization behavior at high charging rates. Therefore, there is still a lack of practical solutions for online prediction of the negative electrode potential, making it impossible to optimize the charging process online and balance fast charging efficiency and safety. Summary of the Invention

[0005] In view of the defects of the prior art, the present invention provides a sodium-ion battery charging optimization method and system, which solves the existing problems.

[0006] The present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for optimizing the charging of sodium-ion batteries, comprising the following steps:

[0008] Real-time acquisition of charging data of sodium-ion batteries at the current sampling point, including current and temperature;

[0009] The current at the current sampling point is input into the charge conservation equation and a forward Euler update is performed to obtain the negative electrode charge state at the current sampling point. The negative electrode charge state at the current sampling point is then input into the negative electrode open-circuit voltage-charge state mapping equation for fitting to obtain the theoretical potential of the negative electrode at the current sampling point. Feature extraction and feature mapping are performed on the vector composed of the current, temperature, current change rate, and negative electrode charge state at the current sampling point to obtain the negative electrode polarization potential. The theoretical potential of the negative electrode at the current sampling point is added to the negative electrode polarization potential to obtain the negative electrode potential relative to Na at the current sampling point. + / Na potential estimate;

[0010] When the estimated potential value is less than the risk threshold, it is determined that there is a risk of sodium deposition; wherein, the risk threshold is constructed by superimposing a zero-volt reference and a dynamic safety margin, and the dynamic safety margin is constructed by real-time temperature and battery health status;

[0011] When a risk of sodium deposition is identified, charging optimization is performed during the constant current and constant voltage phases of the sodium-ion battery. Specifically, during the constant current phase, the charging current is continuously reduced using a closed-loop feedback algorithm until the estimated potential value is not lower than the risk threshold. During the constant voltage phase, the constant voltage target is lowered or the constant voltage duration is shortened using a closed-loop feedback algorithm until the estimated potential value is not lower than the risk threshold.

[0012] Preferably, the negative pole of the current sampling point relative to Na is obtained through a pre-trained physical information neural network model. + The potential estimate of / Na, the physical information neural network model includes an explicit physical layer and a learnable polarization subnetwork;

[0013] The physical layer includes the positive open-circuit voltage-state-of-charge mapping equation, the negative open-circuit voltage-state-of-charge mapping equation, and the charge conservation equation; the specific positive open-circuit voltage-state-of-charge mapping equation and the negative open-circuit voltage-state-of-charge mapping equation are shown below:

[0014] ;

[0015] ;

[0016] In the formula, This is the open-circuit voltage of the positive terminal. This is the open-circuit voltage of the negative terminal, i.e., the theoretical potential of the negative terminal. and These represent the positive and negative states of charge, respectively. , , , , , , , and This is the open circuit factor;

[0017] The charge conservation equation is as follows:

[0018] ;

[0019] In the formula, for t The initial value of the difference in the charge state of the two electrodes when = 0. t For the first t One sampling point, For equivalent reversible capacity, It is Faraday's constant. For current, It is a time variable;

[0020] The learnable polarization subnetwork includes an input layer, multiple hidden layers, and an output layer. The multiple hidden layers extract features from the vector composed of current, temperature, first derivative of current, and negative electrode charge state at the current sampling point. The output layer performs feature mapping on the extracted features to obtain the negative electrode polarization potential.

[0021] Preferably, the charging data also includes the full battery terminal voltage, and the pre-training of the physical information neural network model includes the following steps:

[0022] Collect charging data of sodium-ion batteries at different time points and the corresponding negative electrode relative to Na. + The actual potential value of / Na;

[0023] The charging data at different time points are input into the physical information neural network model to obtain the corresponding potential estimate, the full battery voltage reconstruction, and the negative electrode state of charge change rate.

[0024] The loss function is constructed using the potential estimate, the actual potential value, the reconstructed full cell voltage, the full cell terminal voltage, and the rate of change of the negative electrode state of charge.

[0025] The explicit physical layer parameters are frozen, and the learnable polarimetric subnetwork is trained based on the loss function to obtain a pre-trained physical information neural network model.

[0026] Preferably, the risk threshold is as follows:

[0027] ;

[0028] ;

[0029] In the formula, As a risk threshold, Zero-volt reference, For dynamic safety margin, For temperature sensitivity coefficient, The aging sensitivity coefficient, It indicates a healthy state.

[0030] Preferably, during the constant current phase, the charging current is continuously reduced according to the closed-loop feedback algorithm, specifically including the following steps:

[0031] Obtain the estimated negative electrode potential value of the current sampling point. and risk threshold And calculate the deviation between the two. ;

[0032] If there is a deviation If the value is greater than 0, the current is corrected using the current correction expression to obtain the current correction amount. The specific current correction expression is as follows:

[0033] ;

[0034] In the formula, For the first t Current correction at each sampling point This is the proportionality coefficient. The integral coefficient is... The time variable is At that time, the difference between the estimated negative electrode potential and the risk threshold;

[0035] Apply a limit to the current correction amount:

[0036] ;

[0037] In the formula, This is the current charging current;

[0038] The charging current is corrected based on the current correction amount to obtain the new charging current:

[0039] ;

[0040] In the formula, For the new charging current.

[0041] Preferably, the step of reducing the constant pressure target or shortening the constant pressure duration using the closed-loop feedback algorithm specifically includes the following steps:

[0042] A constant voltage charging test was performed on the sodium-ion battery at a rate of 0.2C within the range of 10%–100% SOC and -10℃–55℃, and the results were recorded. and The curve;

[0043] Piecewise linear regression was used to obtain the local slope for each curve. SOC was interpolated with a step size of 5% and temperature with a step size of 10℃ to obtain the negative electrode potential-full cell voltage slope model.

