Lithium battery test method and system, terminal and medium

Abstract

The invention relates to the field of battery testing, and particularly provides a lithium battery testing method and system, a terminal and a medium, and the method comprises the steps: obtaining nominal parameters, full-life-cycle historical data and multi-mode battery state data of a tested battery; based on the data, a personalized charging and discharging parameter base line is dynamically generated through an expert rule base, and the tested battery is tested based on the base line; in the test execution period, based on the multi-modal time sequence data, the health state of the tested battery is predicted through the multi-scale time sequence-frequency domain fusion network model so as to judge whether a closed-loop correction process is triggered or not; and responding to the starting of the closed-loop correction process, generating an active test script through a Bayesian optimizer based on a prediction result, and issuing the active test script to edge end test equipment for execution. The test efficiency, the test precision and the test information acquisition amount are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of battery testing, in particular to a lithium battery testing method, system, terminal and medium. BACKGROUND

[0002] The conventional lithium battery testing method usually adopts a fixed and preset charging and discharging process and relies on manual data recording and abnormality judgment. This method is inefficient, error-prone, and cannot adapt to different characteristics of different individual batteries due to differences in factory batch, aging degree, use history and real-time environmental temperature, and the accuracy and reliability of the test results are limited.

[0003] The related art solves the defects of the conventional testing method based on automatic testing equipment, and realizes the automatic execution of the charging and discharging process and the data recording through computer control. Although this kind of scheme reduces the manual burden and realizes the automatic acquisition of data, the core of its testing logic is still in the paradigm of static preset. Specifically, the current, voltage threshold, cutoff condition and other key parameters of the charging and discharging are still preset uniformly, and cannot be dynamically adjusted according to the real-time state of the measured object, thereby leading to a lack of adaptability, inability to customize individualized test stress for batteries in different states of health (SOH), different cycle times or different environmental temperatures, and possible over-testing of aged batteries or insufficient testing of healthy batteries; lack of intelligence, the testing process is open-loop, only passive data recording is possible, and online intelligent decision-making and dynamic optimization of the testing process based on real-time data is not possible; lack of depth, the testing purpose is limited to verifying whether the battery meets the preset specifications, and the potential failure modes and boundary performance of the battery cannot be actively stimulated and detected. SUMMARY

[0004] To solve the above problems, the present application provides a lithium battery testing method, system, terminal and medium to improve testing efficiency, testing accuracy and testing information acquisition.

[0005] In a first aspect, the technical solution of the present application provides a lithium battery testing method, comprising the following steps: obtaining the nominal parameters and the full life cycle historical data of the measured battery; real-time acquisition of multi-modal battery state data of the measured battery to form multi-modal time series data; based on the nominal parameters, the full life cycle historical data and the multi-modal battery state data, dynamically calculating and generating an individualized charging and discharging parameter baseline through an embedded expert rule base; downloading the individualized charging and discharging parameter baseline to the edge testing device to test the measured battery; During the test execution, the battery health state of the tested battery is predicted by a multi-scale time-frequency domain fusion network model based on real-time multi-modal time series data, and it is judged whether to trigger a closed-loop correction process according to the prediction result. If yes, the closed-loop correction process is started, and if no, the test is executed normally, and the multi-modal battery state data of the tested battery is continuously collected in real time. In response to the start of the closed-loop correction process, an active test script is generated by a Bayesian optimizer based on the prediction result, and the active test script is sent to the edge test equipment for execution.

[0006] In an optional embodiment, based on the nominal parameters, the full life cycle historical data and the multi-modal battery state data, the personalized charging and discharging parameter baseline is dynamically calculated and generated through the built-in expert rule library, specifically including: Read the target parameters in the nominal parameters of the tested battery, including the nominal maximum charging current, the nominal maximum discharging current, the nominal constant voltage switching voltage of constant current to constant voltage charging and the nominal discharging cutoff voltage, and take these nominal values as the initial reference for parameter calculation; Read the real-time battery surface temperature in the multi-modal battery state data, and determine the current compensation coefficient under the current surface temperature according to the pre-stored temperature-coefficient mapping table in the expert rule library; Read the historical health state and cycle number contained in the full life cycle historical data, and calculate the comprehensive life compensation coefficient through the pre-defined attenuation model in the expert rule library. The calculation logic of the comprehensive life compensation coefficient is configured as: comprehensive life compensation coefficient=f(historical health state, cycle number), and the function f is constructed as a monotonically decreasing function of the historical health state and the cycle number; Read the battery type in the nominal parameters of the tested battery, and query the corresponding chemical system correction factor from the expert rule library; Calculate the final adaptive parameters according to the initial reference, the current compensation coefficient, the comprehensive life compensation coefficient and the chemical system correction factor, and generate a personalized charging and discharging parameter baseline containing all adaptive parameters.

