Battery cognitive twin modeling method and system

Abstract

This invention belongs to the field of battery modeling technology and provides a method and system for building a cognitive twin model of a battery. It employs a hybrid modeling approach to construct a digital twin model of the battery, acquires time-series data of the battery under various operating conditions, extracts battery characteristic parameters, and performs initial calibration of the digital twin model. Simulation verification is then performed; if the simulation error exceeds a preset threshold, a large language model is used to analyze the simulation curves and measured curves, determine the deviation pattern, and generate parameter correction suggestions. The digital twin model is updated until the accuracy requirements are met. Responding to user predictive queries about the future state of the battery, the large language model is used to parse the query and transform it into a simulation task. Predictive simulation is then performed using the updated digital twin model. This invention transforms the traditional modeling process, which relies on human experience, into an intelligent closed loop driven by a large language model, achieving a paradigm shift from passive simulation to cognitive twins.

Description

Technical Field

[0001] This invention belongs to the field of battery modeling technology, specifically relating to a battery cognitive twin modeling method and system. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Batteries exhibit highly nonlinear aging characteristics during actual use, and their operating conditions are complex and variable. Traditional methods based on mechanistic models or purely data-driven approaches struggle to simultaneously meet the demands for high-precision simulation and real-time response. Digital twin technology, by constructing a virtual mirror image of the battery, provides a new approach for achieving state monitoring and performance prediction. However, existing battery modeling methods based on digital twin technology still face several challenges.

[0004] On the one hand, in the process of battery modeling, the calibration of model parameters relies heavily on human experience. When faced with batteries of different material systems and different aging stages, a lot of manual debugging work is required, which is inefficient and difficult to guarantee accuracy.

[0005] On the other hand, as the battery ages during long-term operation, the simulation error will gradually accumulate and increase. However, traditional methods lack automatic diagnosis and correction mechanisms, making it difficult to detect the root cause of the deviation and adjust the parameters in a timely manner.

[0006] Meanwhile, users have an increasing demand for predictive queries about the future state of batteries, such as remaining life assessment and available power estimation. However, existing systems struggle to directly translate users' natural language descriptions into executable simulation tasks, resulting in insufficient intelligence in human-computer interaction and limiting the cognitive capabilities of battery management systems. Summary of the Invention

[0007] To address the aforementioned problems, this invention proposes a battery cognitive twin modeling method and system.

[0008] According to some embodiments, the present invention adopts the following technical solution: A method for building a battery cognitive twin model includes the following steps: A hybrid modeling approach is used to construct a battery digital twin model, which consists of a multiphysics coupling mechanism model and a neural network compensation model. The multiphysics coupling mechanism model is used to realize the electro-thermal-aging coupling, and the neural network compensation model is used to learn the prediction residual of the multiphysics coupling mechanism model to obtain the prediction results of the battery target parameters. Acquire time-series data of the battery under various operating conditions, extract battery characteristic parameters from them, and perform initial calibration of the battery digital twin model based on the battery characteristic parameters; The battery digital twin model after initial calibration is simulated and verified. If the simulation error exceeds the preset threshold, the simulation curve and the measured curve are analyzed using a pre-trained large language model to determine the deviation mode and generate parameter correction suggestions for the multi-physics coupling mechanism model. The battery digital twin model is updated according to the parameter correction suggestions until the accuracy requirements are met; In response to a user's predictive query about the future state of the battery, the query is parsed using a pre-trained large language model and transformed into a simulation task. Based on the simulation task, predictive simulation is performed using the updated battery digital twin model.

[0009] As an alternative implementation method, the process of constructing a battery digital twin model using a hybrid modeling approach includes: the multi-physics coupling mechanism model of the battery digital twin model consists of an equivalent circuit model, a lumped parameter model, and a coupled aging model, and the three achieve electro-thermal-aging coupling through bidirectional data exchange; The neural network compensation model uses a long short-term memory network. It takes the battery's current, voltage, temperature, and historical state as input, learns the prediction residuals of the multiphysics coupling mechanism model, and obtains compensation terms for voltage, temperature, state of charge, and health state. These compensation terms are then added to the predicted values ​​of the multiphysics coupling mechanism model to obtain the final prediction result.

[0010] As an alternative implementation method, the process of acquiring historical time-series data and basic information of the battery under multiple operating conditions is achieved through a pre-trained large language model.

[0011] As an alternative implementation method, the process of acquiring time-series data of the battery under various operating conditions, extracting battery characteristic parameters from it, and performing initial calibration of the battery digital twin model based on the battery characteristic parameters includes: acquiring historical time-series data of the battery under various operating conditions and basic battery information, identifying typical operating condition segments, analyzing voltage response curves, and extracting electrical parameters including ohmic internal resistance and polarization parameters. Identify thermal parameters including specific heat capacity, thermal conductivity, and heat transfer coefficient based on temperature change curves; Based on the battery material system, aging parameters, including the typical parameter range required for the equivalent circuit model and the aging model index, are matched from a pre-built knowledge base. The extracted and matched parameters are integrated into a complete parameter set. Based on the complete parameter set, the multiphysics coupling mechanism model of the digital twin model is initially calibrated. The neural network compensation model is pre-trained using historical operating data to initialize network weights, enabling it to initially possess the ability to compensate for the residuals of the multi-physics coupling mechanism model.