[0044] Obtain the estimated value of the negative electrode potential and risk threshold And calculate the amount of lift between the two. ;

[0045] Based on real-time SOC and temperature, the negative electrode potential-full cell voltage slope model is queried to obtain the current slope. b Based on the lift and current slope b Obtain the target reduction;

[0046] Obtain a new constant pressure target based on the target reduction. At the same time, the duration of constant pressure is shortened by a proportional factor.

[0047] Preferred options also include:

[0048] If the estimated negative electrode potential fails to rise above the risk threshold in three consecutive iterations during the constant current or constant voltage phase, immediately send a stop charging command to the converter, cut off the charging circuit, and set the system status to standby mode.

[0049] During the waiting mode, if a detection occurs within a continuous time period If so, the restart process will begin.

[0050] In a second aspect, the present invention provides a sodium-ion battery charging optimization system, comprising:

[0051] The acquisition module is used to acquire charging data of sodium-ion batteries at the current sampling point in real time. The charging data includes current and temperature.

[0052] The estimation module is used to input the current at the current sampling point into the charge conservation equation and perform a forward Euler update to obtain the negative electrode charge state at the current sampling point. It then inputs the negative electrode charge state at the current sampling point into the negative electrode open-circuit voltage-charge state mapping equation for fitting to obtain the theoretical potential of the negative electrode at the current sampling point. Feature extraction and feature mapping are performed on the vector composed of the current, temperature, current change rate, and negative electrode charge state at the current sampling point to obtain the negative electrode polarization potential. Finally, the theoretical potential of the negative electrode at the current sampling point is added to the negative electrode polarization potential to obtain the negative electrode potential relative to Na at the current sampling point. + / Na potential estimate;

[0053] The judgment module is used to determine the presence of sodium deposition risk when the estimated potential value is less than the risk threshold; wherein, the risk threshold is constructed by superimposing a zero-volt reference and a dynamic safety margin, and the dynamic safety margin is constructed by real-time temperature and battery health status.

[0054] The optimization module is used to optimize charging during the constant current and constant voltage phases of the sodium-ion battery when a risk of sodium deposition is determined. Specifically, during the constant current phase, the charging current is continuously reduced according to a closed-loop feedback algorithm until the estimated potential value is not lower than the risk threshold. During the constant voltage phase, the constant voltage target is lowered or the constant voltage duration is shortened according to a closed-loop feedback algorithm until the estimated potential value is not lower than the risk threshold.

[0055] Compared with the prior art, the above-mentioned at least one technical solution adopted by the present invention can achieve the following beneficial effects:

[0056] This invention first collects real-time charging data of the sodium-ion battery at the current sampling point, and then inputs the charging data of the current sampling point into a pre-trained physical information neural network model to obtain the negative electrode relative to the Na at the current sampling point. + The potential estimate of / Na. Specifically, the current at the current sampling point is input into the charge conservation equation to obtain the negative electrode charge state at the current sampling point. The negative electrode charge state at the current sampling point is then input into the negative electrode open-circuit voltage-charge state mapping equation for fitting to obtain the theoretical potential of the negative electrode at the current sampling point. Feature extraction and feature mapping are performed on the vector composed of the current, temperature, current change rate, and negative electrode charge state at the current sampling point to obtain the negative electrode polarization potential. This invention solidifies the electrochemical physical laws such as the positive and negative electrode open-circuit voltage-charge state curves and the charge conservation equation into explicit constraints of the model, and only models the polarization potential. This method does not require obtaining a large amount of labeled data with reference electrodes, and can accurately describe complex polarization behavior at high magnification.

[0057] When a risk of sodium deposition is detected, the charging current is continuously reduced during the constant current phase using a closed-loop feedback algorithm; during the constant voltage phase, the constant voltage target is lowered or the constant voltage duration is shortened using the same algorithm. This collaborative optimization strategy can quickly raise the negative electrode potential to a safe range with minimal current or voltage drop, thereby eliminating the risk of sodium deposition while maximizing fast charging efficiency and achieving a dynamic balance between safety and charging speed. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of a sodium-ion battery charging optimization method according to the present invention;

[0060] Figure 2 This is a flowchart of the closed-loop current regulation process during the constant current stage of the present invention;

[0061] Figure 3 This is a flowchart of the dynamic pressure reduction during the constant pressure stage of this invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] This invention discloses a method for optimizing sodium-ion battery charging. By combining electrochemical conservation relationships with lightweight neural networks, a physical information neural network model is constructed to estimate the negative electrode potential in real time. When the predicted value approaches the sodium deposition threshold, closed-loop fine-tuning of the charging current and target voltage is performed, achieving a dynamic balance between charging safety and efficiency. This solves the problem of existing technologies struggling to balance fast charging and safety. (Refer to...) Figure 1 Specifically, it includes the following steps:

[0064] Step S1: Data Acquisition: Throughout the entire charging process of the sodium-ion battery, continuously record the full battery terminal voltage, charging current, battery temperature, and cycle count at a preset sampling period. The battery temperature is the temperature of the casing or tabs. Write the collected data into a circular buffer and output it to subsequent modules in real time.

[0065] Specifically, throughout the entire charging process, the battery management system synchronously measures the terminal voltage of the entire battery at a fixed sampling period (e.g., 100 ms). Charging current I and battery temperature T The microcontroller's built-in coulomb counter integrates the real-time current to obtain the cumulative coulomb quantity. When the integrated quantity reaches 100% of the rated capacity, the cycle count is automatically recorded. The measured signals are then sampled by an ADC, digitally filtered, and corrected for temperature drift to form a standard data frame containing a timestamp, voltage, current, temperature, and cycle count.