[0007] In an optional embodiment, the final adaptive parameters are calculated according to the initial reference, the current compensation coefficient, the comprehensive life compensation coefficient and the chemical system correction factor, specifically including: Calculate the adaptive maximum charging current=nominal maximum charging current x current compensation coefficient x comprehensive life compensation coefficient; Calculate the adaptive maximum discharging current=nominal maximum discharging current x current compensation coefficient x comprehensive life compensation coefficient; Calculate the adaptive CV switching voltage=nominal constant voltage switching voltage of constant current to constant voltage charging x(1-α x(1-comprehensive life compensation coefficient)) x chemical system correction factor, where α is an empirical attenuation coefficient; Calculate the adaptive discharge cut-off voltage = nominal discharge cut-off voltage × (1 + β × (1-comprehensive life compensation coefficient)), where β is the empirical protection coefficient.

[0008] In an optional embodiment, based on real-time multimodal time series data, a multi-scale time series-frequency domain fusion network model is used to predict the health status of the tested battery. Based on the prediction results, it is determined whether to trigger a closed-loop correction process, which specifically includes: The sliding time window data of a preset length of multimodal time series data is used as the input of the multi-scale time series-frequency domain fusion network model, and the multi-scale time series-frequency domain fusion network model processes and outputs the predicted health status; Get the actual health status of the battery under test; Compare the deviation between the predicted health state and the actual health state. If the deviation exceeds the threshold, the closed-loop correction process is triggered. Otherwise, the closed-loop correction process is not triggered and the test is performed normally.

[0009] In an optional embodiment, the structure of the multi-scale time-frequency fusion network model includes, in sequence, One-dimensional convolution front-end: three layers of causal convolution for local feature extraction; Frequency domain branch: Perform short-time Fourier transform on the input sequence, obtain the amplitude spectrum, and then feed it into two layers of 2-D convolution and maximum pooling to output frequency domain features; Time domain branch: feeds the one-dimensional convolution front-end output into the bidirectional LSTM to output time domain features; Cross-attention fusion: Using time domain features as query and frequency domain features as key-value, a multi-head attention mechanism is implemented to output fused features; Task head: fully connected network mapped to healthy state.

[0010] In an optional embodiment, the method further comprises: During the execution of the closed-loop correction process, the newly generated multimodal time series data is stored in the experience playback buffer; The newly generated multimodal time series data is used to fine-tune the weights of the last K layers of the multi-scale time series-frequency domain fusion network model to obtain a new multi-scale time series-frequency domain fusion network model; The performance of the new multi-scale time-frequency fusion network model is evaluated on the retained validation set. If its predicted mean absolute percentage error is less than the threshold and no deviation alarm is triggered for several consecutive time windows, the closed-loop correction is considered to have converged. Otherwise, the new multi-scale time-frequency fusion network model is used to predict the health status of the battery under test, and then a new active test script is generated for iterative correction.

[0011] In an optional embodiment, a proactive test script is generated based on the prediction results using a Bayesian optimizer, specifically including: The problem of generating active test scripts is formalized as a constrained optimization problem, including defining optimization variables, constructing an objective function, and defining constraints. The optimization variables are the script parameter vectors to be optimized, and the dimensions of the vectors correspond to the candidate stress modes, including excitation current rate, stress temperature, pulse width, and number of cycles. The objective function aims to maximize the rate of decrease of model prediction error or maximize the model information gain. The constraints define the safety boundary constraints for test safety, including that the maximum surface temperature of the battery must not exceed the absolute safety limit, the maximum voltage during the test must not exceed the upper limit of the electrochemical window, the minimum voltage must not be lower than the lower limit, and the total capacity decay must not exceed the maximum allowable capacity decay threshold. Call the Bayesian optimizer to iteratively solve the defined constrained optimization problem and output the global optimal solution stress parameter vector; The global optimal solution stress parameter vector and the target stress pattern identifier are compiled into an active test script.

[0012] In a second aspect, the technical solution of the present invention provides a lithium battery testing system, comprising: Static data acquisition module, used to obtain the nominal parameters and life cycle historical data of the tested battery; Dynamic data acquisition module, used to collect multimodal battery status data of the tested battery in real time to form multimodal time series data; A charge and discharge parameter baseline generation module, which dynamically calculates and generates personalized charge and discharge parameter baselines based on nominal parameters, full lifecycle historical data, and multimodal battery status data using a built-in expert rule library. The battery testing module is used to send the personalized charge and discharge parameter baseline to the edge test equipment to test the battery under test; The deviation detection module is used to predict the health status of the battery under test based on real-time multimodal time series data through a multi-scale time series-frequency domain fusion network model during test execution. The module determines whether to trigger a closed-loop correction process based on the prediction results. If so, the closed-loop correction process is initiated. If not, the test is performed normally and the multimodal battery status data of the battery under test is continued to be collected in real time. The active test module is used to respond to the start of the closed-loop correction process, generate active test scripts based on the prediction results through the Bayesian optimizer, and send the active test scripts to the edge test device for execution.

[0013] In a third aspect, the technical solution of the present invention provides a terminal, including: A memory for storing a lithium battery test program; A processor is configured to implement the steps of any of the above-mentioned lithium battery testing methods when executing the lithium battery testing program.