[0012] As a further step, the process of identifying typical operating condition segments includes: pre-building a knowledge base, defining morphological feature thresholds for voltage curves or temperature curves for pulse operating conditions, step operating conditions and temperature rise operating conditions respectively, and pre-binding corresponding parameter calculation algorithms for each operating condition. After reading historical time-series data using a large language model, the data is segmented into candidate operating condition segments. Key information such as segment type, current change rate, voltage change amplitude, and temperature rise rate is then extracted from these segments to generate structured segment summaries. The large language model associates the fragment summary with the working condition name in the knowledge base, and compares the numerical features with the preset threshold through the rule engine. If the numerical features fall within the threshold range, the corresponding parameter calculation algorithm is called to complete the deterministic calculation and output the ohmic resistance, polarization parameter and thermal parameter.

[0013] As an optional implementation method, the process of simulating and verifying the battery digital twin model after initial calibration includes: periodically comparing the simulation curves and measured curves of voltage, temperature, state of charge and health output by the battery digital twin model using a large language model, and calculating the root mean square error and mean absolute error. When the error exceeds the preset threshold, the morphological features of the deviation curve are extracted, including several of the following: overall translation, local bulge, slope change, end deviation and abnormal temperature rise rate. Combined with parameter sensitivity analysis, the dominant factors causing the deviation are identified, and correction suggestions for the parameters of the multiphysics coupling mechanism model are generated. For neural network compensation models, historical data is used for incremental training or online weight updates.

[0014] As an alternative implementation, in response to a user's predictive query about the future state of the battery, the process of parsing the query using a pre-trained large language model and transforming it into a simulation task includes: performing deep analysis of the user's input text or voice query using the pre-trained large language model, identifying the user's intent through natural language processing technology, transforming the results of intent recognition and entity extraction into a structured simulation task script, which includes the task type, input parameters, output specifications, and execution constraints, and sending it to the digital twin model. The digital twin model loads the latest model parameters and state and executes the corresponding predictive simulation task.

[0015] As an alternative implementation, the battery digital twin model is deployed in the cloud, and a lightweight multiphysics coupling mechanism model and a neural network compensation model, which have undergone knowledge distillation, are deployed on the edge as a lightweight copy of the battery digital twin model deployed in the cloud, and are updated synchronously with the battery digital twin model in the cloud.

[0016] As a further defined implementation, the cloud and the edge communicate through a collaborative interaction interface.

[0017] As a further defined implementation, the lightweight battery digital twin model deployed on the edge, after knowledge distillation, receives update parameters from the large language model, achieves synchronous updates with the cloud-side model, and performs local inference based on the acquired real-time data to obtain battery state prediction results.

[0018] A battery cognitive twin modeling system, comprising: The battery digital twin model construction module is configured to construct a battery digital twin model using a hybrid modeling method. The battery digital twin model consists of a multi-physics coupling mechanism model and a neural network compensation model. The multi-physics coupling mechanism model is used to realize the electro-thermal-aging coupling, and the neural network compensation model is used to learn the prediction residual of the multi-physics coupling mechanism model to obtain the prediction results of the battery target parameters. The initial calibration module is configured to acquire time-series data of the battery under various operating conditions, extract battery characteristic parameters from it, and perform initial calibration of the battery digital twin model based on the battery characteristic parameters. The simulation correction module is configured to perform simulation verification on the battery digital twin model after initial calibration. If the simulation error exceeds a preset threshold, the simulation curve and the measured curve are analyzed using a pre-trained large language model to determine the deviation mode and generate parameter correction suggestions for the multi-physics coupling mechanism model. The model update module is configured to update the battery digital twin model according to the parameter correction suggestions until the accuracy requirements are met. The intent recognition module is configured to respond to a user's predictive query about the future state of the battery, parse the query using a pre-trained large language model, and transform it into a simulation task. The predictive simulation module is configured to perform predictive simulations based on the simulation task using the updated battery digital twin model.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention transforms the traditional modeling process, which relies on human experience, into an intelligent closed loop driven by a large language model, achieving a paradigm shift from passive simulation to cognitive twins.