[0066] To support the real-time performance of subsequent algorithms, the system establishes a circular buffer in RAM to cache the most recent data frames (e.g., 2048), and triggers a "new data" event upon completion of each frame write. When the buffer write pointer is about to overwrite the read pointer, the oldest data is automatically discarded, ensuring that the algorithm always uses the latest measurement results.

[0067] Step S2: Negative electrode potential estimation: Input real-time data into the Physics-Informed Neural Network (PINN) model.

[0068] The PINN model consists of two parts: an explicit physics layer and a learnable polarization subnetwork. The explicit physics layer, which is not involved in training, includes the positive electrode open-circuit voltage-state-of-charge mapping equation, the negative electrode open-circuit voltage-state-of-charge mapping equation, and the charge conservation equation. These are determined by the battery material properties and fundamental electrochemical laws, and have explicit physical constraints. Therefore, the positive electrode open-circuit voltage-state-of-charge mapping equation, the negative electrode open-circuit voltage-state-of-charge mapping equation, and the charge conservation equation need to be fixed to ensure that the network output always conforms to the basic electrochemical principles. Only the polarization potential is learned in a lightweight manner, and the output is the negative electrode relative to Na at the current sampling point. + Potential estimate of the / Na reference electrode. Single inference delay not exceeding 1ms, estimation error not exceeding 20mV.

[0069] Real-time data frames are read from a buffer and then fed into PINN. Before entering PINN, the data undergoes buffering and filtering to ensure stability and accuracy: voltage and current reflect the battery terminal state and are used in the model loss function calculation; temperature is used to compensate for the open circuit voltage (OCV) curve and dynamic safety threshold; cycle count reflects the battery aging level and is used to adjust the safety margin and assist the model in adapting to aging characteristics. Subsequently, these standardized signals enter the explicit physical layer of PINN to calculate the theoretical negative electrode potential under non-polarized conditions. This value is then superimposed with the output of the learnable polarization subnetwork to obtain the estimated actual negative electrode potential. Through this process, PINN simultaneously integrates physical laws and operational data, ensuring that the estimation results conform to electrochemical principles while possessing good real-time performance and adaptability.

[0070] Specifically, the positive open-circuit voltage-state-of-charge mapping equation is approximated by a piecewise cubic polynomial:

[0071] (1);

[0072] The negative open-circuit voltage-state-of-charge mapping equation uses a hybrid approximation of hyperbolic tangent and linear terms:

[0073] (2);

[0074] in, This is the open-circuit voltage of the positive terminal. This is the open-circuit voltage of the negative terminal, i.e., the theoretical potential of the negative terminal. and These represent the positive and negative states of charge, respectively. and ,coefficient , , , , , , , and It is the open-circuit coefficient, which is calibrated by the open-circuit voltage of the half-cell and then fixed in the firmware.

[0075] The charge conservation equation can be approximated using a single particle as follows:

[0076] (3);

[0077] In the formula, For equivalent reversible capacity, for t The initial value of the difference in the charge state of the two electrodes when = 0. Faraday constant (96485 C·mol) -1), For current (positive value for charging). The time variable is used. The BMS performs a forward Euler update on equation (3) once in each inference cycle:

[0078] (4);

[0079] (5);

[0080] in, For the first k The state of charge of the negative electrode at each sampling point For the first k The state of charge at the positive electrode of each sampling point For the first k The current at each sampling point The inference cycle represents the time interval between each forward computation performed by the model; here, it is taken as 0.1 seconds. i This is the time step index, used to represent each sampling point between the start of charging and the current sampling point 𝑘.

[0081] The model can learn polarimetric subnetworks for training, and in this layer, the model only learns about polarization potentials. Modeling. Its input vector is ,in From the first-order difference The output is the polarization potential. The current and temperature in the data frame are directly used as input, and the rate of change of current... It is calculated by the current difference between two adjacent sampling points. The eigenvectors are obtained recursively from the coulomb quantity and the charge conservation equation. In terms of model output, the polariton network only predicts the polarization potential. It is used to characterize the nonlinear polarization behavior caused by current, temperature and negative electrode filling state during actual charging.

[0082] The learnable polarization subnetwork employs a minimal feedforward fully connected structure, consisting of one input layer, three hidden layers, and one output layer. The specific topology is 4→32→32→16→1, where "4" represents the input dimension and "1" represents the polarization potential. The output is a scalar function with a total of approximately 1800 parameters, and ReLU activation is used after each hidden layer. The extracted features are then mapped through the output layer to obtain the negative polarization potential.

[0083] Ultimately, PINN will use the theoretical negative electrode potential calculated from the explicit physical layer. With polarization potential Adding them together gives a complete estimate of the actual negative electrode potential. The details are as follows:

[0084] (6);

[0085] Simultaneously, the estimated full-cell voltage reconstruction amount can be obtained. :

[0086] (7);

[0087] In the formula, The input vector.

[0088] Model training uses a joint loss function that includes physical constraints:

[0089] (8);

[0090] in, This is the full battery terminal voltage. For a small number of samples with reference electrodes, the labels are... , and As the weight, we can take it here. , The rate of change of the negative electrode's state of charge is represented by the negative electrode's state of charge difference, which is approximated by the difference in negative electrode state of charge between adjacent sampling points. .

