Battery endurance early warning method and system and related components

Abstract

The invention provides a battery endurance early warning method, which comprises the following steps of: acquiring a charge and discharge measurement signal of each battery in a preset battery pack, and determining a charge and discharge curve, a charge state and a health state of the battery according to the charge and discharge measurement signal and a preset equivalent circuit model; on the basis of a double-branch framework, predicting the residual discharge cycle number according to the charge and discharge curve, the charge state and the health state; generating a charging fluctuation diagram according to the residual discharge cycle number and the preset equivalent circuit model; judging whether the battery is abnormal or not according to the charging fluctuation diagram and the residual discharging cycle number; if yes, generating an endurance early warning instruction, and adjusting a charging strategy of the preset battery pack; according to the method, accurate evaluation and prospective early warning of the health state of the battery are realized through deep integration of physical knowledge and an innovative framework driven by data and a systematic multi-step processing flow.

Description

Technical Field

[0001] This invention relates to the field of battery life warning technology, and in particular to a battery life warning method, system, device and storage medium based on BMS. Background Technology

[0002] Currently, battery management systems mainly rely on data-driven methods or traditional physical models based on equivalent circuit models when predicting the remaining battery life and issuing range warnings. However, data-driven methods have poor generalization ability for scenarios with insufficient early life data and are difficult to capture physical processes such as the growth of solid electrolyte interfaces within the battery. Pure physical models, on the other hand, cannot fully learn the complex nonlinear characteristics in actual operation, resulting in limited prediction accuracy and delayed warning response.

[0003] To address these issues, we propose a battery endurance warning method, system, and related components to solve the core problems faced by existing battery management systems (BMS) in predicting remaining battery life (RUL) and issuing endurance warnings, such as scarce early data, unclear physical mechanisms, and delayed warnings. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. To achieve the above objective, this invention adopts the following technical solution: A battery life warning method, the battery life warning method comprising: S101: Collect the charge and discharge measurement signals of each battery in the preset battery pack, and determine the charge and discharge curve, charge state and health state of the battery based on the charge and discharge measurement signals and the preset equivalent circuit model. For example, charge and discharge measurement signals of each battery in a preset battery pack are collected. These signals include time-series data of voltage, current, and temperature, acquired in real-time by a high-precision sensor and stored at a fixed sampling frequency. Based on the charge and discharge measurement signals and a preset equivalent circuit model (which employs a topology including internal resistance, capacitance, and voltage sources), the electrochemical behavior of the battery is dynamically simulated using numerical simulation methods to generate a high-fidelity charge and discharge curve. This curve reflects the battery's response characteristics under various operating conditions. Further, based on the charge and discharge curve, the state of charge is calculated using the ampere-hour integral method, and the health state is estimated using a capacity decay model. The capacity decay model fits the degradation trend through cycle count and internal resistance changes, outputting a battery state parameter set, which includes the charge and discharge curve, state of charge, and health state.

[0005] S102: Based on a dual-branch framework, the remaining discharge cycle number is predicted according to the charge-discharge curve, the charge state, and the health state. The dual-branch framework includes a physical branch and a data-driven branch. The physical branch embeds an interface growth model of the solid electrolyte, and the data-driven branch adopts an extended convolutional neural network. For example, a dual-branch framework is used, consisting of a physical branch and a data-driven branch forming a parallel processing architecture. This framework predicts the remaining discharge cycles based on the charge / discharge curves, charge state, and health state output from the previous step. The physical branch embeds a solid electrolyte interface growth model, which describes the degradation kinetics based on the Arrhenius equation. This model calculates the evolution of the interface layer by introducing temperature-dependent activation energy and reaction rate constants, outputting physical consistency features such as the interface thickness growth rate and lithium-ion loss. The data-driven branch employs an expanded convolutional neural network. This network captures long-term time dependencies through multiple expanded convolutional layers and residual connections, extracting nonlinear degradation features from the original signal. For example, a layer structure with a kernel size of 3 and an expansion rate of 2 is used to enhance the feature receptive field. The outputs of the two branches are fused at the feature layer, using weighted summation or attention mechanisms to integrate physical constraints and data-driven features. The remaining discharge cycles are then regressed through a fully connected layer. For example, using only four early data points, the prediction error can be reduced to within 15 cycles, significantly improving prediction accuracy and generalization ability. The entire prediction process is optimized through a three-step training strategy, including pre-trained data-driven branching, aligning physical branch parameters, and joint fine-tuning, to ensure that the model remains highly robust with limited data, while also supporting the switching of different physical models to adapt to diverse battery chemistry systems.