[0020] This invention enables automatic extraction and initial calibration of parameters for battery digital twin models, reducing manual intervention and improving modeling efficiency and accuracy. It analyzes deviation patterns and generates correction suggestions through a large language model, achieving adaptive calibration of the model and maintaining long-term operational accuracy. It supports users to perform predictive queries in natural language, automatically parsing and driving simulation, improving the level of human-computer interaction intelligence. It adopts an edge-cloud collaborative architecture and lightweight model technology, balancing the high accuracy requirements of the cloud side and the real-time requirements of the edge side.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0022] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0023] Figure 1 This is a flowchart illustrating a battery cognitive twin modeling method combining a large language model and a digital twin model, provided by an embodiment of the present invention. It shows the complete steps from model framework construction, parameter calibration, adaptive calibration to intelligent interaction. Figure 2 This is an architecture diagram of a battery cognitive twin model that combines a large language model and a digital twin model, provided by an embodiment of the present invention. It shows the composition and interaction relationship of the cloud-side digital twin model, the edge-side lightweight model, and the intelligent hub of the large language model based on edge-cloud collaboration. Figure 3 This is a flowchart of parameter extraction and initial calibration of a battery cognitive twin model that combines a large language model and a digital twin model, provided by an embodiment of the present invention. It specifically shows the detailed steps of the intelligent central hub of the large language model automatically extracting electrochemical, thermal and aging parameters from time-series data and performing initial calibration of the digital twin model. Figure 4 These are comparison charts of model initialization and state verification provided in this embodiment of the invention, wherein: a. Multiphysics model parameter diagram; b. LSTM network weight initialization heatmap; c. Battery terminal voltage comparison chart; e. Battery surface temperature comparison chart; f. Battery state of charge (SOC) comparison chart; g. Battery state of health (SOH) comparison chart; h. Model error comparison chart; Figure 5 This invention provides an adaptive calibration flowchart for a battery cognitive twin model that combines a large language model and a digital twin model, detailing the closed-loop process of analyzing bias patterns, generating parameter correction suggestions, and updating the model using the large language model. Figure 6These are comparison charts before and after adaptive calibration provided in this embodiment of the invention, including: a. Comparison table of multiphysics model parameters before and after correction; b. Chart of average weight change amplitude of LSTM input features; c. Comparison chart of LSTM network weight distribution before and after correction; d. Comparison chart of battery terminal voltage before and after correction with the true value; e. Comparison chart of battery surface temperature before and after correction with the true value; f. Comparison chart of battery state of charge (SOC) before and after correction with the true value; g. Comparison chart of battery state of health (SOH) before and after correction with the true value; h. Comparison table of model error before and after correction; Figure 7 This is a comparison chart of battery state estimation under different aging levels provided in the embodiments of the present invention, showing the comparison of the prediction accuracy of the model at different aging stages of battery capacity decay; Figure 8 This is a comparison chart of battery state estimation under different operating ambient temperatures provided in the embodiments of the present invention, showing the comparison of the prediction accuracy of the model under different ambient temperature conditions; Figure 9 This is a comparison chart of state estimation of different individual batteries provided in the embodiments of the present invention, showing the comparison of the prediction effect of the model on different individual batteries, and verifying the model's ability to adapt to individual differences; Figure 10 This is a comparison chart of battery state estimation under different operating conditions provided in the embodiments of the present invention, showing the comparison of the prediction accuracy of the model under different operating conditions; Figure 11 This is a flowchart of intelligent task parsing and predictive simulation execution provided in an embodiment of the present invention, which shows the complete closed-loop process from user query input to simulation result output; Figure 12 This is a schematic diagram of the battery intelligent operation and maintenance assistant interactive interface provided in the embodiment of the present invention, which shows the multimodal interaction method of users querying battery status through natural language dialogue, and the system returning text analysis results and capacity decay curves; Figure 13 This is a schematic diagram of the real-time monitoring interface of the battery cognitive twin system provided in this embodiment of the invention, which shows the multimodal output effect of the system in presenting battery voltage, temperature, current and other state parameters through visualization curves and real-time data monitoring panel during operation; Figure 14 This is a schematic diagram of the cloud-side model performance monitoring and edge-to-edge collaborative verification interface provided in this embodiment of the invention. It shows the output comparison curves of the cloud digital twin model and the edge quantization model, the model accuracy evaluation index, the key parameter table, and the network structure adjustment interface. Figure 15 This is a functional module diagram of a battery cognitive twin modeling system that combines a large language model and a digital twin model, provided by an embodiment of the present invention. It shows the eight core modules included in the system and their functional divisions. Figure 16 This is a system architecture diagram of a battery cognitive twin model that combines a large language model and a digital twin model, provided by an embodiment of the present invention. It shows the physical deployment architecture based on edge-cloud collaboration and the data interaction relationship between each layer. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0027] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0028] Example 1 A method for building a battery cognitive twin model by combining a large language model and a digital twin model, such as... Figure 1 As shown, the method includes: S101 completes the construction of the battery digital twin model. The model adopts a hybrid modeling method, integrating a multi-physics coupling mechanism model and a neural network compensation model.

[0029] S102 automatically extracts battery characteristic parameters from the collected time-series data and performs initial calibration of the digital twin model based on the extracted parameters.

[0030] In model verification, when the simulation error exceeds a preset threshold, the S103 analyzes the voltage curve deviation pattern using a large language model and generates parameter correction suggestions to achieve adaptive calibration of the model.

[0031] The S104 large language model intelligent hub analyzes users' predictive queries about the future state of the battery and transforms them into simulation tasks.

[0032] S105 communicates bidirectionally with the updated digital twin model through a collaborative interaction interface to perform predictive simulations.

[0033] S106 verifies the model's prediction accuracy under different working conditions using public datasets, and presents the analysis results in a multimodal interactive manner through a visual interface and natural language dialogue.

[0034] Further, in step S101, the battery digital twin model architecture is as follows: Figure 2 As shown, an edge-cloud collaborative architecture is adopted. On the cloud side, a cloud battery digital twin model and a large language model intelligent hub are deployed. The cloud battery digital twin model consists of a high-precision multi-physics coupling mechanism model and a neural network compensation model. The multi-physics coupling mechanism model comprises an equivalent circuit model, a lumped parameter model, and a coupled aging model, which achieve electro-thermal-aging coupling through bidirectional data exchange. The neural network compensation model uses a long short-time memory network, taking the battery's current, voltage, temperature, and historical state as input, learning the mechanism model's prediction residuals, and outputting compensation terms for voltage, temperature, state of charge, and health state. These are then added to the mechanism model's predicted values ​​item by item to obtain the final prediction result. On the edge side, a lightweight physical field coupling model and a lightweight neural network, after knowledge distillation, are deployed as a lightweight copy of the cloud battery digital twin model. Through a collaborative interaction interface, they receive updated parameters from the large language model intelligent hub in real time, achieving synchronous updates with the cloud-side model. Based on real-time data collected on the edge side, they perform local rapid inference and output battery state prediction results.