[0091] In this invention, the Physical Information Neural Network (PINN) is trained using an "offline pre-training + online incremental learning" approach. In the offline phase, a small amount of experimental data with reference electrodes is first collected as supervisory samples, and combined with a large amount of conventional full-cell operating data to construct a training set containing features such as battery terminal voltage, charge / discharge current, temperature, and capacity. During training, the explicit layers of the network (including the positive and negative electrode OCV–SOC curves and the charge conservation equation) remain frozen, and only the weight parameters of the lightweight polarimetric subnetwork are optimized. The loss function is designed as a joint loss of physical constraints, including full-cell terminal voltage reconstruction error, negative electrode potential supervision error, and SOC–current consistency constraints, ensuring that the model can still satisfy electrochemical consistency under limited data. The optimization algorithm uses Adam, with an initial learning rate set to 1×10⁻⁶. -3 The error gradually decreases with iteration. After offline pre-training and deployment to the battery management system (BMS), the model performs lightweight incremental learning using low-rate or static fragment data every 100 charge-discharge cycles: the explicit physical layer is frozen, only 5 gradient updates are performed on the polarization network, and a residual verification mechanism is used to determine whether to accept the update. This maintains the model's adaptability to complex polarization behavior, avoids parameter drift, and ensures that the negative electrode potential estimation error is always controlled within ±20mV.

[0092] Step S3: Sodium precipitation risk assessment: Set a sodium precipitation risk threshold. ,in" "This is a reference calibration method for electrochemical potential, indicating "relative to a sodium / sodium ion reference electrode". The voltage adaptively adjusts within the 20mV–80mV range as ambient temperature rises or cell aging increases. This adjustment is made when the estimated negative electrode potential... When the negative electrode approaches the thermodynamic conditions for sodium deposition, continuing to charge is likely to produce metallic sodium deposition on the electrode surface, thus generating a sodium deposition risk signal.

[0093] To ensure fast charging efficiency while suppressing sodium deposition, this invention employs a "zero-volt reference + dynamic safety margin" approach for sodium deposition criterion. The system first fixes the thermodynamic reference. Add dynamic safety margin Constitutes a risk threshold . It adapts to two types of state variables: instantaneous temperature T and state of health (SOH) representing the degree of aging. Under the baseline conditions of room temperature (25℃) and a new cell (SOH=100%), The lower limit is set to 20mV; when the temperature deviates from 25℃ or the SOH decreases, the system uses a linear gain model:

[0094] (9);

[0095] in, The temperature sensitivity coefficient, used to describe how sensitive the safety margin is to temperature changes, is set to 0.6 mV·℃. -1 , The aging sensitivity coefficient describes the rate at which the safety margin increases as the cell's health deteriorates, and is set to 40mV. The calculated... After a 2-second first-order low-pass filter, it can track sudden temperature changes while avoiding threshold jitter; subsequently... An interval limit is imposed to ensure that it always falls within the range of 20mV–80mV. The final risk threshold is given by the following formula:

[0096] (10);

[0097] When the real-time negative electrode potential is estimated Less than At that time, the system immediately sets the sodium precipitation risk flag Flag=1, and sends the flag along with the current Temperature and SOH are fed together into the subsequent closed-loop control logic; if If the flag is cleared, charging will continue as originally set. Through a temperature-aging dual-factor dynamic margin and filtering strategy, this step maintains a safe potential redundancy of no less than 20mV throughout the entire lifespan and temperature range, while avoiding reduced charging efficiency due to excessive conservatism.

[0098] Step S4: Closed-loop current adjustment during constant current stage: If the current is in the constant current charging stage and there is a risk of sodium deposition, the proportional-integral algorithm is executed to slightly reduce the charging current to obtain a new current. , The current charging current. Until... And the current drop is minimal; the specific process is shown in the attached diagram. Figure 2 As shown.

[0099] Specifically, in this embodiment, the current control during the constant current stage employs an improved proportional-integral algorithm to effectively suppress the risk of sodium deposition at the negative electrode while ensuring charging efficiency. Its control logic is as follows:

[0100] First, once a sodium precipitation risk indicator is received during the constant current phase, the system initiates closed-loop current regulation logic. Specifically, the controller reads the latest estimated negative electrode potential within a fixed 100ms control cycle. Sodium risk threshold Calculate the instantaneous deviation:

[0101] (11);

[0102] when If a risk of sodium precipitation is detected, the controller initiates flow adjustment calculations.

[0103] The current regulation uses an improved proportional-integral algorithm, and the current correction is:

[0104] (12);

[0105] Among them, the proportionality coefficient Integral coefficient The integration window is fixed at 500ms. This integration window limit ensures that the integration term is only effective within a short time interval, avoiding excessive current suppression caused by long-term accumulation.

[0106] To prevent the current from dropping too much at once, the system applies a limit to the correction amount:

[0107] (13);

[0108] Cut off excessive correction amounts to ensure that the current reduction in a single operation does not exceed 20% of the current charging current, thus avoiding excessive current reduction that could affect charging efficiency.

[0109] get Then, the new charging current is set as follows:

[0110] (14);

[0111] (15);

[0112] Thus satisfying If the next control cycle detects... Then keep the current No further decline; if If risks still exist, continue iterating until the deviation is eliminated or the current drop reaches the preset minimum safe current of 0.2C.

[0113] when After the current returns to above the threshold and remains above it for three cycles, the controller automatically resets the integral term and slowly releases the current limit. It is gradually restored to 1 with an increment of 0.05.

