Physical battery management method, terminal and storage medium

Abstract

The application provides a physical battery management method, a terminal and a storage medium. Running data of a physical battery is collected and preprocessed to obtain a standardized feature data set; a digital twin model including a multi-dimensional state feature evolution model and a mechanism-data fusion electro-thermal coupling model is constructed; the running data, the standardized feature data set and expected working condition information are input into the digital twin model, first output data are output by the multi-dimensional state feature evolution model, and second output data are output by the mechanism-data fusion electro-thermal coupling model; the two output data are checked and fused to obtain a full state vector; strategy searching and optimization are performed on the digital twin model based on the full state vector, an optimization target, a constraint condition and a preset action space to obtain a target action and issue the target action to a battery management system for execution, so that accurate state prediction and active optimization control of the physical battery can be realized, and the safety and service life of the physical battery in operation are improved.

Description

Technical Field

[0001] This invention relates to the field of physical battery technology, and in particular to physical battery management methods, terminals, and storage media. Background Technology

[0002] With the rapid development of large-scale energy storage systems, the performance, safety, and economic management of energy storage batteries, as core energy storage units, throughout their entire lifecycle has become crucial. Traditional battery management systems (BMS) mainly rely on real-time or historical operating data (voltage, current, temperature, etc.) for state estimation (such as SOC, SOH) and passive protection (such as overcharge, over-discharge, and over-temperature protection), essentially a "passive response" mechanism based on threshold judgment. In recent years, although the concept of intelligent operation and maintenance has been introduced, and some fault warnings have been achieved through data analysis, its analytical dimensions, predictive depth, and proactive decision-making are still insufficient. Specifically, the parameters of existing BMS models (such as internal resistance and capacity) are mostly factory-calibrated values ​​or calibrated under specific health conditions, which cannot be adaptively updated to follow the performance degradation throughout the battery's entire lifecycle. This leads to the "mismatch" between the model and the physical entity, which intensifies with usage time. At the same time, operation and maintenance strategies heavily rely on fixed safety thresholds, allowing intervention only when an anomaly occurs or afterward, which is a "sudden stop during the event" or "remedial action after the event," lacking the ability to proactively predict and mitigate potential risks. Furthermore, predictions of remaining useful life (RUL) and internal critical states (such as lithium deposition and SEI film growth) are mostly based on simplified empirical models or single data sources, which lack accuracy and reliability, and are particularly poor in adaptability to complex operating conditions and individual differences. Charging strategies, thermal management strategies, and equalization strategies are mostly preset or offline optimized, and cannot be dynamically, online, and globally optimized according to the real-time status of the battery, environmental changes, and usage requirements.

[0003] In summary, existing technologies lack a high-fidelity virtual mirror model that can span the entire process from battery production and use to retirement. They also fail to dynamically generate optimal management strategies based on the individual characteristics of battery degradation in real time. As a result, the battery's potential is not fully released, lifespan prediction is inaccurate, safety risk warnings are delayed, and maintenance efficiency is low. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a physical battery management method, terminal and storage medium that can realize accurate state prediction and active optimization control of physical batteries, thereby improving the safety and service life of physical battery operation.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A physical battery management method, comprising: Collect the operating data of the physical battery and preprocess the operating data to obtain a standardized feature dataset; A digital twin model of the physical battery is constructed, the digital twin model including a multidimensional state feature evolution model and a mechanism-data fusion electrothermal coupling model; Obtain expected operating condition information, input the operating data, the standardized feature dataset, and the expected operating condition information into the digital twin model, the multidimensional state feature evolution model outputs first output data based on the standardized feature dataset and the expected operating condition information, the mechanism-data fusion electrothermal coupling model outputs second output data based on the operating data and the expected operating condition information, and the first output data and the second output data are verified and fused to obtain a full state vector; The optimization objective, constraints, and preset action space are obtained. Based on the full state vector, the optimization objective, the constraints, and the preset action space, a strategy search and optimization is performed on the digital twin model to obtain the target action. The target action is then sent to the battery management system, which executes the target action on the physical battery.