[0035] As a typical embodiment, step S101 specifically includes: On the cloud side, firstly, a high-fidelity digital twin model integrating the electrochemical-thermal-aging multi-physics coupling mechanism and deep neural network compensation is constructed. Then, using this model as a simulation environment, the optimal charging strategy is generated through reinforcement learning training, and at the same time, a lightweight model is obtained on the edge side through knowledge distillation technology. The strategy and lightweight model generated on the cloud side are sent to the edge side, where the lightweight model performs rapid simulation and safety verification of the charging strategy. The verified strategy is finally sent to the edge side for execution by the battery management system. At the same time, the real-time running data collected on the edge side is fed back to the cloud, driving the parameter update and structural self-correction of the high-fidelity model, thereby forming a continuously optimized integrated closed loop of "perception-decision-verification-execution".

[0036] The battery digital twin modeling method based on edge-cloud collaboration specifically includes: In the cloud, a basic white-box model is established based on the electrochemical-thermal-aging multiphysics coupling mechanism, considering lithium-ion concentration distribution, electrode reaction kinetics, heat conduction, and capacity decay mechanisms; simultaneously, a deep neural network is introduced to construct a black-box compensation model, which adopts a long short-term memory network structure to learn the unknown nonlinear dynamics in the residuals of the basic model; by weighted fusion of the mechanism output of the white-box model and the compensation output of the black-box model, a high-fidelity gray-box digital twin model is constructed, the mathematical model of which can be expressed as: (1) in, For the multivariate output of the digital twin model, This represents the output of the multiphysics coupling model. This represents the compensation part of the unknown nonlinear neural network, where Input to the model, This refers to the internal state of the system. For physical parameter vectors, These are the weight parameters of the neural network.

[0037] The multiphysics coupling model is established based on the battery's electro-thermal-aging state coupling mechanism. Specifically, it includes a coupled second-order equivalent circuit model, a coupled thermal model, a coupled aging model, and a coupled ampere-hour integral model. The electrical model uses a second-order equivalent circuit, and its voltage output is affected by temperature. The calculation of the thermal model depends on the heat generation power of the electrical model. The decay rate of the aging model is driven by the current and voltage of the electrical model and the temperature of the thermal model. Through bidirectional coupling of state variables and parameters, each model achieves high-fidelity synchronous simulation of the battery's external characteristics and internal state.

[0038] A dual-time-scale extended Kalman filter algorithm is employed for online model calibration. This algorithm comprises two cooperative estimation loops: a fast-changing loop responsible for rapidly tracking dynamic states such as SOC and voltage; and a slow-changing loop responsible for identifying and correcting physical parameters that change slowly with aging, such as ohmic resistance and thermal resistance. This mechanism effectively solves the problem of time-varying drift of model parameters, ensuring the prediction accuracy of the digital twin throughout its entire lifecycle.

[0039] The neural network compensation The network architecture employs Long Short Period Memory (LSTM), and the network input includes both the model input and the system's internal state. (2) The network input at the previous moment is represented as: The external input of the model at the previous time step is represented as: The model's internal input at the previous time step is represented as: This 8-dimensional input vector comprehensively covers the electrical, thermal, and state characteristics of the battery, enabling LSTM to effectively capture the battery's dynamic memory effect and historical dependence through its internal control mechanism, thereby accurately outputting residual compensation terms for voltage, temperature, and SOC.

[0040] Specifically, these compensation values ​​are added to the prediction values ​​of the multiphysics coupling model to form the final output of the digital twin model, as shown in Equation 1.

[0041] For network training, a stochastic gradient descent algorithm is used to iteratively optimize the weight parameters of the LSTM neural network. The training process is based on battery operation data uploaded from the device, and the data is divided into multiple mini-batches for efficient training. A weighted loss function is defined to simultaneously consider the prediction errors of voltage, temperature, and SOC. By minimizing the above loss function, the optimized network structure and parameters are obtained.

[0042] The construction of a lightweight digital twin model on the edge side inherits the electrochemical-thermal-aging multiphysics physical model structure and parameters from the high-fidelity model on the cloud side, ensuring that the edge-side model follows the physical laws of battery operation. Simultaneously, the complete LSTM network compensation module on the cloud side is structurally pruned, removing redundant neuron connections and hidden layer units to obtain a lightweight LSTM student model while maintaining time-series prediction capabilities. Specifically, this includes: By using width pruning and depth pruning, a method can be obtained with a depth range from... Layer reduction to The number of layers and nodes per layer from Reduce to The lightweight student model structure is significantly less complex than the cloud-based teacher model. This structure fundamentally reduces computational complexity and memory consumption. Knowledge distillation training is then performed based on this structure.