[0006] S103: Generate a charging fluctuation diagram based on the remaining discharge cycle number and the preset equivalent circuit model; For example, a charging fluctuation diagram is generated based on the remaining discharge cycle number and a preset equivalent circuit model. The remaining discharge cycle number serves as a key input for adjusting the parameters of the equivalent circuit model, such as optimizing charging current and voltage limits to simulate actual operating conditions. The charging process is simulated based on the updated equivalent circuit model, collecting dynamic fluctuation data of voltage and current. This data is recorded in time series format and includes transient response information. A charging fluctuation diagram is constructed using signal processing techniques such as Fast Fourier Transform or sliding window analysis. This diagram characterizes abnormal modes during the charging process, such as overshoot or oscillation, and is output in a visual form for use by subsequent modules.

[0007] S104: Determine whether the battery is abnormal based on the charging fluctuation diagram and the remaining discharge cycle count; For example, the presence of battery anomalies is determined based on the charging fluctuation graph and the remaining discharge cycle count. The charging fluctuation graph provides voltage and current fluctuation characteristics, including quantitative indicators such as peak value, frequency, and duration. These indicators are extracted through sliding window analysis and statistical methods (such as calculating the root mean square error or peak-to-valley difference) and compared with preset thresholds to identify preliminary anomalies. For example, a voltage drop exceeding 0.5V or a current oscillation frequency exceeding 100Hz are considered risk signals. Simultaneously, a comprehensive risk assessment is performed using the remaining discharge cycle count. Multivariate decision models such as logistic regression or random forests are used to quantify the probability of anomalies. For instance, if the remaining discharge cycle count is below 100 and the fluctuation characteristics exceed the safe range, the battery is determined to have degradation anomalies. The anomaly judgment process also includes time-series pattern matching, comparing the current fluctuation data with a historical normal pattern library to detect abnormal segments deviating from the baseline. The judgment result and anomaly description are output, such as "accelerated SEI growth leads to increased internal resistance" or "lithium deposition causes a sudden capacity drop." A severity level (e.g., 1-5) is assigned to each anomaly type to ensure the accuracy and operability of the judgment, while providing refined input for subsequent early warnings.

[0008] S105: If an anomaly is found, generate a battery life warning command and adjust the charging strategy of the preset battery pack.

[0009] For example, if an anomaly is detected, a battery life warning command is generated, and the charging strategy for the preset battery pack is adjusted. The battery life warning command is categorized by anomaly level and type, with different priorities. For instance, a low-level warning (levels 1-2) prompts a reduction in charging rate or a pause in fast charging, while a high-level warning (levels 3-5) triggers a user interface alarm and recommends immediate maintenance. The charging strategy is dynamically adjusted based on the warning command, employing model predictive control to optimize the charging parameters of the equivalent circuit model. This includes adjusting the maximum charging current (e.g., reducing it from 2C to 1C), the upper voltage limit (e.g., reducing it from 4.2V to 4.0V), or enabling an active balancing mechanism to balance differences between battery packs. The adjustment process is fed back to the front-end model in real time for online optimization, such as updating the weights of the dual-branch framework through incremental learning or recalibrating the equivalent circuit model parameters to form a closed-loop control that adapts to battery degradation. Simultaneously, warning logs and strategy change history are recorded to support subsequent analysis and auditing.

[0010] This invention proposes an innovative framework that deeply integrates physical knowledge and data-driven approaches, enabling accurate assessment and proactive early warning of battery health status through a systematic, multi-step processing flow.

[0011] A battery life warning system, based on the aforementioned warning method, includes: The signal analysis module is used to collect the charge and discharge measurement signals of each battery in the preset battery pack, and determine the charge and discharge curve, charge state and health state of the battery based on the charge and discharge measurement signals and the preset equivalent circuit model. The discharge analysis module is used to predict the remaining discharge cycle number based on the charge-discharge curve, the charge state, and the health state using a dual-branch framework. The dual-branch framework includes a physical branch and a data-driven branch. The physical branch embeds an interface growth model of a solid electrolyte, and the data-driven branch uses an dilated convolutional neural network. The fluctuation analysis module is used to generate a charging fluctuation diagram based on the remaining discharge cycle number and the preset equivalent circuit model. An anomaly detection module is used to determine whether the battery is abnormal based on the charging fluctuation diagram and the remaining discharge cycle number. The instruction generation module is used to generate a battery life warning instruction if an anomaly is found, and to adjust the charging strategy of the preset battery pack.