[0006] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A physical battery management terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the various steps of the physical battery management method described above.

[0007] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A computer storage medium storing a computer program that is executed by a processor to implement the steps of the physical battery management method described above.

[0008] The beneficial effects of this invention are as follows: By collecting and preprocessing the operating data of physical batteries, a standardized feature dataset is obtained, which can eliminate noise, outliers, and missing values ​​in the original data, providing a reliable data foundation for subsequent models; a digital twin model is constructed, which includes a multi-dimensional state feature evolution model and a mechanism-data fusion electrothermal coupling model. The dual models output first and second output data based on the standardized feature dataset, operating data, and expected operating condition information. The first and second output data are verified and fused to generate a full state vector, which can effectively suppress the prediction bias of a single model under complex operating conditions and improve the robustness and accuracy of physical battery state estimation; optimization objectives, constraints, and preset action spaces are obtained, and strategy search and optimization are performed on the digital twin model based on the full state vector to obtain the target action and send it to the battery management system for execution, realizing the transformation from passive response to active optimization. This enables the battery management strategy to adapt to the real-time state and degradation characteristics of the battery, improving the safety and service life of physical battery operation. Attached Figure Description

[0009] Figure 1 This is a flowchart of a physical battery management method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a physical battery management terminal according to an embodiment of the present invention; Detailed Implementation Definitions:

[0010] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0011] In existing technologies, battery management systems for energy storage batteries primarily rely on passive protection based on real-time or historical data such as voltage, current, and temperature for state estimation and threshold triggering. These systems typically use fixed model parameters (such as internal resistance and capacity) calibrated at the factory or under specific health conditions, failing to adaptively update with performance degradation throughout the battery's lifespan. This exacerbates the mismatch between the model and the actual battery. Furthermore, maintenance strategies often depend on fixed safety thresholds, representing in-process or post-process intervention, lacking accurate prediction of critical internal states such as lithium deposition and SEI film growth, as well as remaining battery life. Simultaneously, charging, thermal management, and equalization strategies are preset or optimized offline, unable to dynamically adjust according to real-time battery status and environmental changes, resulting in inaccurate lifespan predictions, delayed safety risk warnings, and insufficient release of battery potential.

[0012] To address at least the aforementioned issues, this paper constructs a digital twin containing a multidimensional state feature evolution model and a mechanism-data fusion electrothermal coupling model by collecting and preprocessing the operational data of the physical battery. The multidimensional state feature evolution model outputs first output data using expected operating condition information and a standardized feature dataset, while the mechanism-data fusion electrothermal coupling model outputs second output data using the original operational data and expected operating condition information. The outputs of the two models are cross-validated and fused to form a full state vector. Combining the full state vector, optimization objectives, constraints, and a preset action space, online strategy search and optimization are performed on the digital twin model, and the obtained target actions are then sent to the battery management system for execution. This approach enables accurate state prediction and proactive optimization control of the physical battery, allowing the battery management strategy to adapt to the real-time battery state and degradation characteristics, thereby improving the safety and lifespan of the physical battery.

[0013] The following describes a physical battery management method of the present invention in detail. Please refer to [link / reference]. Figure 1 The method 100 includes steps 101 to 104: Step 101: Collect the operating data of the physical battery and preprocess the operating data to obtain a standardized feature dataset.

[0014] Specifically, by integrating high-precision sensors and a BMS (Battery Management System) at the physical battery level, real-time operational data of the physical battery is collected. This data directly reflects the battery's current instantaneous state, including the voltage and current of each cell, the temperature at each measurement point, timestamps, battery pack operation logs, operating condition information (such as charge / discharge rate and ambient temperature), and historical maintenance records. The collected time-series data is then cleaned, aligned, and feature-extracted (such as incremental capacity analysis IC curve features and temperature distribution entropy) to form a standardized feature dataset, providing a reliable data foundation for subsequent models.

[0015] Step 102: Construct a digital twin model of the physical battery. The digital twin model includes a multidimensional state feature evolution model and a mechanism-data fusion electrothermal coupling model.