[0043] A teacher-student distillation framework is constructed, using a cloud-side complete LSTM network as the teacher model and an edge-side lightweight LSTM network as the student model. A temperature regulation mechanism is employed to soften the output distribution of the teacher model. Let the outputs of the teacher model and the student model be respectively... and Temperature parameters are The distribution of soft targets is as follows: (3) An alternating training strategy is employed, fixing the teacher model parameters and updating only the student model parameters. Stochastic gradient descent is used to optimize the student model. After training until the student model's performance on the validation set reaches over 95% of the teacher model's, the distilled lightweight LSTM network is integrated with the distributed physical model parameters at the edge, forming a lightweight digital twin model on the edge side. Figure 3 As shown in the figure, this comparison illustrates the relationship between the predicted outputs and actual measured values ​​of the two models for key state parameters of battery voltage, temperature, and SOC under the same input conditions. The comparison curves clearly demonstrate that the cloud-side high-fidelity model has higher prediction accuracy, while the edge-side lightweight model achieves a significant improvement in computational efficiency while maintaining relative accuracy.

[0044] Furthermore, in step S102, the specific process for parameter extraction and initial calibration is as follows: Figure 3As shown, the process includes: a large language model intelligent central hub reads historical time-series data and basic battery information under multiple operating conditions, identifies typical operating condition segments, and extracts ohmic internal resistance and polarization parameters by analyzing voltage response curves; identifies thermal parameters such as specific heat capacity, thermal conductivity, and heat transfer coefficient based on temperature change curves; combines the battery material system to match the typical parameter range required for the equivalent circuit model and aging parameters such as aging model index from a pre-built knowledge base; integrates the extracted and matched parameters into a complete parameter set, and sends it to the digital twin model through a collaborative interaction interface to complete the initial calibration of the equivalent circuit model and the lumped parameter thermal model. Simultaneously, historical operating data is used to pre-train the neural network compensation model, initializing network weights to enable it to initially compensate for the residuals of the mechanistic model.

[0045] Furthermore, in step S102, the identification process is implemented through a rule-guided automatic algorithm scheduling method, specifically including: First, a knowledge base is pre-built to define the morphological characteristic thresholds of voltage curves or temperature curves for pulse conditions, step conditions, and temperature rise conditions, respectively. Then, corresponding parameter calculation algorithms are pre-bound for each condition. For example, the ohmic internal resistance is calculated by the voltage difference before and after the current change, the polarization resistance and polarization capacitance are extracted by exponential fitting of the voltage recovery curve, or the specific heat capacity and heat transfer coefficient are inverted based on the lumped parameter thermal model.

[0046] After reading historical time-series data, the big language model intelligent hub first divides it into candidate operating condition segments, and then extracts key information such as segment type, current change rate, voltage change amplitude, and temperature rise rate from them to generate structured segment summaries.

[0047] Subsequently, the big language model intelligent hub uses semantic understanding capabilities to associate these summaries with the operating condition names in the knowledge base. At the same time, it compares the numerical features with preset thresholds through the rule engine. Once the numerical features fall within the threshold range, it automatically calls the parameter calculation algorithm bound to the operating condition, and the algorithm completes the deterministic calculation, outputting the ohmic resistance, polarization parameters, and thermal parameters.

[0048] The large language model intelligent hub combines the battery's material system and cycle number to match the typical parameter range required for the equivalent circuit model and aging parameters such as the aging model index from the knowledge base.

[0049] Furthermore, in this embodiment, the pre-built knowledge base includes three parts: the first part is a typical operating condition feature template library, which selects typical operating condition segments such as pulse, step, and temperature rise from public datasets such as NASA and CALCE, extracts features such as voltage change rate, current change amplitude, and temperature rise rate, and determines the threshold range, and organizes the operating condition name, feature description, and threshold conditions into JSON format and stores them in the database.

[0050] The second part is a parameter acquisition algorithm library, which pre-writes calculation methods for different operating conditions. For example, under pulsed operating conditions, the ohmic internal resistance is calculated based on the voltage difference; under static operating conditions, the polarization parameters are calculated by performing exponential fitting on the voltage recovery curve; and under temperature rise operating conditions, the thermal parameters are calculated by back-calculating the thermal model. These methods are encapsulated into functions and associated with the corresponding operating conditions.

[0051] The third part is the parameter range and aging knowledge base. It organizes the industry-recognized value ranges of parameters such as ohmic internal resistance, polarization resistance, and polarization capacitance of equivalent circuit models under different material systems into tables. At the same time, based on publicly available aging datasets such as NASA and CALCE, it fits the corresponding aging model indices under different conditions and stores them in the database.

[0052] To verify the initial calibration results described above, this embodiment used the NASA battery dataset for testing. Figure 4 As shown, after automatic parameter extraction and calibration of the large language model, the key parameters of the multiphysics model ( Figure 4 a) and the weight distribution of the LSTM network ( Figure 4 b, Figure 4 c) were all successfully initialized. Figure 4 d to Figure 4 The curves of the initial model simulation and the actual values ​​were compared. The results show that the simulated curves of the initial model's terminal voltage, surface temperature, SOC, and SOH are basically consistent with the actual curves, indicating that the parameters automatically extracted through the large language model have high accuracy. Further analysis... Figure 4 Comparing the errors of h, the root mean square error (RMSE) of the initial model has been controlled within a small range, proving the effectiveness of the automatic calibration method proposed in this invention and greatly reducing the need for manual intervention.