[0114] Step S5: Dynamic voltage reduction during constant voltage stage: If the current stage is constant voltage charging and there is a risk of sodium deposition, the constant voltage target is adjusted according to the negative electrode potential-full cell voltage slope model. Or shorten the duration of constant pressure, so that and satisfy The specific process is as follows: Figure 3 As shown.

[0115] When the charging process enters the constant voltage stage, the current gradually decreases as polarization decays. However, if the polarization component is too large, the negative electrode potential may still continue to rise. To raise the negative electrode potential without significantly extending the charging time, this embodiment introduces a "negative electrode potential – full cell voltage slope model" and uses it to implement a modified constant voltage target. The model is as follows:

[0116] During the offline calibration phase, representative cells were subjected to constant voltage charging tests at a low rate of 0.2C within the range of 10%–100% SOC and -10℃–55℃, and the synchronization curves were recorded. and For each curve, piecewise linear regression is used to obtain the local slope:

[0117] (16);

[0118] And b The values ​​are fixed in the processor in the form of a two-dimensional table. The SOC is interpolated in a step size of 5% and the temperature is interpolated in a step size of 10℃ to obtain the negative electrode potential-full cell voltage slope model.

[0119] During online operation, once the risk of sodium precipitation is detected during the constant pressure phase ( The controller immediately looks up the current slope based on the real-time SOC and temperature. b Let the required lifting amount be:

[0120] (17);

[0121] Then and The corresponding target reduction is:

[0122] (18);

[0123] Therefore, the new constant pressure target is calculated:

[0124] (19);

[0125] At the same time, the duration of constant pressure will be shortened proportionally. ):

[0126] (20);

[0127] Shorten the constant voltage duration to avoid prolonged stay in the high polarization region. Re-evaluate every 100ms thereafter. and :like If the value has exceeded the threshold and remained above it for three cycles, then maintain the current state. Until the current decays to the cutoff value; if If it is still below the threshold, query the slope model again for iterative adjustment. The maximum cumulative decrease does not exceed 4% of the rated constant current voltage. After three consecutive iterations If the circuit still fails to return to the safe range, it is determined that there is a risk of polarization anomaly or thermal runaway, and the process immediately proceeds to step S6, the safe charging stop procedure.

[0128] Step S6: Safe Charging Stop and Restart: If the risk of sodium deposition cannot be eliminated by closed-loop current adjustment during the constant current stage or dynamic voltage reduction during the constant voltage stage, charging should be stopped immediately and the system should enter a waiting mode; during the waiting period, monitoring should be performed at a low sampling frequency. ,when consistently higher After setting the time, the normal charging process will automatically resume.

[0129] When the closed-loop current regulation in the constant current stage or the dynamic voltage reduction in the constant voltage stage fails to estimate the negative electrode potential in three consecutive iterations... Raise to dynamic threshold The above indicates that the current polarization anomaly or local hot spot may pose an unacceptable risk of sodium deposition if charging continues. The controller immediately sends a stop charging command to the converter, disconnects the charging circuit, and sets the system status to "safe waiting." After entering the waiting mode, the sampling period is reduced from 100ms to 2s to reduce the load on the microprocessor unit and bus; only voltage, current, and temperature measurements are retained, and the coulomb integration is frozen in the firmware to prevent invalid integration from affecting the SOC estimation.

[0130] During the waiting period, the control logic... Perform sliding window monitoring: if continuous Internal detection The "safe platform" phenomenon indicates that polarization has been fully released, meeting the restart conditions. To prevent oscillations, the system is then observed for a 30-second freeze period; if within the freeze period... If the device falls into the risk zone again, it will wait for the timer to reset to zero and restart. The restart process can only begin when the entire freeze period ends and the negative electrode potential remains above the threshold.

[0131] Upon restart, the controller first resumes constant current charging at a low current of 0.2C, maintaining this for 60 seconds as a "soft start" phase; if during this period... Maintain safety, and the rate of temperature rise. If the current is less than 0.1℃ / min, it will be gradually increased to the original fast charging current in 0.5C increments, and the S4 and S5 closed-loop regulation logic will be reactivated. If during soft start... If the threshold is triggered again, the system immediately reverts to standby mode and lowers the soft-start current limit for the next restart to 0.1C to avoid frequent start-stop cycles. The entire stop-and-restart cycle is allowed to be executed a maximum of three times; if normal charging cannot be restored after three attempts, the BMS records a fault code and reports it to the main controller, requiring manual inspection or switching to a constant-temperature slow charging solution.

[0132] Through this safe process of "stop charging - wait - soft start - recovery", the present invention can fully release electrochemical polarization and heat accumulation under extreme working conditions, prevent continuous deposition of metallic sodium or local overheating, and at the same time use soft start step current to quickly restore the charging progress, taking into account both safety and usage efficiency.

[0133] Step S7: Online self-calibration: Every 100 charge-discharge cycles, low-rate or open-circuit data segments are used to incrementally learn the physical information neural network, and the model parameters are corrected in real time. This ensures that the negative electrode potential estimation error never exceeds 20mV.

[0134] To ensure that the physical information neural network maintains a negative electrode potential estimation accuracy within ±20mV throughout the entire lifespan of the device, the system automatically triggers incremental learning once every 100 charge-discharge cycles. Upon triggering, the controller prioritizes extracting a 5-minute data segment from the 0.2C constant current discharge segment or the 30-minute resting segment between the two most recent charges; this segment has minimal polarization and an approximately open-circuit voltage, and can be considered a "semi-supervised" tag.