[0016] Specifically, a dynamic digital twin model of the physical battery's entire lifecycle is constructed. This involves importing prior knowledge such as battery design parameters, factory test data, and material properties into the digital twin model. This model is a high-fidelity dynamic mapping of the physical battery in virtual space and includes two core sub-models: a multidimensional state characteristic evolution model and a mechanism-data fusion electrothermal coupling model. The multidimensional state characteristic evolution model is used to predict the evolution trajectory of the battery's macroscopic state indicators and key microscopic degradation characteristics over time. The mechanism-data fusion electrothermal coupling model is used to estimate mechanism parameters and simulate physical quantities. In this way, a dual-model collaborative mechanism can be used to achieve high-precision, full-cycle prediction of the battery's internal state and remaining lifespan, providing a reliable basis for battery management decisions.

[0017] Step 103: Obtain expected operating condition information. Input the operating data, the standardized feature dataset, and the expected operating condition information into the digital twin model. The multidimensional state feature evolution model outputs first output data based on the standardized feature dataset and the expected operating condition information. The mechanism-data fusion electrothermal coupling model outputs second output data based on the operating data and the expected operating condition information. Verify and fuse the first output data and the second output data to obtain the full state vector.

[0018] Specifically, the system obtains preset operating condition information for a future period of time from user input / trip planning (users directly set ("I want to charge to 80% within 30 minutes")), historical behavior prediction (prediction based on time-series prediction models of historical operating data), and external dispatch signals (power commands issued by the power grid or energy storage dispatch system). This includes the expected charge and discharge power curve, ambient temperature change curve, and user behavior patterns (such as fast charging before work and slow charging at night) for a future period of time (such as the next hour or the next complete discharge cycle). The operating data, standardized feature dataset, and expected operating conditions over a future period are input into a digital twin model. The multi-dimensional state feature evolution model within the digital twin model, using a deep learning network, outputs first output data based on the standardized feature dataset and expected operating conditions. This output data includes the current values ​​of macroscopic state indicators, predicted sequences for future time steps (T), and predicted values ​​of microscopic degradation features. The mechanism-data fusion electrothermal coupling model within the digital twin model performs self-updates based on the operating data and expected operating conditions, predicting physical quantities to obtain second output data including mechanism parameters and physical quantities. The first and second output data are then validated and fused to obtain the full state vector. This approach effectively suppresses prediction biases of a single model under complex operating conditions, improving the robustness and accuracy of physical battery state estimation.

[0019] Step 104: Obtain the optimization objective, constraints, and preset action space. Based on the full state vector, the optimization objective, the constraints, and the preset action space, perform strategy search and optimization on the digital twin model to obtain the target action. Send the target action to the battery management system, which then executes the target action on the physical battery.

[0020] Specifically, the full-state vector is used as the current state. A multi-objective reward function is constructed based on optimization objectives (such as maximizing the charging speed, minimizing the current cycle capacity decay, and balancing cell temperature differences) and constraints. These constraints include safety boundaries (such as voltage < 4.2V, temperature < 55℃, and lithium plating risk probability < 5%) and hardware limitations (maximum charging current and cooling power limits). Using a preset action space as the search domain, reinforcement learning algorithms (such as DDPG and PPO) or optimization algorithms are employed in the virtual simulation environment of the digital twin model for rapid, large-scale simulation and optimization. The resulting target strategy parameters maximize long-term cumulative rewards, forming the target action, are then sent to the BMS for execution. This approach enables low-risk, high-efficiency virtual strategy exploration using a high-fidelity digital twin model, avoiding direct trial and error on the physical battery. Furthermore, the management strategy can be adjusted online based on real-time status, allowing the battery management strategy to adapt to the real-time battery status and degradation characteristics, thereby improving the safety and lifespan of the physical battery.

[0021] In one embodiment of the present invention, step 101 includes step 1011: Step 1011: The running data includes time-series data; the time-series data is cleaned and aligned, and macroscopic state indicators and microscopic degradation features are extracted from the cleaned and aligned time-series data. The macroscopic state indicators are used to represent the overall performance indicators of the physical battery, and the microscopic degradation features are used to represent the indirect characteristics or state quantities of the aging mechanism or material changes inside the physical battery. A standardized feature dataset is generated based on the macroscopic state indicators and the microscopic degradation features.