[0053] Further, in step S103, the specific process of the model adaptive calibration is as follows: Figure 5 As shown, the process includes: during model operation, the digital twin model outputs simulated voltage, temperature, state of charge, and health status in real time; the intelligent central processing unit of the large language model periodically compares the simulated curves with the measured curves, and calculates the root mean square error and mean absolute error; when the error exceeds a preset threshold, the morphological features of the deviation curve are extracted, including overall translation, local bulges, slope changes, end deviations, abnormal temperature rise rates, etc., and combined with parameter sensitivity analysis, the dominant factors causing the deviation are identified, and correction suggestions for the parameters of the multi-physics coupling mechanism model are generated. At the same time, for the neural network compensation model, incremental training or online weight updates are performed using historical data; the parameter updates of the digital twin model are triggered through a collaborative interaction interface, and the error is re-verified after the update, iterating until the accuracy requirements are met.

[0054] To verify the effectiveness of adaptive calibration, this embodiment calibrated the model exhibiting significant errors. For example... Figure 6 As shown, the difference before and after calibration is obvious: Figure 6 A demonstrates that key mechanism parameters such as ohmic internal resistance and polarization capacitance have been accurately corrected. Figure 6 b and Figure 6 The result shows that the LSTM model dynamically adjusts its attention to different input features and its internal weight distribution. After calibration, from Figure 6 d to Figure 6 As can be seen from g, the simulation curve of the corrected battery state has a significantly improved fit with the real value. Figure 6 The error comparison table for h quantifies this improvement. After calibration, the voltage RMSE decreased from the initial 0.092V to 0.0099V, the temperature RMSE decreased from 0.83℃ to 0.102℃, and the SOC estimation accuracy also improved significantly. This result demonstrates that the large language model can effectively diagnose the root causes of biases and guide the model to complete adaptive calibration, maintaining the model's accuracy in long-term operation.

[0055] To verify the generalization ability of the model established in this invention, tests were conducted under different aging levels of batteries, different ambient temperatures, different individual batteries, and different operating conditions. The results are as follows: Figures 7 to 10 As shown. Figure 7 This demonstrates that the model can still maintain high accuracy in state estimation during the middle and late stages of battery capacity decay; Figure 8 The results show that, within a wide temperature range of 4℃, 24℃, and 43℃, the model's simulations of voltage and temperature are highly consistent with the actual values. Figure 9 This indicates that the model can accurately estimate the state of different individual batteries in the same batch without additional debugging, demonstrating good adaptability to individual differences. Figure 10 This verifies the prediction accuracy of the model under various typical driving conditions, including US06, HWFET, UDDS, and Cycle_1. These examples collectively demonstrate that the battery cognitive twin model constructed using the proposed method exhibits good generalization ability and robustness across different application scenarios.

[0056] Further, steps S104 and S106, as follows: Figure 11 As shown, the intelligent central processing unit of the large language model first performs deep analysis on the text or voice query input by the user, identifies the user's intent through natural language processing technology, and transforms the results of intent recognition and entity extraction into a structured simulation task script. This script includes the task type, input parameters, output specifications and execution constraints, and is sent to the digital twin model through a collaborative interaction interface. The digital twin model loads the latest model parameters and state, and executes the corresponding predictive simulation tasks, such as extrapolating the capacity decay curve based on the current state through the aging model.

[0057] Further, in step S106, the large language model intelligent hub performs multimodal conversion on the simulation results returned by the digital twin model. Ultimately, as... Figure 12 The image shows the interactive interface of the battery intelligent operation and maintenance assistant. The system not only responds through natural language dialogue, but also simultaneously generates a capacity degradation curve, intuitively displaying the future capacity change trend. Figure 13 The real-time monitoring interface presents battery voltage, temperature, current, and other status parameters in the form of visual curves and a real-time data monitoring panel, supporting multimodal interaction. Meanwhile, as... Figure 14 The cloud-side model performance monitoring and edge-to-edge collaborative verification interface shown displays the output comparison curves of the cloud digital twin model and the edge quantization model, model accuracy evaluation indicators, key parameter tables, and network structure adjustment interface. It provides system operation and maintenance personnel with a visual tool for evaluating the effect of lightweight model, verifying consistency between the cloud and the edge, and optimizing the network structure, thereby constructing a multi-role, multi-level, multimodal interaction system for end users, operation and maintenance personnel, and developers.

[0058] Example 2 A functional module diagram of a battery cognitive twin modeling system that combines a large language model and a digital twin model, as shown below. Figure 15 As shown, it includes: The digital twin model building module is configured to build a battery digital twin model, which includes a multiphysics coupling model and a compensation model. The multiphysics coupling model is used to make basic predictions of voltage response, temperature distribution and state of charge changes based on the battery's electro-thermal-aging coupling mechanism. The compensation model is used to learn and characterize the unknown nonlinear characteristics and coupling effects in the battery's dynamic process, and to compensate for the basic predictions to obtain the final prediction results. The model lightweighting module extracts key features from the battery digital twin model through knowledge distillation technology and distills them to obtain a lightweight side battery digital twin model. The large language model intelligent hub module is configured to drive the entire process of battery cognitive twin modeling, including: automatically extracting battery characteristic parameters from the collected time-series data and sending them to the basic model building module for initial calibration; analyzing voltage curve deviation patterns and generating parameter correction suggestions when the model verification error exceeds the threshold; parsing user predictive queries about the future state of the battery and converting them into simulation tasks; and performing multimodal conversion on the simulation results and feeding them back to the user. The parameter extraction and calibration module is configured as the intelligent hub of the collaborative large language model. It extracts electrochemical parameters, thermal parameters and aging parameters from historical time series data, matches the initial parameters from the pre-built knowledge base, and integrates them to complete the initial calibration of the multi-physics coupling model. The adaptive calibration module is configured to compare the simulation output with the measured data in real time. When the error exceeds the preset threshold, deviation analysis is triggered. The intelligent central hub of the large language model generates correction suggestions for the parameters of the multi-physics coupled model, and at the same time triggers the online weight update of the neural network compensation model. The human-computer interaction module is configured to provide a multimodal interactive interface between the user and the system, including a visual interaction unit and a natural language dialogue unit. The visual interaction unit is used to generate charts such as battery state curves, temperature distribution maps, capacity decay trend maps, and lifetime prediction probability distribution maps to graphically display simulation results. The natural language dialogue unit supports users to query via text or voice input and converts the simulation results returned by the system into natural language descriptions to respond to users in a dialogue format, thereby achieving dynamic interaction.