[0135] The incremental learning process is executed asynchronously during the microcontroller's idle time slots: First, the precise state of charge is obtained by integrating the segment current according to the conservation equation; then, the segment voltage sequence is fitted with the fixed OCV-SOC curve using least squares to obtain the "true" negative electrode potential of the calibration stage. During the training phase, the OCV curve and the charge-conserving layer were frozen, and Adam optimization was performed 5 times on only about 1.8k weights of the polarization subnetwork (learning rate 1×10). -4 (Batch size 64), the loss function uses the three-term joint physical loss from offline pre-training. Immediately after the update, the average residual is calculated on the independent validation segment; if the residual decreases by ≥2mV compared to before the update, the new weights are written to Flash and activated; otherwise, the old weights are rolled back to ensure network stability. The entire calibration process takes <3 seconds, with no perceptible delay to the main charging process.

[0136] While performing weight correction, the system dynamically adjusts the safety margin based on the temperature within the calibration segment and the difference between the old and new residuals. If the residual decreases significantly at the current temperature, then the value should be appropriately reduced. (Not less than 20mV), conversely, in the temperature range where the residual increases, The voltage increases by 5mV until it caps at 80mV. Utilizing this periodic incremental learning and margin fine-tuning mechanism, the model adaptively tracks internal resistance growth, capacity decay, and temperature drift. Even after 800 cycles of long-term testing, it maintains a negative electrode potential estimation error of ≤19mV, ensuring accurate and reliable subsequent sodium deposition risk assessment.

[0137] Example 2

[0138] Based on the same concept, the present invention also provides a sodium-ion battery charging optimization system, including an operation acquisition module, an estimation module, a judgment module and an optimization module.

[0139] The acquisition module is used to collect real-time charging data of the sodium-ion battery at the current sampling point. The charging data includes current and temperature.

[0140] The estimation module inputs the current at the current sampling point into the charge conservation equation and performs a forward Euler update to obtain the negative electrode charge state at the current sampling point. It then inputs the negative electrode charge state into the negative electrode open-circuit voltage-charge state mapping equation for fitting to obtain the theoretical potential of the negative electrode at the current sampling point. Feature extraction and feature mapping are performed on the vector composed of the current, temperature, current change rate, and negative electrode charge state at the current sampling point to obtain the negative electrode polarization potential. Finally, the theoretical potential of the negative electrode at the current sampling point is added to the negative electrode polarization potential to obtain the negative electrode potential relative to Na at the current sampling point. + The potential estimate of / Na.

[0141] The judgment module is used to determine the presence of sodium deposition risk when the estimated potential value is less than the risk threshold. The risk threshold is constructed by superimposing a zero-volt reference and a dynamic safety margin, which is built by considering the real-time temperature and the health status of the battery.

[0142] The optimization module is used to optimize charging during the constant current and constant voltage phases of the sodium-ion battery when a risk of sodium deposition is identified. Specifically, during the constant current phase, the charging current is continuously reduced according to a closed-loop feedback algorithm until the estimated potential value is not lower than the risk threshold. During the constant voltage phase, the constant voltage target is lowered or the constant voltage duration is shortened according to a closed-loop feedback algorithm until the estimated potential value is not lower than the risk threshold.

[0143] Furthermore, the acquisition module is located at the front end of the battery management system and is used to acquire battery terminal voltage, charging current, temperature, cumulative capacity and cycle count in real time according to a fixed sampling period, and send the data to the bus.

[0144] The acquisition module consists of a high-precision analog front-end and embedded sampling firmware: voltage measurement uses a 16-bit differential-integrating analog-to-digital converter with a 1MΩ / 100kΩ precision voltage divider network; current measurement uses a 0.5mΩ shunt resistor with a zero-drift programmable gain amplifier; temperature signals are acquired by a 10kΩ NTC thermistor attached to the housing and tabs. The microprocessor's built-in coulomb counter integrates the current in real time to generate a cumulative capacity, and automatically increments the cycle count when the integrated value equals the rated capacity. All channels are physically sampled at 1kHz, digitally averaged with a 100ms logic period, and filtered with infinite impulse response to form timestamped data frames. The data frames are first written to a 2048×16B circular buffer, and then pushed to the system bus via DMA to ensure that subsequent modules always acquire the latest measurement values.

[0145] Furthermore, the estimation module communicates with the acquisition module and runs on the floating-point coprocessor core of the microprocessor: its explicit layer solidifies the positive and negative electrode OCV-SOC curves (positive electrode fourth-order polynomial, negative electrode hyperbolic tangent function) and the charge conservation differential equation; the learnable layer adopts a 4-32-32-16-1 fully connected topology (total parameters 1761, ReLU activation), and only fits the polarization potential. The model inputs are current, temperature, rate of change of current, and negative electrode SOC; the output is the estimated value of the current negative electrode potential. Each inference attempt takes 0.6ms, and the result is sent to the risk assessment module via shared memory.

[0146] Furthermore, the judgment module and the estimation module are connected in communication to determine whether there is a risk of sodium precipitation based on a dynamic threshold and to generate a risk signal. The safety margin is calculated by looking up a table based on temperature and SOH. (20 mV–80 mV), and combined with a fixed reference of 0 V to obtain a dynamic threshold. ;like This generates a 1-bit sodium precipitation risk signal flag, along with the current... Temperature and SOH are broadcast to the closed-loop collaborative control module via an internal message queue; otherwise, Flag is 0.