[0022] Specifically, time-series data includes high-frequency electrical sequences (continuously acquired voltage, current, and temperature time-series data (e.g., acquired once per second)), operating condition history sequences (charge and discharge rate changes over time) and event sequences (maintenance actions (equilibrium start / end), fault alarm occurrence time and duration). The standardized feature dataset includes macroscopic state indicators and microscopic degradation features. The macroscopic state indicators include the current state of charge (SOC), battery health state (SOH) (estimated by the capacity increment method), and internal resistance (calculated by ohmic voltage drop), which are calculated directly or indirectly. The microscopic degradation features include IC (incremental capacity) curve features (key peak height and peak position extracted from the voltage-capacity differential curve of a partial charge-discharge segment), temperature distribution entropy (disorder index of temperature distribution at each temperature measurement point in the cell or module, reflecting thermal non-uniformity), voltage relaxation features (slope of the voltage recovery curve over time after charging and discharging stops, and time constant), cycle accumulation (cumulative charge and discharge energy, and cumulative overcharge and over-discharge time), and operating condition context features (average depth of discharge, average C-rate, and ambient temperature statistics (mean and variance) over the past N cycles).

[0023] The time-series data undergoes data cleaning, including noise reduction, outlier removal, and missing value handling: median filtering, Kalman filtering, or wavelet transform are used to filter out sensor noise in the time-series data; outliers exceeding physical limits (such as voltage jumping instantaneously from 3.7V to 5V) are deleted or filled by interpolation; for data missing due to communication failures, linear interpolation or forward fill is used to fill short time gaps, while long missing segments are marked and discarded. After cleaning, data alignment is performed, including timestamp normalization and time alignment: all sensor data (voltage, current, multiple temperature points) are resampled according to a unified time base (e.g., 1Hz); for data with non-equal intervals, time-weighted averaging or nearest-neighbor interpolation is used; the timestamps of each event (such as equalization, charge / discharge switching) are aligned to the above-mentioned base time line as 0 / 1 markers. Macroscopic state indicators and microscopic degradation features are extracted from the cleaned and aligned time-series data: A fixed-length (e.g., the past hour) sliding time window is used to calculate statistical characteristics (mean, variance, rate of change) within the window; a complete segment of constant-current charging data is selected to calculate the dQ / dV curve, and a local peak detection algorithm is used to extract the height and position of key peaks to obtain the IC (incremental capacity) curve features; after normalizing the readings of all temperature sensors at the same time, the temperature distribution entropy is calculated; some features (such as polarization resistance and diffusion coefficient) can be estimated online by inputting real-time running data into a simplified electrochemical equivalent circuit model (ECM). In this way, the raw time-series data can be transformed into a standardized feature dataset with clear physical meaning and aging sensitivity, improving the interpretability and characterization accuracy of battery internal degradation mechanisms (such as SEI film growth, lithium deposition, and thermal non-uniformity), providing reliable input data for digital twin models, and thus enhancing the reliability of state estimation and strategy optimization throughout the entire battery lifecycle.

[0024] In one embodiment of the present invention, step 103 includes steps 1031 to 1033: Step 1031: Using the standardized feature dataset and the expected working condition information as input, a deep learning network is used to predict the evolution trajectory of the macroscopic state index and the microscopic degradation feature over time, and the current value of the macroscopic state index and the predicted sequence of the future preset time steps, as well as the predicted value of the microscopic degradation feature, are output as the first output data.

[0025] Specifically, the multidimensional state feature evolution model employs deep learning networks (such as LSTM and Transformer) as input, using standardized feature datasets and expected operating condition information to learn and predict the evolution trajectory of the battery's macroscopic state and microscopic degradation features over time / cycles. It outputs the current values ​​of macroscopic state indicators and predicted sequences for future preset time steps, such as the current values ​​and predicted sequences for the next T time steps of the state of charge (SOC), state of health (SOH), and state of charge / discharge (SOP), as well as predicted values ​​of microscopic degradation features, such as the percentage of available lithium-ion loss, the percentage of active material loss, and the predicted SEI film thickness increase. In this way, a unified prediction of the battery's macroscopic performance degradation and internal aging mechanisms can be achieved, providing a forward-looking state basis for strategy optimization.