[0059] The database management module is configured to store and manage battery operating data and a battery-related knowledge base. The battery operating data includes time-series data such as current, voltage, and temperature uploaded from the terminal side. The battery-related knowledge base includes templates for electrochemical parameters, thermal parameters, aging parameters, deviation patterns and parameter correction mapping relationships, and templates for common user questions for different material systems, providing knowledge support for the intelligent hub of the large language model.

[0060] A system architecture diagram for a battery cognitive twin modeling system that combines a large language model with a digital twin model, as shown below. Figure 16 As shown, it includes: The edge-side physical sensing layer is the foundation for the system's interaction with the physical battery. This layer includes the physical energy storage battery and its associated charge / discharge management and temperature control equipment. The bidirectional power supply is responsible for the battery's charge / discharge management, while the temperature control equipment is used to precisely control the battery's operating temperature. Through the data acquisition and transmission module, the edge side collects real-time time-series data such as the battery's current, voltage, and surface temperature. This data is then processed by the industrial control computer and uploaded to the edge side, providing realistic physical data support for the digital twin model.

[0061] The edge-side perceptual neural layer undertakes real-time inference and local loop closure tasks. This layer deploys a lightweight edge digital twin model that has undergone knowledge distillation, enabling rapid and lightweight simulation based on real-time data uploaded from the edge, achieving real-time estimation of battery state. Simultaneously, the edge layer possesses parameter verification and adaptive correction capabilities, receiving updated parameters from the cloud and performing local verification, maintaining model accuracy while ensuring real-time performance.

[0062] The cloud-based cognitive brain layer is the intelligent core of the entire system. Based on a battery knowledge base, this layer integrates functional modules such as natural language parsing, model accuracy evaluation, knowledge distillation, adaptive correction, indicator prediction, and decision suggestion generation. Users input natural language queries through a human-computer interaction window. The cloud-based layer interprets the intent and drives the digital twin model to perform predictive simulations, feeding back the results to the user through report generation and decision suggestion modules. The cloud-based layer is also responsible for online accuracy evaluation of the model, optimizing model parameters through adaptive correction mechanisms, and using knowledge distillation technology to generate lightweight models for deployment on the edge.

[0063] The three layers form a complete closed-loop optimization system: the edge provides real data, the edge enables real-time reasoning, the cloud performs intelligent decision-making and model optimization, and the optimized parameters and lightweight model are then sent to the edge for updates, enabling the system to have continuous learning and adaptive capabilities, and realizing a paradigm shift from passive simulation to cognitive twins.

[0064] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).

[0065] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for modeling a battery cognitive twin, characterized in that, Includes the following steps: A hybrid modeling approach is used to construct a battery digital twin model, which consists of a multiphysics coupling mechanism model and a neural network compensation model. The multiphysics coupling mechanism model is used to realize the electro-thermal-aging coupling, and the neural network compensation model is used to learn the prediction residual of the multiphysics coupling mechanism model to obtain the prediction results of the battery target parameters. Acquire time-series data of the battery under various operating conditions, extract battery characteristic parameters from them, and perform initial calibration of the battery digital twin model based on the battery characteristic parameters; The battery digital twin model after initial calibration is simulated and verified. If the simulation error exceeds the preset threshold, the simulation curve and the measured curve are analyzed using a pre-trained large language model to determine the deviation mode and generate parameter correction suggestions for the multi-physics coupling mechanism model. The battery digital twin model is updated according to the parameter correction suggestions until the accuracy requirements are met; In response to a user's predictive query about the future state of the battery, the query is parsed using a pre-trained large language model and transformed into a simulation task. Based on the simulation task, predictive simulation is performed using the updated battery digital twin model.

2. The battery cognitive twin modeling method as described in claim 1, characterized in that, The process of constructing a battery digital twin model using a hybrid modeling method includes: the multi-physics coupling mechanism model of the battery digital twin model consists of an equivalent circuit model, a lumped parameter model, and a coupled aging model, which achieve electro-thermal-aging coupling through bidirectional data exchange; The neural network compensation model uses a long short-term memory network. It takes the battery's current, voltage, temperature, and historical state as input, learns the prediction residuals of the multiphysics coupling mechanism model, and obtains compensation terms for voltage, temperature, state of charge, and health state. These compensation terms are then added to the predicted values ​​of the multiphysics coupling mechanism model to obtain the final prediction result.