[0014] In a fourth aspect, the technical solution of the present invention provides a computer-readable storage medium, on which a lithium battery testing program is stored. When the lithium battery testing program is executed by a processor, the steps of the lithium battery testing method as described in any one of the above items are implemented.

[0015] It can be seen from the above technical solutions that this application has the following advantages: 1. By introducing an expert rule base, this application can integrate the battery's nominal parameters, historical full life cycle data, and real-time multimodal sensor data, dynamically calculate and generate a charge and discharge parameter baseline that best suits the current individual state of the battery being tested, and change the static test mode so that the test intensity matches the battery's current "physical condition." This not only protects aging batteries from overstress damage, but also fully taps the performance of healthy batteries within a safe range, thereby improving the accuracy and safety of the test.

[0016] 2. This application deploys a multi-scale time-frequency fusion network model to perform real-time health status prediction, and intelligently determines whether to trigger closed-loop correction based on the prediction results. This makes the test system no longer an "automated" system that simply executes instructions, but has "intelligent" decision-making capabilities. It can actively intervene in the test process based on new information emerging during the test (such as prediction deviation), and guide the test in a direction that can better reveal battery performance or problems.

[0017] 3. This application introduces active test scripts generated based on a Bayesian optimizer, which can actively apply optimized and targeted stress conditions (such as specific high-rate pulses and temperature cycles) to the battery. The active testing method can effectively stimulate potential defects and failure modes that are difficult to expose under standard testing procedures, thereby achieving a more in-depth evaluation of the battery's boundary performance and reliability, and improving the effectiveness of the test.

[0018] 4. This application deeply integrates personalized baselines, online intelligent decision-making and active detection to form an efficient intelligent test closed loop. On the basis of liberating manpower, it further achieves a comprehensive improvement in test efficiency, test accuracy and test information acquisition, providing technical means for high-precision sorting, in-depth quality assessment and failure analysis of lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for the description. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 A lithium battery test method flowchart is provided for the embodiment of the present application.

[0021] Figure 2 A lithium battery test system structure schematic diagram is provided for the embodiment of the present application.

[0022] Figure 3 A terminal structure schematic diagram is provided for the embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to make the application purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions protected by the present application will be described in detail below with specific embodiments and drawings. Obviously, the following described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0024] Unless otherwise defined, all technical and scientific terms used in the present application have the same meanings as those commonly understood by those skilled in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments of the present application, and are not intended to limit the present application.

[0025] Figure 1 A lithium battery test method flowchart is provided for the embodiment of the present application. Among them, Figure 1 The execution subject can be a lithium battery test system. The lithium battery test method provided by the embodiment of the present application is executed by a computer device, and accordingly, the lithium battery test system runs in the computer device. According to different needs, the order of steps in the flowchart can be changed, and some can be omitted.

[0026] As Figure 1 shown, the method comprises the following steps.

[0027] S1, obtaining the nominal parameters and full life cycle historical data of the battery to be tested.

[0028] S2, real-time acquisition of multi-modal battery state data of the battery to be tested, to form multi-modal time series data.

[0029] S3, based on the nominal parameters, full life cycle historical data and multi-modal battery state data, dynamically calculating and generating individualized charging and discharging parameter baseline through the built-in expert rule base.

[0030] S4, issuing the individualized charging and discharging parameter baseline to the edge test device to test the battery to be tested.

[0031] S5, during the test execution, based on the real-time multi-modal time series data, the battery health state under test is predicted by a multi-scale time series-frequency domain fusion network model, and it is judged whether to trigger a closed-loop correction process according to the prediction result, if yes, the closed-loop correction process is started, if not, the test is executed normally, and the multi-modal battery state data of the battery under test is continuously collected in real time.

[0032] S6, in response to the start of the closed-loop correction process, an active test script is generated by a Bayesian optimizer based on the prediction result, and the active test script is issued to the edge test device for execution.

[0033] As a refinement and extension of the above embodiment, in order to fully describe the specific implementation process in this embodiment, the following will give a non-limiting description of the specific implementation of the above steps.

[0034] A local area network test environment is built, and a control computer is connected to the battery test system device through a USB line and a signal line. The control computer can be connected to a local area network or the Internet for remote control. The control computer is installed with test software, and key parameter data information can be obtained by collecting test software log information to draw dynamic change curves, such as voltage and current dynamic change curves, thereby improving human-computer interaction. During the test process, abnormal signals such as excessive current, failure to reach 50% of the battery capacity after 5 hours of charging, frequent switching of indicator light signals during charging, and stagnation after charging to 75% are collected. When the parameter value exceeds the set normal interval range, the log throws and records error information, and simulates rapid and continuous pressing of the Shift key five times to trigger the control computer buzzer to emit sound, reminding the test personnel of the occurrence of abnormal scenarios and avoiding test accidents. After the test process is displayed, the test log file can be read to capture the alarm information, battery capacity, battery consumption, and test time during the test process. According to these key parameter information, the performance of the lithium battery can be quickly understood.