[0026] Step 1032: Based on the operating data, the mechanism parameters are estimated using a particle filter. The mechanism parameters are then used to update the mechanism-data fusion electrothermal coupling model. The updated mechanism-data fusion electrothermal coupling model predicts physical quantities based on the operating data and the expected operating condition information. The mechanism parameters and the physical quantities are then used as the second output data.

[0027] Specifically, the mechanism-data fusion electrothermal coupling model adds a real-time data-driven update layer to the basic mechanism model to achieve adaptive updates of model parameters. The basic mechanism model is constructed by coupling a simplified P2D electrochemical model with a three-dimensional lumped or one-dimensional heat conduction equation. In the electrochemical part, the positive and negative electrode plates are equivalent to two spherical particles with an electrolyte in between. The diffusion of lithium ions inside the particles is described by the solid-phase diffusion equation (Fick's second law), and the insertion / extraction reaction of lithium ions at the electrode-electrolyte interface is described by the Butler-Volmer equation, which includes the exchange current density and reaction rate constant. The convection and diffusion of Li+ in the electrolyte are described by the liquid-phase lithium ion concentration distribution equation, and Ohm's law is introduced to describe the solid-phase potential and liquid-phase potential distribution. The heat transfer part includes Ohmic heat (internal resistance heating), polarization heat, reaction heat, and convective heat transfer terms between the battery and the environment (coolant, air). The temperature is corrected for electrochemical parameters such as diffusion coefficient and reaction rate constant by the Arrhenius formula. In real-time operation, a particle filter (or Kalman filter, neural network calibrator) is used to dynamically update the mechanistic parameters of the mechanistic model online based on real-time operational data. These parameters include the diffusion coefficients of the positive and negative electrodes and the reaction rate constant. For example, as the battery ages, the solid-phase diffusion coefficient may decrease, and the particle filter will adjust this coefficient in real time to minimize the residual between the model output voltage and the actual voltage. Furthermore, when updating parameters, it is mandatory that the updated parameters conform to the physical laws of electrochemistry-thermodynamics (e.g., the decrease in diffusion coefficient cannot exceed a certain physical upper limit, and it must be positively correlated with the growth trend of the SEI film), avoiding physically meaningless overfitting caused by purely data-driven approaches.

[0028] The updated model, based on the input operating data and expected operating conditions, simulates and obtains the voltage of each cell, the total voltage, and the temperature rise curve as physical quantities. Mechanism parameters (such as diffusion coefficient, reaction rate constant, and SEI film impedance, which are time-varying model parameters corresponding to the microscopic degradation characteristics of the multidimensional state feature evolution model) and physical quantities estimated online through particle filtering are used as the second output data. In this way, the mechanism-data fusion electrothermal coupling model can adaptively calibrate with battery aging, maintaining high-fidelity simulation capabilities, and cross-validate with the microscopic degradation characteristics of the multidimensional state feature evolution model, improving the accuracy and robustness of the full state vector.

[0029] Step 1033: Determine whether the first output data and the second output data satisfy a preset mapping relationship. If they do, use a Bayesian fusion framework or a Kalman filter to fuse the first output data and the second output data through a covariance cross-fusion algorithm to obtain a full state vector. Otherwise, generate a contradictory signal. The contradictory signal is used to indicate that at least one of the multidimensional state feature evolution model or the mechanism-data fusion electrothermal coupling model needs to be corrected.