[0012] An energy storage device, comprising a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the steps of the early warning method as described in any of the preceding embodiments.

[0013] A computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the early warning method as described in any of the preceding claims.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes a dual-branch framework combined with a physical model and an extended convolutional neural network to effectively leverage prior knowledge of equivalent circuit models and solid electrolyte interface growth mechanisms. Simultaneously, it learns degradation characteristics from data, enabling high-precision prediction of remaining discharge cycles even under limited early-stage data conditions. Furthermore, it generates early warning commands based on charging fluctuation maps and real-time anomaly detection, achieving agile response to battery anomalies and dynamic optimization of charging strategies, significantly improving the reliability and adaptability of the range warning system. Attached Figure Description

[0015] Figure 1 A schematic flowchart illustrating the battery life warning method; Figure 2 This is a schematic block diagram of a battery life warning system. Detailed Implementation

[0016] To more clearly illustrate the technical solution of this application, the technical solution of this application will be described below through specific embodiments. It should be noted that the specific embodiments are used to expand the description of the technical solution of this application, and are not intended to limit this application.

[0017] A battery life warning method, referring to Figure 1 , In some embodiments, determining the charge-discharge curve, state of charge, and health state of the battery based on the charge-discharge measurement signal and a preset equivalent circuit model includes: performing dynamic response simulation on the charge-discharge measurement signal based on the equivalent circuit model to generate a charge-discharge curve; integrating the charge-discharge curve to obtain the state of charge of the battery; and determining the health state of the battery based on the real-time state of charge and the internal resistance parameters of the equivalent circuit model.

[0018] For example, dynamic response simulation of charge-discharge measurement signals is performed based on an equivalent circuit model to generate charge-discharge curves. The equivalent circuit model adopts a second-order RC network structure including ohmic internal resistance, polarization resistance, and double-layer capacitance. The dynamic characteristics of the battery are described by establishing a state-space equation. The differential equation is solved using a fourth-order Runge-Kutta numerical integration method with a step size of 1 millisecond to simulate the voltage response behavior of the battery under varying current conditions, generating high-precision charge-discharge curves containing time, voltage, and current values. The charge-discharge curves are integrated, and the charge change is accumulated by combining the ampere-hour integration method with real-time current sampling values. The integration results are corrected by a calibration curve of open-circuit voltage and charge state (based on polynomial fitting), and a real-time charge state value with an accuracy within ±2% is calculated. The battery health status is determined based on the real-time state of charge and the internal resistance parameters of the equivalent circuit model. The internal resistance parameters are obtained by fitting the electrochemical impedance spectroscopy test data using a nonlinear least squares method. The health status is calculated using a capacity decay model. By comparing the ratio of the current maximum usable capacity (calculated based on full charge and discharge test data) to the rated capacity, combined with the internal resistance growth model (e.g., for every 10% increase in internal resistance, the health status decreases by 8%), the degree of battery degradation is comprehensively evaluated. The output is a complete set of battery state parameters including charge and discharge curves, state of charge values ​​(0-100% range), and health status values ​​(0-100% range).

[0019] In some embodiments, the dual-branch framework further includes a hidden layer and an output layer. Based on the dual-branch framework, predicting the remaining discharge cycles according to the charge-discharge curve, charge state, and health state includes: extracting degradation features from the charge-discharge curve, charge state, and health state through an expanded convolutional neural network in the data-driven branch; generating physical consistency features through an interface growth model, wherein the interface growth model calculates degradation parameters based on the Arrhenius equation; fusing the degradation features and physical consistency features in the hidden layer to obtain a fused feature vector; and predicting the remaining discharge cycles through the output layer based on the fused feature vector.