[0035] The nominal parameters include nominal capacity, nominal voltage, battery type, and nominal maximum charging current (I_charge_standard), nominal maximum discharging current (I_discharge_standard), nominal constant current to constant voltage charging switching voltage (V_cv_standard), and nominal discharging cutoff voltage (V_cutoff_standard).

[0036] The digital twin platform database is used to query the full lifecycle historical data associated with the tested battery's unique identifier (e.g., QR code), including historical state of health (SOH_historical) and cycle count. Cycle count is a core indicator of battery aging and lifespan consumption. It refers to the total number of times a battery completes a full 100% deep charge and discharge cycle.

[0037] The battery testing system and its integrated multi-sensor system simultaneously collect multimodal battery status data from the battery under test at high frequency (sampling period Δt1 ≤ 1 s), generating multimodal time series data. This data includes real-time voltage, current, surface temperature, expansion force, and ultrasonic time-of-flight. This encrypted data is uploaded to the cloud-based digital twin platform in real time.

[0038] In some optional embodiments, step S3 dynamically calculates and generates a personalized charge and discharge parameter baseline based on nominal parameters, full life cycle historical data and multimodal battery status data through a built-in expert rule library, specifically including the following steps.

[0039] S3.1, read the target parameters from the nominal parameters of the battery under test.

[0040] The target parameters include the nominal maximum charge current (I_charge_standard), the nominal maximum discharge current (I_discharge_standard), the nominal constant current to constant voltage charging switching voltage (V_cv_standard), and the nominal discharge cut-off voltage (V_cutoff_standard). These nominal values ​​are used as the initial benchmark for parameter calculation.

[0041] S3.2, read the real-time battery surface temperature in the multimodal battery status data, and determine the current compensation coefficient at the current surface temperature according to the temperature-coefficient mapping table pre-stored in the expert rule library.

[0042] The battery surface temperature is T_current, the current compensation coefficient is k_T, and the temperature-coefficient mapping table is configured as follows: When T_current is lower than the first temperature threshold T_low, k_T takes a value less than 1 to prevent lithium precipitation; When T_current is in the optimal temperature range [T_opt_low, T_opt_high], k_T takes a value of 1 or slightly greater than 1 to allow full-rate or accelerated testing; When T_current is higher than the second temperature threshold T_high, k_T takes a value smaller than 1 again to suppress the risk of thermal runaway.

[0043] S3.3, read the historical health status and cycle count contained in the full life cycle history data, calculate the comprehensive life compensation coefficient through the pre-defined attenuation model in the expert rule base.

[0044] The calculation logic of the comprehensive life compensation coefficient is configured as: comprehensive life compensation coefficient = f (historical health status, cycle count), the function f is constructed as a monotonically decreasing function with respect to the historical health status and cycle count, so that for the battery with low health degree (small value of historical health status SOH_historical) or high cycle count (large value of cycle count Cycle_count), the calculation result of the comprehensive life compensation coefficient k_SOH tends to be less than 1, so as to reduce the test stress and slow down the aging process.

[0045] S3.4, read the battery type in the nominal parameters of the battery under test, and query the corresponding chemical system correction factor from the expert rule base.

[0046] According to the battery type Type, the corresponding chemical system correction factor δ_Type is queried from the expert rule base. The correction factor is used to fine-tune the voltage threshold sensitivity of different chemical system batteries, for example: For lithium iron phosphate (LFP) battery, its δ_Type is configured to be insensitive to voltage change, mainly controlled by capacity and temperature; For ternary lithium (NMC) battery, its δ_Type is configured to be highly sensitive to voltage change, and the voltage window needs to be strictly adjusted.

[0047] S3.5, calculate the final adaptive parameters according to the initial reference, current compensation coefficient, comprehensive life compensation coefficient and chemical system correction factor, and generate a personalized charge and discharge parameter baseline containing all adaptive parameters, including: Calculate the adaptive maximum charging current = nominal maximum charging current × current compensation coefficient × comprehensive life compensation coefficient; Calculate the adaptive maximum discharging current = nominal maximum discharging current × current compensation coefficient × comprehensive life compensation coefficient; Calculate the adaptive CV switching voltage = nominal constant current to constant voltage charging switching voltage × (1-α×(1-comprehensive life compensation coefficient)) × chemical system correction factor, where α is an empirical attenuation coefficient; Calculate the adaptive discharge cut-off voltage = nominal discharge cut-off voltage × (1+β×(1-comprehensive life compensation coefficient)), where β is an empirical protection coefficient.

[0048] The generated personalized charge and discharge parameter baseline set is issued to the battery test system device as the execution reference of the current test period, ensuring the safety and basic personalization of the initial stage of the test.

[0049] In some optional embodiments, in step S5, the battery health state to be measured is predicted based on real-time multi-modal time series data by a multi-scale time series-frequency domain fusion network model, and whether to trigger a closed-loop correction process is determined according to a prediction result, specifically including the following steps.

[0050] S5.1. The multi-scale time series-frequency domain fusion network model is input with sliding time window data of a preset length of multi-modal time series data, and the health state is predicted by processing the multi-scale time series-frequency domain fusion network model.