3. The battery cognitive twin modeling method as described in claim 1, characterized in that, The process of acquiring historical time-series data and basic information about the battery under multiple operating conditions is achieved through a pre-trained large language model.

4. The battery cognitive twin modeling method as described in claim 1, characterized in that, The process of acquiring time-series data of the battery under various operating conditions, extracting battery characteristic parameters from it, and performing initial calibration of the battery digital twin model based on the battery characteristic parameters includes: acquiring historical time-series data of the battery under various operating conditions and basic battery information, identifying typical operating condition segments, analyzing voltage response curves, and extracting electrical parameters including ohmic internal resistance and polarization parameters. Identify thermal parameters including specific heat capacity, thermal conductivity, and heat transfer coefficient based on temperature change curves; Based on the battery material system, aging parameters, including the typical parameter range required for the equivalent circuit model and the aging model index, are matched from a pre-built knowledge base. The extracted and matched parameters are integrated into a complete parameter set. Based on the complete parameter set, the multiphysics coupling mechanism model of the digital twin model is initially calibrated. The neural network compensation model is pre-trained using historical operating data to initialize network weights, enabling it to initially possess the ability to compensate for the residuals of the multi-physics coupling mechanism model.

5. The battery cognitive twin modeling method as described in claim 4, characterized in that, The process of identifying typical operating condition segments includes: pre-building a knowledge base, defining morphological feature thresholds for voltage curves or temperature curves for pulse operating conditions, step operating conditions, and temperature rise operating conditions respectively, and pre-binding corresponding parameter calculation algorithms for each operating condition. After reading historical time-series data using a large language model, the data is segmented into candidate operating condition segments. Key information such as segment type, current change rate, voltage change amplitude, and temperature rise rate is then extracted from these segments to generate structured segment summaries. The large language model associates the fragment summary with the working condition name in the knowledge base, and compares the numerical features with the preset threshold through the rule engine. If the numerical features fall within the threshold range, the corresponding parameter calculation algorithm is called to complete the deterministic calculation and output the ohmic resistance, polarization parameter and thermal parameter.

6. The battery cognitive twin modeling method as described in claim 1, characterized in that, The process of simulating and verifying the battery digital twin model after initial calibration includes: periodically comparing the simulation curves and measured curves of voltage, temperature, state of charge and health output by the battery digital twin model using a large language model, and calculating the root mean square error and mean absolute error. When the error exceeds the preset threshold, the morphological features of the deviation curve are extracted, including several of the following: overall translation, local bulge, slope change, end deviation and abnormal temperature rise rate. Combined with parameter sensitivity analysis, the dominant factors causing the deviation are identified, and correction suggestions for the parameters of the multiphysics coupling mechanism model are generated. For neural network compensation models, historical data is used for incremental training or online weight updates.

7. The battery cognitive twin modeling method as described in claim 1, characterized in that, In response to a user's predictive query about the future state of the battery, the process of parsing the query using a pre-trained large language model and transforming it into a simulation task includes: performing deep analysis of the user's input text or voice query using the pre-trained large language model; identifying the user's intent using natural language processing technology; transforming the intent recognition and entity extraction results into a structured simulation task script, which includes the task type, input parameters, output specifications, and execution constraints, and sending it to the digital twin model; and loading the latest model parameters and state into the digital twin model to execute the corresponding predictive simulation task.

8. The battery cognitive twin modeling method as described in claim 1, characterized in that, The battery digital twin model is deployed in the cloud, and a lightweight multiphysics coupling mechanism model and a neural network compensation model, which have undergone knowledge distillation, are deployed on the edge as a lightweight copy of the battery digital twin model deployed in the cloud. The battery digital twin model is updated synchronously with the battery digital twin model in the cloud. The cloud and edge communicate through a collaborative interaction interface.

9. The battery cognitive twin modeling method as described in claim 7, characterized in that, The lightweight battery digital twin model, which is a knowledge distillation model deployed on the edge, receives updated parameters from the large language model, achieves synchronous updates with the cloud-side model, and performs local inference based on the acquired real-time data to obtain battery state prediction results.

10. A battery cognitive twin modeling system, characterized in that, include: The battery digital twin model construction module is configured to construct a battery digital twin model using a hybrid modeling method. The battery digital twin model consists of a multi-physics coupling mechanism model and a neural network compensation model. The multi-physics coupling mechanism model is used to realize the electro-thermal-aging coupling, and the neural network compensation model is used to learn the prediction residual of the multi-physics coupling mechanism model to obtain the prediction results of the battery target parameters. The initial calibration module is configured to acquire time-series data of the battery under various operating conditions, extract battery characteristic parameters from it, and perform initial calibration of the battery digital twin model based on the battery characteristic parameters. The simulation correction module is configured to perform simulation verification on the battery digital twin model after initial calibration. If the simulation error exceeds a preset threshold, the simulation curve and the measured curve are analyzed using a pre-trained large language model to determine the deviation mode and generate parameter correction suggestions for the multi-physics coupling mechanism model. The model update module is configured to update the battery digital twin model according to the parameter correction suggestions until the accuracy requirements are met. The intent recognition module is configured to respond to a user's predictive query about the future state of the battery, parse the query using a pre-trained large language model, and transform it into a simulation task. The predictive simulation module is configured to perform predictive simulations based on the simulation task using the updated battery digital twin model.

Images