[0020] For example, degradation features are extracted from charge-discharge curves, charge states, and health states using an expanded convolutional neural network (ECN) in the data-driven branch. The ENN employs an 8-layer deep architecture, with each layer containing 64 3×1 convolutional kernels. The expansion rate gradually increases from 1 to 8 to capture multi-scale temporal features. ReLU activation and batch normalization are used to enhance nonlinear expressive power, and skip connections prevent gradient vanishing, outputting a 512-dimensional degradation feature vector. Physical consistency features are generated using an interface growth model based on the Arrhenius equation, incorporating a temperature compensation factor (calculated according to the Arrhenius formula) and a reaction rate constant (obtained through fitting experimental data). The model simulates the change in solid electrolyte interface layer thickness with cycle number and temperature at a 1-minute time step, outputting a physical feature vector containing interface thickness, lithium-ion loss, and degradation rate. Degradation features and physical consistency features are fused in the hidden layer. A multi-head attention mechanism is used to calculate the correlation weights between features, generating a 1024-dimensional fused feature vector through weighted summation. Dropout (with a dropout rate of 0.2) is used to prevent overfitting. The remaining discharge cycles are predicted by the output layer based on the fused feature vector. The output layer adopts a fully connected structure with the Sigmoid activation function to map the features to the 0-1000 cycle range and outputs the integer value of the remaining discharge cycles and its 90% confidence interval.

[0021] In some embodiments, generating a charging fluctuation map based on the remaining discharge cycle number and a preset equivalent circuit model includes: adjusting the charging parameters of the equivalent circuit model based on the remaining discharge cycle number to generate balanced charging information; constructing a charging simulation process based on the balanced charging information, and collecting charging fluctuation data during the charging simulation process; performing time series analysis on the charging fluctuation data to obtain a charging fluctuation change chart. For example, the charging parameters of the equivalent circuit model are adjusted according to the remaining discharge cycle number. These parameters include the maximum allowable charging current (adjustable from 0.1C to 2C), the cutoff voltage (adjustable from 3.0V to 4.2V), and the temperature compensation coefficient (range -0.5mV / °C to -1.5mV / °C). For instance, when the remaining discharge cycle number is below 150, the charging current is linearly reduced from 1.2C to 0.5C, and the cutoff voltage is adjusted from 4.2V to 4.0V. A charging simulation process is constructed based on balanced charging information. The state equations of the equivalent circuit model are solved using the Euler method. The charging behavior under the adjusted parameters is simulated in 10-millisecond steps, and the voltage, current, and temperature values ​​at each time point are recorded. Charging fluctuation data is collected during the charging simulation. A 14-bit ADC is used to synchronously acquire voltage and current signals at a 1kHz sampling rate, forming a high-resolution time series containing 10,000 data points. Time series analysis was performed on the charging fluctuation data. Wavelet transform was used to extract time-frequency features, and statistical indicators such as root mean square error, peak factor and waveform factor were calculated. High-frequency noise was eliminated by moving average filtering, and a chart of charging fluctuation changes containing time domain, frequency domain and statistical features was generated.

[0022] In some embodiments, determining whether a battery is abnormal based on a charging fluctuation graph and the remaining discharge cycle count includes: extracting fluctuation features from the charging fluctuation graph, including voltage drop peaks and current oscillation frequencies; comparing the fluctuation features with a preset threshold to determine the abnormality level, and marking it as a preliminary abnormality if it exceeds the preset threshold; performing a risk assessment based on the remaining discharge cycle count to determine the risk level, and determining that the battery is degraded abnormally if the remaining discharge cycle count is lower than a safety threshold and a preliminary abnormality exists; and generating an abnormality judgment result based on the abnormality level and the risk level.

[0023] For example, fluctuation features are extracted from the charging fluctuation graph. These features include voltage drop peaks (obtained by finding local minima and calculating the difference from the reference voltage, with an accuracy of 0.01V) and current oscillation frequency (analyzed using Fast Fourier Transform to identify the main frequency components, with a resolution of 0.1Hz). Auxiliary features such as rise time, overshoot, and steady-state error are also extracted. The fluctuation features are compared with preset thresholds: a voltage drop peak threshold of 0.3V (exceeding this threshold indicates an anomaly), and a current oscillation frequency threshold of 50Hz (exceeding this threshold indicates an anomaly). Anomalies are classified into levels 1-5 based on the degree of exceedance (e.g., 20% exceedance is level 1, 100% exceedance is level 5). Risk assessment is performed based on the remaining discharge cycle count, with a safety threshold of 100 cycles (adjustable according to battery type). A risk matrix model maps the anomaly level to the remaining cycle count as a risk value (range 0-100). For example, an anomaly level of 3 with 80 remaining cycles results in a risk value of 75. If the remaining discharge cycle count is below the safety threshold and a preliminary anomaly exists (risk value greater than 60), the battery is determined to have degradation anomalies. Based on the anomaly level and risk level, an anomaly judgment result is generated, which includes the anomaly type (such as SEI growth anomaly, lithium deposition, etc.), severity (level 1-5), and handling suggestions (such as reducing the charging rate, equalization maintenance, etc.).