[0051] S5.2. The actual health state of the measured battery is obtained.

[0052] S5.3. The deviation between the predicted health state and the actual health state is compared, and if the deviation exceeds a threshold value, it is determined that the closed-loop correction process is triggered, otherwise, the closed-loop correction process is not triggered, and the test is normally performed.

[0053] The structure of the multi-scale time series-frequency domain fusion network model includes, in sequence, one-dimensional convolution front end: three layers of causal convolution, with convolution kernel sizes of 5, 3, and 3, respectively, and channel numbers of 64→128→256, for local feature extraction; frequency domain branch: short-time Fourier transform is performed on the input sequence, and the amplitude spectrum is taken to be input into two layers of 2-D convolution and maximum pooling, and the frequency domain features are output; time domain branch: the output of the one-dimensional convolution front end is input into a bidirectional LSTM, and the time domain features are output; cross-attention fusion: taking the time domain features as Query and the frequency domain features as Key-Value, performing multi-head attention mechanism, and outputting the fusion features; task head: full connection network mapping to health state.

[0054] It should be noted that the health state (State of Health, SoH) is a scalar index for quantifying the performance degradation of a lithium battery relative to its brand-new state, and its value is a specific numerical value in the interval [0, 1] or [0%, 100%].

[0055] The standard calculation formula is: SoH = (battery current maximum available capacity C_current / battery nominal initial capacity C_rated) × 100% SoH_k_pred (predicted health state) is output by the deployed multi-scale time series-frequency domain fusion network model in real time, and the model analyzes high-frequency voltage, current, temperature and other multi-modal battery state sensing data to indirectly infer the current capacity degradation.

[0056] SoH_k_real (actual state of health) is calculated based on the end-of-life capacity, which is the actual discharge capacity (Ah) divided by the nominal capacity (Ah) during the test.

[0057] The deviation between the predicted state of health and the actual state of health refers to the difference between the predicted SoH value and the actual SoH value.

[0058] In some optional embodiments, step S6 generates an active test script based on the prediction result through a Bayesian optimizer, specifically including the following steps.

[0059] S6.1, formalize the generation problem of the active test script as a constrained optimization problem, including defining optimization variables, constructing an objective function, and defining constraint conditions.

[0060] The optimization variable is the script parameter vector to be optimized, and the dimension of the vector corresponds to the candidate stress mode, including the excitation current rate (C-rate), stress temperature (T_stress), pulse width (t_pulse), and cycle number (N_cycle).

[0061] The objective function aims to maximize the model prediction error reduction rate or maximize the model information gain. The expression of the objective function f(x) is: f(x) = (E_before - E_after(x)) / E_before Where E_before is the average prediction error of the current model before triggering, and E_after(x) is the expected prediction error of the model after fine-tuning after injecting the stress script defined by the parameter x.

[0062] The constraint condition defines a set of safety boundary constraints for test safety, including that the maximum surface temperature of the battery must not exceed the absolute safety limit, the maximum voltage during the test must not exceed the upper limit of the electrochemical window, the minimum voltage must not be lower than the lower limit, and the total capacity attenuation must not exceed the maximum allowed capacity attenuation threshold. S6.2, call the Bayesian optimizer to iteratively solve the defined constrained optimization problem to output the globally optimal stress parameter vector.

[0063] S6.21: Surrogate Model Construction. Based on the historical stress-response data set D_historical = { (x_i, f(x_i))}, a surrogate model Surrogate Model of the objective function f(x) is constructed using Gaussian process regression, which is used to predict the objective function value and its uncertainty at unknown points x_*.

[0064] S6.22: Acquisition function maximization. Use Expected Improvement (EI) or Upper Confidence Bound (UCB) as the acquisition function a(x). The acquisition function balances exploitation and exploration, and the next optimal candidate test point x_next is recommended by solving the argmax a(x) subproblem.

[0065] S6.23: Parallel safety check. After recommending x_next, first check whether it satisfies all safety constraints g(x_next) ≤ 0 in the simulation environment. If the check fails, remove the point from the search space and return to S6.22 to recommend again; if the check passes, go to S6.24.

[0066] S6.24: Iteration and convergence. Add the recommended candidate point x_next and its target function value f(x_next) estimated by simulation or prior knowledge to the historical data set D_historical, and update the surrogate model. Repeat steps S6.21-S6.23 until the maximum number of iterations N_iter is reached or the improvement of the target function value is less than the convergence threshold τ. Finally, output the global optimal solution stress parameter vector x_optimal.

[0067] S6.3, compile the global optimal solution stress parameter vector together with the target stress mode identifier into an active test script.

[0068] Compile the optimal parameter vector x_optimal obtained by optimization together with the target stress mode identifier (indicating high rate, high temperature, or lithium precipitation excitation, etc.) into a structured active test script that can be parsed and executed by the edge controller. The script is a JSON or XML format document containing instruction sequences and parameters, which is essentially a programmed description of a series of controlled stress conditions applied to verify the specific performance of the battery or to excite potential fault modes.