[0030] Specifically, determining whether the first and second output data satisfy a preset mapping relationship includes: substituting the microscopic degradation characteristics (such as the percentage of usable lithium ion loss and the percentage of active material loss) in the first output data and the mechanistic parameters (such as the positive and negative electrode solid-phase diffusion coefficients, reaction rate constants, and SEI film impedance) in the second output data into a mapping function (such as the functional relationship between diffusion coefficient and usable lithium loss and temperature) that has been experimentally calibrated or derived using first principles for physical consistency verification; if the mapping deviation between the two is within a preset threshold (e.g., 5%), it is determined that the mapping relationship is satisfied, and a Bayesian fusion framework or a dual Kalman filter is then used. The algorithm uses a filter to take the first output data (such as SOH and lithium loss) as the observation for state estimation and the second output data (voltage residual and identified parameters) as the observation for state update. Through a covariance cross-fusion algorithm, a confidence-weighted state estimate is output, resulting in a full state vector. This full state vector includes external physical quantities (corrected SOC, SOH, SOP, and remaining lifetime RUL) and internal mechanistic quantities (available lithium loss, active material loss, lithium plating risk probability, thermal runaway warning level, internal resistance (ohmic + polarization + diffusion), and temperature distribution). If the mapping deviation exceeds a threshold, a conflict signal is generated. This conflict signal triggers collaborative optimization: simultaneously correcting the first and second output data using an extended Kalman filter. This approach avoids state distortion caused by a single model bias, ensures the reliability of the full state vector generation, and provides a reliable, comprehensive, and interpretable state benchmark for subsequent management optimization.

[0031] In one embodiment of the present invention, step 104 includes step 1041: Step 1041: Using the digital twin model as a virtual simulation environment, with the full state vector as the current state, the optimization objective and the constraints are quantified into a reward function. Based on the preset action space, iterative simulation and optimization are performed using a reinforcement learning algorithm until the target action is converged.

[0032] Specifically, the preset action space refers to the range or discrete option set of all possible actions (i.e., controllable variables, such as the starting point, inflection point multiplier, and constant voltage stage voltage of the charging current curve; the PWM duty cycle of the cooling pump; and the SOC threshold for equalization activation) set by the reinforcement learning network during initialization. For example, the action space of the charging current curve can be defined as a vector space containing segmented multipliers (0.1C to 3C) and segmented times (0 to 120 minutes); the action space of the cooling system power is a continuous value range (0 to 100% PWM duty cycle); and the action space of the equalization trigger condition can be a discrete value (such as equalization activation when SOC equals 10%, 20%, etc.) or a continuously adjustable threshold range. Using a digital twin model as a virtual simulation environment, with the full-state vector as the current state, a pre-constructed reinforcement learning network samples candidate actions from a pre-defined action space based on the current state (e.g., charging at 2C for 10 minutes, then charging at 1C for 20 minutes). These candidate actions and the current state are then input into the digital twin model for simulation, i.e., the model extrapolates forward (e.g., simulating the next hour) to obtain the next state (the full-state vector at the end of the simulation). The reward value is then calculated based on the next state and the reward function (quantized according to the optimization objective and constraints). For example, the charging efficiency target is calculated based on the next state. The reward function is calculated by substituting the calculated values ​​of charging efficiency target achievement, capacity decay increment, and safety constraint violation penalty into the reward function: Reward Value = w1 × Charging Efficiency Target Achievement - w2 × Capacity Decay Increment - w3 × Safety Constraint Violation Penalty, where w1, w2, and w3 are weighting coefficients. The parameters of the reinforcement learning network (including the Actor and Critic networks) are updated based on the current state, candidate actions, reward value, and next state. The next state is then used as the current state for resampling and simulation, until the reward value is maximized, yielding the target action (such as an optimized multi-stage constant current-constant voltage charging curve). In this way, within the safety boundary, a management strategy best suited to the current state of the physical battery can be adaptively generated, improving the safety and lifespan of the physical battery operation.

[0033] In one embodiment of the present invention, after step 104, the method includes: obtaining the operating data of the battery management system after executing the target action, comparing the operating data after executing the target action with the output result of the digital twin model under the target action, and updating the digital twin model based on the comparison result.