[0024] In some embodiments, if an anomaly is detected, a battery life warning instruction is generated, and the charging strategy of the preset battery pack is adjusted. This includes: generating a battery life warning instruction based on the anomaly detection result, wherein the battery life warning instruction includes: reducing the charging rate or triggering equalization charging; adjusting the charging strategy of the preset battery pack based on the battery life warning instruction; outputting the battery life warning instruction to the user interface or control system, and recording the warning log for historical analysis.

[0025] For example, a range warning command is generated based on the anomaly assessment result. This command uses a structured data format, including the warning level (levels 1-5 correspond to different color codes), the effective timestamp (accurate to milliseconds), and specific operation commands (such as "reduce charging current to 0.5C", "start active balancing", etc.). The command data is verified using CRC to ensure transmission reliability. Based on the range warning command, the charging strategy of the preset battery pack is adjusted. Parameters are dynamically adjusted by modifying the control register values ​​of the battery management system (such as the charging current register and voltage limit register), while simultaneously triggering the equalization charging function (using a flying capacitor or inductor equalization topology) to eliminate differences between batteries within the pack. The range warning command is output to the user interface or control system, transmitting command data packets at a rate of 500kbps using the CAN 2.0B protocol, and storing warning logs in local non-volatile memory. The warning logs are recorded for historical analysis. The log content includes timestamps, battery serial numbers, warning commands, parameter changes before and after adjustment, and execution results. SQL queries and data visualization analysis are supported for subsequent model optimization and warning strategy improvement.

[0026] In some embodiments, the charge / discharge measurement signals include: charging voltage, charging current, charging temperature, discharging voltage, discharging current, and discharging temperature.

[0027] For example, the charge and discharge measurement signals include charging voltage, charging current, charging temperature, discharging voltage, discharging current, and discharging temperature. The charging voltage range covers 2.7V to 4.2V with a sampling accuracy of ±0.1%. The charging current is measured using a Hall sensor with a range of -50A to +50A. The temperature is measured using an NTC thermistor with an accuracy of ±0.5°C. All signals are synchronously acquired at a sampling frequency of 100Hz and noise interference is eliminated through digital filtering to form a time series dataset.

[0028] A battery life warning system based on a battery management system (BMS) is disclosed. The BMS-based battery life warning device 200 is used to execute the aforementioned BMS-based battery life warning method. The BMS-based battery life warning device 200 can be configured in a server.

[0029] The server can be a standalone server, a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0030] like Figure 2 As shown, the battery life warning device 200 based on BMS includes: a signal analysis module 201, a discharge analysis module 202, a fluctuation analysis module 203, an anomaly judgment module 204, and an instruction generation module 205.

[0031] The signal analysis module 201 is used to collect the charge and discharge measurement signals of each battery in the preset battery pack, and determine the charge and discharge curve, state of charge and health of the battery based on the charge and discharge measurement signals and the preset equivalent circuit model.

[0032] The discharge analysis module 202 is used to predict the remaining discharge cycle number based on the charge-discharge curve, charge state and health state according to the dual-branch framework. The dual-branch framework includes a physical branch and a data-driven branch. The physical branch embeds the interface growth model of the solid electrolyte, and the data-driven branch adopts an expanded convolutional neural network.

[0033] The fluctuation analysis module 203 is used to generate a charging fluctuation diagram based on the remaining discharge cycle number and the preset equivalent circuit model.

[0034] The anomaly detection module 204 is used to determine whether there is an anomaly in the battery based on the charging fluctuation graph and the remaining discharge cycle number.

[0035] The instruction generation module 205 is used to generate a battery life warning instruction and adjust the charging strategy of the preset battery pack if an abnormality occurs.