[0069] In some optional embodiments, during the active test injection process, the edge side continuously performs the high-frequency data acquisition and upload of step S200, and stores the newly generated data stream into the experience replay buffer D. Then, an incremental meta-learning algorithm (such as MAML-LoRA) is used to quickly fine-tune the last K layers of the multi-scale time series-frequency domain fusion network model, so that the model can quickly adapt to the new characteristics or new emerging degradation modes of the current battery, which includes the following steps.

[0070] S7.1, during the closed-loop correction process, store the newly generated multi-modal time series data in the experience replay buffer.

[0071] S7.2, fine-tuning the last K layers of the multi-scale time-frequency fusion network model using the newly generated multi-modal time series data to obtain a new multi-scale time-frequency fusion network model.

[0072] S7.3, performing performance evaluation on the new multi-scale time-frequency fusion network model while reserving the validation set, if the predicted mean absolute percentage error is less than a threshold value and deviation alarm is not triggered in continuous time windows, it is determined that the closed-loop correction converges, otherwise, using the new multi-scale time-frequency fusion network model to predict the battery state of health of the measured battery, and then generating a new active test script for iterative correction.

[0073] The incremental meta-learning algorithm quickly fine-tunes the last K layers of the multi-scale time-frequency fusion network model, and the optimization goal is to minimize the following fine-tuning loss function L: L = L_SoH + λ·||Δθ|| 2 Wherein, L_SoH is the loss function of the state of health prediction task, and the mean absolute error (MAE) or mean square error (MSE) is used for calculation; Δθ is the incremental update of the model to be fine-tuned weight; λ is a regularization hyperparameter, used to control the model update amplitude, prevent overfitting; ||Δθ|| 2 The L2 regularization term is used to punish the size of the weight update, so as to ensure the stability of the fine-tuning process.

[0074] The above describes an embodiment of a lithium battery test method in detail, based on the lithium battery test method described in the above embodiment, the embodiment of the present application also provides a lithium battery test system corresponding to the method.

[0075] Figure 2 A structural schematic block diagram of a lithium battery test system provided by the embodiment of the present application, in the embodiment, the lithium battery test system 200 can be divided into a plurality of functional modules according to its functions. The module referred to by the present application refers to a series of computer program segments which can be executed by at least one processor and can complete fixed functions, which are stored in the memory.

[0076] The static data acquisition module 210 is used for acquiring the nominal parameters and full life cycle historical data of the measured battery.

[0077] The dynamic data acquisition module 220 is used for real-time acquisition of multi-modal battery state data of the measured battery to form multi-modal time series data.

[0078] The charge and discharge parameter baseline generation module 230 is used to dynamically calculate and generate a personalized charge and discharge parameter baseline based on nominal parameters, full life cycle historical data and multimodal battery status data through a built-in expert rule base.

[0079] The battery testing module 240 is used to send the personalized charge and discharge parameter baseline to the edge testing device to test the battery under test.

[0080] The deviation detection module 250 is used to predict the health status of the battery under test based on real-time multimodal time series data through a multi-scale time series-frequency domain fusion network model during test execution, and determine whether to trigger a closed-loop correction process based on the prediction result. If so, the closed-loop correction process is started. If not, the test is performed normally and the multimodal battery status data of the battery under test continues to be collected in real time.

[0081] The active test module 260 is used to respond to the start of the closed-loop correction process, generate an active test script through the Bayesian optimizer based on the prediction results, and send the active test script to the edge test device for execution.

[0082] In some optional embodiments, the system 200 also includes a mode adaptation module 270, which is used to store the newly generated multimodal time series data in the experience replay buffer during the execution of the closed-loop correction process; use the newly generated multimodal time series data to fine-tune the last K layers of weights of the multi-scale time series-frequency domain fusion network model to obtain a new multi-scale time series-frequency domain fusion network model; perform performance evaluation on the new multi-scale time series-frequency domain fusion network model in the retained validation set, and if its predicted average absolute percentage error is less than a threshold and no deviation alarm is triggered for several consecutive time windows, it is determined that the closed-loop correction has converged; otherwise, the new multi-scale time series-frequency domain fusion network model is used to predict the health status of the battery under test, and then generate a new active test script for iterative correction.

[0083] The lithium battery testing system of this embodiment is used to implement the aforementioned lithium battery testing method. Therefore, the specific implementation of the system can be seen in the embodiment section of the lithium battery testing method in the previous text. Therefore, its specific implementation can refer to the description of the corresponding embodiments of each part and will not be elaborated here.

[0084] In addition, since the lithium battery testing system of this embodiment is used to implement the aforementioned lithium battery testing method, its function corresponds to that of the aforementioned method and will not be described in detail here.

[0085] Figure 3A structure schematic diagram of a terminal 300 provided by the embodiment of the present application, comprising: a processor 310, a memory 320 and a communication unit 330. The processor 310 is used to realize the flow steps of the lithium battery test method embodiment when realizing the lithium battery test program stored in the memory 320.