[0034] Specifically, the system collects actual operating data of the physical batteries after the battery management system executes the target action, including the voltage, current, temperature response curves, and SOC trajectory of each cell. It also obtains the simulation output results of the digital twin model corresponding to the target action. By comparing the actual operating data with the simulation output results, the spatiotemporal distribution of the prediction error (e.g., the root mean square error of SOC estimation and the magnitude of temperature prediction error) is calculated. This is used to evaluate the effectiveness of the strategy and store it in an "adaptive strategy library." The adaptive strategy library not only stores executed battery management actions but also their corresponding current battery state labels (e.g., SOC range, SOH level, temperature setting) to facilitate case-based reasoning in similar scenarios (e.g., the same ambient temperature and the same battery aging stage). It allows for rapid invocation without the need for repeated searches, reducing the time spent on online optimization. On the other hand, through backpropagation or system identification methods, it provides feedback to correct the worst-performing sub-model in the digital twin model (e.g., if the temperature prediction deviation is large, the heat transfer coefficient of the thermal model is fine-tuned first; if the SOC deviation is large, the reaction kinetic parameters of the electrochemical model are adjusted), thereby achieving the self-evolutionary capability of "the more it is used, the more accurate the model becomes".

[0035] In summary, this invention provides a physical battery management method. By collecting and preprocessing the operational data of the physical battery, a standardized feature dataset is obtained, which eliminates noise, outliers, and missing values ​​in the original data, providing a reliable data foundation for subsequent models. A digital twin model is constructed, comprising a multi-dimensional state feature evolution model and a mechanism-data fusion electrothermal coupling model. The dual models output first and second output data based on the standardized feature dataset, operational data, and expected operating condition information. The first and second output data are verified and fused to generate a full state vector, which can effectively suppress the prediction bias of a single model under complex operating conditions and improve the robustness and accuracy of physical battery state estimation. The optimization objective, constraints, and preset action space are obtained. Based on the full state vector, strategy search and optimization are performed on the digital twin model to obtain the target action, which is then sent to the battery management system for execution. This realizes the transformation from passive response to active optimization, enabling the battery management strategy to adapt to the real-time state and degradation characteristics of the battery, thereby improving the safety and lifespan of the physical battery operation.

[0036] The present invention has the following beneficial effects: 1. Accurate prediction and transparent management: The dynamic digital twin model enables high-precision, full-cycle prediction of the battery's internal state and remaining lifespan, transforming the battery state from a "black box" to a "white box," making management decisions more data-driven.

[0037] 2. Proactive early warning and risk mitigation: Through early and quantitative simulation prediction of potential faults (such as lithium plating risk and thermal runaway risk), the operation and maintenance mode can be completely upgraded from "passive protection" to "proactive early warning and intervention", which greatly improves safety.

[0038] 3. Intelligent optimization and performance enhancement: The adaptive strategy can dynamically generate target control commands within the safety boundary according to different scenarios (fast charging, battery life, lifespan priority), which can significantly shorten charging time, extend battery life, and improve energy efficiency.

[0039] 4. Dynamic Adaptation and Personalized Management: The solution can automatically adapt to the initial differences and performance variations of different individual batteries, providing a "tailor-made" management strategy for each battery to achieve refined operation and maintenance.

[0040] 5. Unified Platform and Full Lifecycle Value Mining: This method provides a unified digital platform and model foundation for the design verification, online management, residual value assessment, and tiered utilization of energy storage batteries, opening up the entire lifecycle data chain and maximizing the value of battery assets.

[0041] Please refer to Figure 2 The present invention also provides a physical battery management terminal 200, including a memory 201, a processor 202, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of a physical battery management method as described above.

[0042] The present invention also provides a computer-readable storage medium. This storage medium stores a computer program that is executed by a processor to implement a serial port adaptive method for an embedded power supply device as described above.

[0043] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A physical battery management method, characterized in that, include: Collect the operating data of the physical battery and preprocess the operating data to obtain a standardized feature dataset; A digital twin model of the physical battery is constructed, the digital twin model including a multidimensional state feature evolution model and a mechanism-data fusion electrothermal coupling model; Obtain expected operating condition information, input the operating data, the standardized feature dataset, and the expected operating condition information into the digital twin model, the multidimensional state feature evolution model outputs first output data based on the standardized feature dataset and the expected operating condition information, the mechanism-data fusion electrothermal coupling model outputs second output data based on the operating data and the expected operating condition information, and the first output data and the second output data are verified and fused to obtain a full state vector; The optimization objective, constraints, and preset action space are obtained. Based on the full state vector, the optimization objective, the constraints, and the preset action space, a strategy search and optimization is performed on the digital twin model to obtain the target action. The target action is then sent to the battery management system, which executes the target action on the physical battery.