[0036] This application provides an energy storage device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement a battery life warning method based on a BMS as described in any of the embodiments of this application.

[0037] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it causes the processor to implement a battery life warning method based on a BMS, as described in any of the embodiments of this application.

[0038] This invention achieves accurate assessment and proactive early warning of battery health status through a systematic, multi-step processing flow by deeply integrating physical knowledge with a data-driven innovative framework.

[0039] This invention utilizes a dual-branch framework combined with a physical model and an extended convolutional neural network to effectively leverage prior knowledge of equivalent circuit models and solid electrolyte interface growth mechanisms. Simultaneously, it learns degradation characteristics from data, enabling high-precision prediction of remaining discharge cycles even under limited early-stage data conditions. Furthermore, it generates early warning commands based on charging fluctuation maps and real-time anomaly detection, achieving agile response to battery anomalies and dynamic optimization of charging strategies, significantly improving the reliability and adaptability of the range warning system.

[0040] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A battery endurance warning method, characterized by, The battery endurance early warning method comprises: S101: collecting the charge-discharge measurement signals of each battery in the preset battery pack, and determining the charge-discharge curve, state of charge and state of health of the battery according to the charge-discharge measurement signals and a preset equivalent circuit model; S102: predicting the remaining discharge cycle number according to the charge-discharge curve, the state of charge and the state of health based on a double-branch framework, wherein the double-branch framework comprises: a physical branch and a data-driven branch, the physical branch is embedded with an interface growth model of a solid electrolyte, and the data-driven branch adopts an expanded convolutional neural network; S103: generating a charging fluctuation map according to the remaining discharge cycle number and the preset equivalent circuit model; S104: determining whether the battery is abnormal according to the charging fluctuation map and the remaining discharge cycle number; S105: if abnormal, generating an endurance early warning instruction and adjusting the charging strategy of the preset battery pack.

2. The battery endurance warning method of claim 1, wherein, The double-branch framework further comprises: a hidden layer and an output layer, and the remaining discharge cycle number is predicted according to the charge-discharge curve, the state of charge and the state of health based on the double-branch framework, comprising: extracting degradation features from the charge-discharge curve, the state of charge and the state of health through the expanded convolutional neural network in the data-driven branch; generating physical consistency features through the interface growth model, wherein the interface growth model calculates degradation parameters based on the Arrhenius equation; fusing the degradation features and the physical consistency features in the hidden layer to obtain a fusion feature vector; predicting the remaining discharge cycle number based on the fusion feature vector through the output layer.

3. The method of claim 1, wherein, The charge-discharge measurement signals comprise: charging voltage, charging current, charging temperature, discharging voltage, discharging current and discharging temperature.

4. The method of claim 1, wherein, The charging fluctuation map comprises voltage sudden drop peak value and current oscillation frequency.

5. The method of claim 1, wherein, The endurance early warning instruction comprises reducing the charging rate or triggering equalization charging.

6. A battery endurance warning system based on the warning method of claim 1, characterized by, Comprise: The signal analysis module is configured to collect the charge-discharge measurement signals of each battery in the preset battery pack, and determine the charge-discharge curve, state of charge and state of health of the battery according to the charge-discharge measurement signals and a preset equivalent circuit model; The discharge analysis module is configured to predict the remaining discharge cycle number according to the charge-discharge curve, the state of charge and the state of health based on a double-branch framework, wherein the double-branch framework comprises: a physical branch and a data-driven branch, the physical branch is embedded with an interface growth model of a solid electrolyte, and the data-driven branch adopts an expanded convolutional neural network; The fluctuation analysis module is configured to generate a charging fluctuation map according to the remaining discharge cycle number and the preset equivalent circuit model; The abnormality determination module is configured to determine whether the battery is abnormal according to the charging fluctuation map and the remaining discharge cycle number; The instruction generation module is configured to generate an endurance early warning instruction and adjust the charging strategy of the preset battery pack if abnormal.

7. An energy storage device, characterized by, Comprise a memory and a processor; The memory is configured to store a computer program; The processor is configured to execute the computer program and implement the steps of the early warning method according to any one of claims 1-5 when executing the computer program.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is configured to implement the steps of the early warning method according to any one of claims 1-5 when executed by the processor.

Images