[0086] The terminal 300 comprises the processor 310, the memory 320 and the communication unit 330. These components communicate through one or more buses. Those skilled in the art can understand that the structure of the server shown in the figure does not constitute a limitation to the present application. It can be a bus structure or a star structure. It can also comprise more or less components than shown in the figure, or combine some components, or have different component arrangement.

[0087] The memory 320 can be used to store the execution instructions of the processor 310. The memory 320 can be realized by any type of volatile or non-volatile storage terminal or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the terminal 300 can execute part or all of the steps in the following method embodiments.

[0088] The processor 310 is the control center of the storage terminal. It connects all parts of the electronic terminal through various interfaces and lines. It executes the software programs and / or modules stored in the memory 320 and calls the data stored in the memory to perform various functions of the electronic terminal and / or process data. The processor can be composed of integrated circuits (ICs). For example, it can be composed of a single packaged IC or multiple packaged ICs with the same function or different functions. For example, the processor 310 can only include a central processing unit (CPU). In the embodiment of the present application, the CPU can be a single operation core or can include multiple operation cores.

[0089] The communication unit 330 is used to establish a communication channel so that the storage terminal can communicate with other terminals. It receives user data sent by other terminals or sends user data to other terminals.

[0090] The application further provides a computer storage medium, wherein the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM) and the like.

[0091] The application further provides a computer storage medium, wherein the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM) and the like.

[0092] The computer storage medium stores a lithium battery test program, and the lithium battery test program is executed by the processor to realize the flow steps of the lithium battery test method embodiment.

[0093] Those skilled in the art can clearly understand that the technology in the embodiments of the application can be realized by means of software and necessary general hardware platforms. Based on such understanding, the technical solutions in the embodiments of the application can be embodied in the form of a software product, which is stored in a storage medium such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and the like, and includes a plurality of instructions for causing a computer terminal (which can be a personal computer, a server, or a second terminal, a network terminal, and the like) to execute all or part of the steps of the method described in the embodiments of the application.

[0094] In several embodiments provided by the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely schematic, for example, the division of the units is merely a logical function division, and an actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0095] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0096] In addition, each functional unit in various embodiments of the application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0097] The above description of disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined in this application can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to these embodiments shown in this application, but will conform to the widest scope consistent with the principles and novel features disclosed in this application.

Claims

1. A lithium battery testing method, characterized in that: The following steps are involved: Obtain the nominal parameters and life cycle historical data of the tested battery; Collect multi-modal battery status data of the tested battery in real time to form multi-modal time series data; Based on nominal parameters, full life cycle historical data and multi-modal battery status data, the built-in expert rule base dynamically calculates and generates personalized charge and discharge parameter baselines; Send the personalized charge and discharge parameter baseline to the edge test equipment to test the battery under test; During the test execution, the health status of the tested battery is predicted based on real-time multimodal time series data through a multi-scale time series-frequency domain fusion network model. Based on the prediction results, it is determined whether to trigger the closed-loop correction process. If so, the closed-loop correction process is started. If not, the test is performed normally and the multimodal battery status data of the tested battery continues to be collected in real time. The response closed-loop correction process is initiated, and an active test script is generated through the Bayesian optimizer based on the prediction results, and the active test script is sent to the edge test device for execution.

2. The lithium battery testing method according to claim 1, wherein: Based on nominal parameters, lifecycle historical data, and multimodal battery status data, the built-in expert rule base dynamically calculates and generates personalized charge and discharge parameter baselines, including: Read the target parameters from the nominal parameters of the tested battery, including the nominal maximum charge current, nominal maximum discharge current, nominal constant-current to constant-voltage charging switching voltage, and nominal discharge cut-off voltage, and use these nominal values ​​as the initial benchmark for parameter calculation; Read the real-time battery surface temperature from the multimodal battery status data and determine the current compensation coefficient at the current surface temperature based on the temperature-coefficient mapping table pre-stored in the expert rule library; The historical health status and number of cycles contained in the full life cycle historical data are read, and the comprehensive life compensation coefficient is calculated using the attenuation model predefined in the expert rule library. The calculation logic of the comprehensive life compensation coefficient is configured as follows: comprehensive life compensation coefficient = f (historical health status, number of cycles). Function f is constructed as a monotonically decreasing function of the historical health status and number of cycles. Read the battery type from the nominal parameters of the tested battery and query the corresponding chemical system correction factor from the expert rule library; The final adaptive parameters are calculated based on the initial benchmark, current compensation coefficient, comprehensive life compensation coefficient, and chemical system correction factor to generate a personalized charge and discharge parameter baseline that includes all adaptive parameters.