[0147] Furthermore, the optimization module and the judgment module are connected for communication, used to perform micro-current adjustment based on risk signals during the constant current phase, and to perform dynamic voltage reduction or shorten the constant voltage time during the constant voltage phase based on risk signals. Charging is stopped when either type of adjustment is ineffective three times consecutively. It includes a constant current PI current adjustment subunit and a constant voltage slope reduction subunit: during the constant current phase, when Flag=1, the PI controller... , Calculate the current correction amount and limit it to Within this range, new current settings are issued in real time; during the constant voltage phase, based on the slope table... k (SOC, T The target voltage reduction is calculated, and the CV-SET command of the energy storage converter is updated or the constant voltage duration is shortened. If the Flag is still 1 after three consecutive adjustments of any sub-unit, the control module sends a STOP_CHARGE to the energy storage converter via CAN and calls the safe charging stop procedure.

[0148] Furthermore, a sodium-ion battery charging optimization system also includes an online self-calibration module and an execution and communication module.

[0149] The online self-calibration module communicates with the estimation module to adjust neural network parameters and safety margins using low-magnification data within a preset cyclic interval. Incremental updates are performed to compensate for temperature drift and capacity decay. The online self-calibration module starts a cycle counter, triggering calibration upon reaching 100 charge-discharge cycles or detecting that the model residual exceeds a threshold: the module retrieves the most recent 5 minutes of low-rate or static data, freezes explicit layers, and performs 5-step Adam updates (learning rate 1e-4) only on the polarization network weights; after the update, the RMSE is calculated on the validation fragment, and if it decreases by ≥2mV, the new weights are written, and the model is simultaneously recalibrated based on the residual-temperature curve. Make a 5mV level fine-tuning; otherwise, roll back the old weights and maintain the original margin.

[0150] The execution and communication module communicates bidirectionally with the aforementioned modules. It is configured to run unit functions on the microcontroller or embedded processor and interact with an external host via CAN, UART, or Ethernet interfaces to remotely upgrade network weights and control parameters. The execution and communication module is integrated into the main microcontroller's RTOS task manager: on one hand, it schedules the execution priority and time slices of the aforementioned functional units; on the other hand, it provides parameter read / write, firmware OTA, and network weight upgrade interfaces to the external host via dual CAN (500kbps), UART (115200bps), and 100M Ethernet. All external commands are subject to CRC verification and authorization authentication before taking effect, ensuring the system's safe and reliable operation.

[0151] This invention can accurately determine the negative electrode potential without a reference electrode: the physical information neural network embeds the open-circuit voltage-charge state curve and charge conservation equation into the network structure, and only performs lightweight learning on the polarization potential, which can achieve sub-millisecond inference on a typical microcontroller; the estimation error is stably controlled within 20mV, avoiding the high cost and packaging damage problems of traditional reference electrode schemes, and simplifying the assembly and system integration of sodium-ion batteries from the hardware level.

[0152] The charging current-voltage dual-channel closed-loop regulation balances fast charging efficiency and safety: When the negative electrode potential approaches the sodium deposition threshold, the PI closed-loop current regulation algorithm first dynamically reduces the current in the constant current stage with the smallest amplitude; if there is still a risk, the constant voltage target is simultaneously lowered or the constant voltage time is shortened by using the negative electrode potential-full battery voltage slope model, so that the negative electrode potential can quickly rise while the current and time decreases are minimized.

[0153] Dynamic safety margin and incremental learning mechanism ensure reliability throughout the entire life cycle: Safety margin Adaptive adjustment based on temperature and aging status, incremental learning per... Each cycle automatically fine-tunes the polarization parameters to continuously compensate for model shifts caused by rising internal resistance and capacity decay.

[0154] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0155] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for sodium-ion battery charging optimization, characterized in that, The method comprises the following steps: Real-time acquisition of charging data of the sodium-ion battery at the current sampling point, wherein the charging data comprises current and temperature; inputting the current of the current sampling point into the charge conservation equation and performing forward Euler updating to obtain the negative electrode state of charge of the current sampling point, inputting the negative electrode state of charge of the current sampling point into the negative electrode open circuit voltage-state of charge mapping equation to perform fitting to obtain the negative electrode theoretical potential of the current sampling point; performing feature extraction and feature mapping on a vector composed of the current, temperature, current change rate and negative electrode state of charge of the current sampling point to obtain the negative electrode polarization potential; and adding the negative electrode theoretical potential of the current sampling point and the negative electrode polarization potential to obtain the potential estimation value of the negative electrode relative to Na + of the current sampling point. When the potential estimation value is less than the risk threshold value, it is determined that there is a risk of sodium precipitation; wherein the risk threshold value is composed by superimposing a zero-volt reference and a dynamic safety margin, and the dynamic safety margin is constructed by real-time temperature and the state of health of the battery; When it is determined that there is a risk of sodium precipitation, charging optimization is performed in the constant-current stage and the constant-voltage stage of the sodium-ion battery; wherein in the constant-current stage, the charging current is continuously reduced according to a closed-loop feedback algorithm until the potential estimation value is not less than the risk threshold value; in the constant-voltage stage, the constant-voltage target is reduced or the constant-voltage duration is shortened according to the closed-loop feedback algorithm until the potential estimation value is not less than the risk threshold value.