2. The physical battery management method according to claim 1, characterized in that, The runtime data is preprocessed to obtain a standardized feature dataset, including: The operational data includes time-series data; The time-series data is cleaned and aligned. Macroscopic state indicators and microscopic degradation features are extracted from the cleaned and aligned time-series data. The macroscopic state indicators are used to represent the overall performance indicators of the physical battery. The microscopic degradation features are used to represent the indirect characteristics or state quantities of the aging mechanism or material changes inside the physical battery. A standardized feature dataset is generated based on the macroscopic state indicators and the microscopic degradation features.

3. The physical battery management method according to claim 2, characterized in that, The multidimensional state feature evolution model outputs first output data based on the standardized feature dataset and the expected operating condition information, including: Using the standardized feature dataset and the expected working condition information as input, a deep learning network is used to predict the evolution trajectory of the macroscopic state indicators and the microscopic degradation features over time, and outputs the current value of the macroscopic state indicators and the predicted sequence of future preset time steps, as well as the predicted value of the microscopic degradation features, as the first output data.

4. The physical battery management method according to claim 1, characterized in that, The mechanism-data fusion electrothermal coupling model outputs second output data based on the operating data and the expected operating condition information, including: Based on the operational data, the mechanism parameters are estimated using a particle filter. The mechanism parameters are then used to update the mechanism-data fusion electrothermal coupling model. The updated mechanism-data fusion electrothermal coupling model predicts physical quantities based on the operational data and the expected operating conditions. The mechanism parameters and the physical quantities are then used as the second output data.

5. The physical battery management method according to claim 1, characterized in that, The first output data and the second output data are verified and fused to obtain a full state vector, including: Determine whether the first output data and the second output data satisfy a preset mapping relationship. If they do, use a Bayesian fusion framework or a Kalman filter to fuse the first output data and the second output data through a covariance cross-fusion algorithm to obtain a full state vector. Otherwise, generate a contradictory signal. The contradictory signal is used to indicate that at least one of the multidimensional state feature evolution model or the mechanism-data fusion electrothermal coupling model needs to be corrected.

6. The physical battery management method according to claim 1, characterized in that, Based on the full state vector, the optimization objective, the constraints, and the preset action space, a strategy search and optimization are performed on the digital twin model to obtain the target action, including: Using the digital twin model as a virtual simulation environment, with the full state vector as the current state, the optimization objective and the constraints are quantified into a reward function. Based on the preset action space, reinforcement learning algorithms are used for iterative simulation and optimization until the target action is obtained.

7. The physical battery management method according to claim 6, characterized in that, Based on the preset action space, reinforcement learning algorithms are used for iterative simulation and optimization until the target action is converged, including: By using a pre-built reinforcement learning network, candidate actions are sampled from a preset action space based on the current state. The candidate actions and the current state are then input into the digital twin model for simulation and deduction to obtain the next state. The reward value is then calculated based on the next state and the reward function. The parameters of the reinforcement learning network are updated based on the current state, the candidate action, the reward value, and the next state. The next state is then used as the current state, and the network is resampled and simulated until the reward value is maximized to obtain the target action.

8. The physical battery management method according to claim 1, characterized in that, The battery management system performs the target action on the physical battery, followed by: The system acquires the operational data of the battery management system after executing the target action, compares the operational data after executing the target action with the output result of the digital twin model under the target action, and updates the digital twin model based on the comparison result.

9. A physical battery management terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step of the physical battery management method according to any one of claims 1 to 8.

10. A computer storage medium storing a computer program, characterized in that, The computer program is executed by a processor to implement the steps of a physical battery management method according to any one of claims 1 to 8.

Images