3. The lithium battery testing method according to claim 2, wherein: The final adaptive parameters are calculated based on the initial benchmark, current compensation coefficient, comprehensive life compensation coefficient, and chemical system correction factor, including: Calculate the adaptive maximum charging current = nominal maximum charging current × current compensation coefficient × comprehensive life compensation coefficient; Calculate the adaptive maximum discharge current = nominal maximum discharge current × current compensation coefficient × comprehensive life compensation coefficient; Calculate the adaptive CV switching voltage = nominal constant current to constant voltage charging switching voltage × (1-α × (1-comprehensive life compensation coefficient)) × chemical system correction factor, where α is the empirical attenuation coefficient; Calculate the adaptive discharge cut-off voltage = nominal discharge cut-off voltage × (1 + β × (1-comprehensive life compensation coefficient)), where β is the empirical protection coefficient.

4. The lithium battery testing method according to claim 1, wherein: Based on real-time multimodal time series data, the health status of the tested battery is predicted through a multi-scale time series-frequency domain fusion network model. The prediction results are used to determine whether to trigger a closed-loop correction process, including: The sliding time window data of a preset length of multimodal time series data is used as the input of the multi-scale time series-frequency domain fusion network model, and the multi-scale time series-frequency domain fusion network model processes and outputs the predicted health status; Get the actual health status of the battery under test; Compare the deviation between the predicted health state and the actual health state. If the deviation exceeds the threshold, the closed-loop correction process is triggered. Otherwise, the closed-loop correction process is not triggered and the test is performed normally.

5. The lithium battery testing method according to claim 1, wherein: The structure of the multi-scale time-frequency fusion network model includes: One-dimensional convolution front-end: three layers of causal convolution for local feature extraction; Frequency domain branch: Perform short-time Fourier transform on the input sequence, obtain the amplitude spectrum, and then feed it into two layers of 2-D convolution and maximum pooling to output frequency domain features; Time domain branch: feeds the one-dimensional convolution front-end output into the bidirectional LSTM to output time domain features; Cross-attention fusion: Using time domain features as query and frequency domain features as key-value, a multi-head attention mechanism is implemented to output fused features; Task head: fully connected network mapped to healthy state.

6. The lithium battery testing method according to claim 1, wherein: The method further includes: During the execution of the closed-loop correction process, the newly generated multimodal time series data is stored in the experience playback buffer; The newly generated multimodal time series data is used to fine-tune the weights of the last K layers of the multi-scale time series-frequency domain fusion network model to obtain a new multi-scale time series-frequency domain fusion network model; The performance of the new multi-scale time-frequency fusion network model is evaluated on the retained validation set. If its predicted mean absolute percentage error is less than the threshold and no deviation alarm is triggered for several consecutive time windows, the closed-loop correction is considered to have converged. Otherwise, the new multi-scale time-frequency fusion network model is used to predict the health status of the battery under test, and then a new active test script is generated for iterative correction.

7. The lithium battery testing method according to claim 1, wherein: Based on the prediction results, the Bayesian optimizer generates active test scripts, including: The problem of generating active test scripts is formalized as a constrained optimization problem, including defining optimization variables, constructing an objective function, and defining constraints. The optimization variables are the script parameter vectors to be optimized, and the dimensions of the vectors correspond to the candidate stress modes, including excitation current rate, stress temperature, pulse width, and number of cycles. The objective function aims to maximize the rate of decrease of model prediction error or maximize the model information gain. The constraints define the safety boundary constraints for test safety, including that the maximum surface temperature of the battery must not exceed the absolute safety limit, the maximum voltage during the test must not exceed the upper limit of the electrochemical window, the minimum voltage must not be lower than the lower limit, and the total capacity decay must not exceed the maximum allowable capacity decay threshold. Call the Bayesian optimizer to iteratively solve the defined constrained optimization problem and output the global optimal solution stress parameter vector; The global optimal solution stress parameter vector and the target stress pattern identifier are compiled into an active test script.

8. A lithium battery testing system, characterized in that: include: Static data acquisition module, used to obtain the nominal parameters and life cycle historical data of the tested battery; Dynamic data acquisition module, used to collect multimodal battery status data of the tested battery in real time to form multimodal time series data; A charge and discharge parameter baseline generation module, which dynamically calculates and generates personalized charge and discharge parameter baselines based on nominal parameters, full lifecycle historical data, and multimodal battery status data using a built-in expert rule library. The battery testing module is used to send the personalized charge and discharge parameter baseline to the edge test equipment to test the battery under test; The deviation detection module is used to predict the health status of the battery under test based on real-time multimodal time series data through a multi-scale time series-frequency domain fusion network model during test execution. The module determines whether to trigger a closed-loop correction process based on the prediction results. If so, the closed-loop correction process is initiated. If not, the test is performed normally and the multimodal battery status data of the battery under test is continued to be collected in real time. The active test module is used to respond to the start of the closed-loop correction process, generate active test scripts based on the prediction results through the Bayesian optimizer, and send the active test scripts to the edge test device for execution.

9. A terminal, characterized in that: include: A memory for storing a lithium battery test program; A processor is configured to implement the steps of the lithium battery testing method according to any one of claims 1 to 7 when executing the lithium battery testing program.

10. A computer-readable storage medium, characterized in that The readable storage medium stores a lithium battery testing program, which, when executed by a processor, implements the steps of the lithium battery testing method according to any one of claims 1 to 7.

Images