2. The sodium-ion battery charge optimization method of claim 1, wherein, obtaining an estimated value of the potential of the negative electrode relative to Na + at the current sampling point through a pre-trained physical information neural network model, the physical information neural network model comprising an explicit physical layer and a learnable polaron network; The explicit physical layer comprises a positive electrode open-circuit voltage-state-of-charge mapping equation, a negative electrode open-circuit voltage-state-of-charge mapping equation, and a charge conservation equation; the positive electrode open-circuit voltage-state-of-charge mapping equation and the negative electrode open-circuit voltage-state-of-charge mapping equation are specifically as follows: ； ； wherein Voc is the open circuit voltage of the positive electrode, Voc is the open circuit voltage of the negative electrode, i.e. the negative electrode theoretical potential, and Soc and Sod represent the state of charge of the positive and negative electrodes, respectively, , , , , , , , and is the open circuit coefficient; The charge conservation equation is specifically as follows: ； wherein is t the initial value of the difference in state of charge of the two electrodes for t is the t is the is the equivalent reversible capacity, is the Faraday constant, is the current, is the time variable; The learnable polarization subnetwork comprises an input layer, a plurality of hidden layers, and an output layer; the vector composed of the current, the temperature, the first-order derivative of the current, and the negative electrode state of charge at the current sampling point is subjected to feature extraction through the plurality of hidden layers, and the extracted features are subjected to feature mapping through the output layer to obtain the negative electrode polarization potential.

3. The sodium-ion battery charge optimization method of claim 2, wherein, The charging data further comprises full-cell terminal voltage, and the pre-training of the physical information neural network model comprises the following steps: Charging data of sodium-ion batteries at different time points and the corresponding negative electrode potentials relative to Na + / Na actual value; The charging data at different time points is input into the physical information neural network model to obtain the corresponding potential estimation value, full-cell voltage reconstruction quantity, and negative electrode state of charge change rate; A loss function is constructed by the potential estimation value, the actual potential value, the full-cell voltage reconstruction quantity, the full-cell terminal voltage, and the negative electrode state of charge change rate; The explicit physical layer parameters are frozen, and the learnable polarization subnetwork is trained based on the loss function to obtain the pre-trained physical information neural network model.

4. The sodium-ion battery charge optimization method of claim 1, wherein, The risk threshold value is specifically as follows: ； ； wherein, is a risk threshold, is a zero volt reference, is a dynamic safety margin, is a temperature sensitivity coefficient, is an aging sensitivity coefficient, is a health state.

5. The sodium-ion battery charge optimization method of claim 1, wherein, In the constant-current stage, the charging current is continuously reduced according to a closed-loop feedback algorithm, specifically comprising the following steps: obtaining a negative electrode potential estimation value of the current sampling point and a risk threshold value and calculating the deviation between them ; If there is a deviation If the value is greater than 0, the current is corrected using the current correction expression to obtain the current correction amount. The specific current correction expression is as follows: ； In the formula, is the current correction amount of the mth t sampling point, is a proportional coefficient, is an integral coefficient, is a time variable , the difference between the negative electrode potential estimation value and the risk threshold value; The current correction quantity is subjected to amplitude limiting: ； In the formula, is the current charging current; The charging current is corrected according to the current correction quantity to obtain a new charging current: ； In the formula, is the new charging current.

6. The sodium-ion battery charge optimization method of claim 1, wherein, The constant-voltage target is reduced or the constant-voltage duration is shortened according to a closed-loop feedback algorithm, specifically comprising the following steps: The sodium-ion battery was tested at a rate of 0.2C in the range of 10%-100% SOC, -10°C-55°C, and the curves of vs. were recorded. A local slope is obtained by piecewise linear regression for each curve, and the negative electrode potential-full-cell voltage slope model is obtained by interpolation of the state of charge with a 5% step and the temperature with a 10℃ step; Obtaining a negative electrode potential estimate and a risk threshold and calculating the lift between them ; querying the negative electrode potential-full cell voltage slope model based on the real-time SOC and temperature to obtain a current slope b ; obtaining a target drop based on the lifting amount and the current slope b ​ Obtain new constant pressure target according to target reduction At the same time, shorten the constant pressure duration by the proportional factor.

7. The sodium-ion battery charge optimization method of claim 5, wherein, Further comprising: When the negative electrode potential estimation value fails to be lifted above the risk threshold value for three consecutive iterations in the constant-current stage or the constant-voltage stage, a stop charging instruction is immediately sent to the current transformer to cut off the charging circuit and set the system state to a waiting mode; During the waiting mode, if it is detected that then the restart procedure is entered.

8. A sodium-ion battery charge optimization system, characterized by, Comprising: The acquisition module is configured to acquire, in real time, charging data of the sodium-ion battery at a current sampling point, wherein the charging data comprises current and temperature; The estimation module is configured to input the current of the current sampling point into a charge conservation equation and perform a forward Euler update to obtain a negative electrode state of charge of the current sampling point, input the negative electrode state of charge of the current sampling point into a negative electrode open circuit voltage-state of charge mapping equation to perform fitting, and obtain a negative electrode theoretical potential of the current sampling point; perform feature extraction and feature mapping on a vector composed of the current, temperature, current rate of change and negative electrode state of charge of the current sampling point to obtain a negative electrode polarization potential; and add the negative electrode theoretical potential of the current sampling point and the negative electrode polarization potential to obtain an estimated value of the potential of the negative electrode relative to Na + / Na of the current sampling point. A judgment module is configured to determine that there is a sodium precipitation risk when the potential estimation value is less than a risk threshold value; wherein the risk threshold value is formed by superimposing a zero-volt reference and a dynamic safety margin, and the dynamic safety margin is constructed by a real-time temperature and a state of health of the battery; An optimization module is configured to perform charging optimization in a constant-current stage and a constant-voltage stage of the sodium-ion battery when it is determined that there is a sodium precipitation risk; wherein in the constant-current stage, the charging current is continuously reduced according to a closed-loop feedback algorithm until the potential estimation value is not less than the risk threshold value; and in the constant-voltage stage, the constant-voltage target is reduced or the constant-voltage duration is shortened according to the closed-loop feedback algorithm until the potential estimation value is not less than the risk threshold value.

Images