Training data set construction method and device, equipment, storage medium and program product

Abstract

The application discloses a training data set construction method and device, equipment, a storage medium and a program product. The method comprises the following steps: determining a target battery parameter based on battery-related information, determining a plurality of historical battery models matched with the target battery model, and constructing a migration rule set; performing physical quantity migration and message timestamp migration on historical operation data according to the migration rule set to obtain migrated operation data. Residual error analysis is performed on the migrated operation data and target operation data, and residual error correction is performed on the migrated operation data to obtain corrected operation data. A training data set is constructed according to the corrected operation data and the target operation data. The model trained based on the data set constructed according to the technical scheme of the application can accurately predict the state of the battery after actual deployment, effectively reduces the problem of missed reports or false reports, significantly improves the model training efficiency and training effect, and provides a practical and reliable battery state prediction model for new battery models.

Description

Technical Field

[0001] This application belongs to the field of battery technology, and in particular relates to methods, apparatus, devices, storage media and program products for constructing training datasets. Background Technology

[0002] The Battery Management System (BMS) is the core control unit of a battery. In today's battery industry, battery state prediction models deployed in the BMS can be used to achieve important functions such as accurate thermal runaway early warning, State of Health (SOH) estimation, and Remaining Useful Life (RUL) prediction.

[0003] However, in the process of developing intelligent applications for the battery management system of new battery models, the lack of sufficient real-world operational data makes it difficult to train an effective battery state prediction model for deployment. Furthermore, a large amount of data from older battery models differs significantly from that of newer models in many aspects, rendering the older model data unusable. Existing training data construction methods have serious flaws; the resulting datasets cannot fully conform to the actual specifications of the battery management system corresponding to the new battery models. Models trained on this data frequently experience false alarms and missed alarms after deployment, severely impacting normal battery operation. On the other hand, training the model based on only a small amount of operational data from the new battery models leads to overfitting, severely affecting the model's practical application performance.

[0004] Therefore, how to construct an accurate and effective training dataset to provide sufficient and practical training data for the training process of the corresponding battery management system model is an important problem that urgently needs to be solved. Summary of the Invention

[0005] This application provides a training dataset construction method, apparatus, device, storage medium, and program product to construct an accurate and effective training dataset, providing sufficient and practical training data for the training process of the corresponding model of the battery management system.

[0006] In a first aspect, embodiments of this application provide a method for constructing a training dataset, including: Based on the battery-related information of the target battery model, determine the target battery parameters corresponding to the target battery model; Based on the target battery parameters, multiple historical battery models that match the target battery model are identified, and a migration rule set between the target battery model and the multiple historical battery models is constructed. The migration rule set includes physical feature transformation rules and control strategy transformation rules. Based on the migration rule set, physical quantity migration and message timestamp migration are performed on the historical operation data of multiple historical battery models to obtain the migration operation data. Residual analysis is performed on the migration operation data and the target operation data of the target battery model. Based on the residual analysis results, residual correction is performed on the migration operation data to obtain the corrected operation data. Based on the corrected operating data and the target operating data, a training dataset is constructed. The training dataset is used to train the battery state prediction model in the battery management system corresponding to the target battery model.

[0007] Secondly, embodiments of this application provide a training dataset construction apparatus, comprising: The parameter determination module is used to determine the target battery parameters corresponding to the target battery model based on battery-related information of the target battery model. The rule building module is used to determine multiple historical battery models that match the target battery model based on the target battery parameters, and to build a migration rule set between the target battery model and the multiple historical battery models. The migration rule set includes physical feature transformation rules and control strategy transformation rules. The data migration module is used to perform physical quantity migration and message timestamp migration on historical operating data of multiple historical battery models according to the migration rule set, so as to obtain the migration operating data. The residual correction module is used to perform residual analysis on the migration operation data and the target operation data of the target battery model, and to perform residual correction on the migration operation data based on the residual analysis results to obtain corrected operation data. The dataset building module is used to build a training dataset based on the corrected running data and the target running data. The training dataset is used to train the battery state prediction model in the battery management system corresponding to the target battery model.

[0008] Thirdly, embodiments of this application provide a terminal device, the device including: a processor and a memory storing computer program instructions; When the processor executes computer program instructions, it implements the training dataset construction method as described in the first aspect.

[0009] Fourthly, embodiments of this application provide a computer storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, they implement the training dataset construction method as described in the first aspect.

[0010] Fifthly, embodiments of this application provide a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to perform the training dataset construction method as described in the first aspect.

[0011] The technical solutions provided by the embodiments of this application have at least the following beneficial effects: This application provides a method for constructing a training dataset, comprising: determining target battery parameters based on battery-related information of a target battery model; determining multiple historical battery models matching the target battery model based on the target battery parameters, and constructing a migration rule set between the target battery model and the multiple historical battery models; then, performing physical quantity migration and message timestamp migration on the historical operating data of the multiple historical battery models according to the migration rule set to obtain migrated operating data; further, performing residual analysis on the migrated operating data and the target operating data of the target battery model, and correcting the residuals in the migrated operating data to obtain corrected operating data; and constructing a training dataset for training a battery state prediction model in the battery management system corresponding to the target battery model based on the corrected operating data and the target operating data.

[0012] The technical solution provided in this application utilizes existing historical operating data of historical battery models and a small amount of target operating data of the target battery model for precise transfer and combination. This constructs a high-quality training dataset that can be used to train a battery state prediction model for new battery models, effectively solving the problem that new batteries cannot train effective models due to a lack of sufficient operating data. Through dual transfer of physical quantities and message timestamps, as well as residual correction, this application ensures that the constructed training dataset not only matches the target battery model in terms of physical characteristics, but also that the message timestamps and decision specifications at the control logic level match the battery management system corresponding to the target battery model. The model trained on the dataset constructed based on this application's technical solution can accurately predict the battery state after actual deployment, effectively reducing false alarms or missed alarms, significantly improving model training efficiency and effectiveness, and providing a practical and reliable battery state prediction model for new battery models.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating a training dataset construction method provided in some embodiments of this application; Figure 2 A schematic diagram of the structure of a training dataset construction apparatus provided in some embodiments of this application; Figure 3This is a schematic diagram of the structure of a terminal device provided in some embodiments of this application. Detailed Implementation

[0016] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0017] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0018] It should be noted that the acquisition, storage, use, and processing of data in this application embodiment all comply with the relevant provisions of national laws and regulations.

[0019] Furthermore, it should be noted that in the embodiments of this application, certain software, components, models, and other existing solutions in the industry may be mentioned. These should be considered as exemplary, and their purpose is only to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0020] The Battery Management System (BMS) is the core control unit of a battery. In today's battery industry, battery state prediction models deployed in the BMS can be used to achieve important functions such as accurate thermal runaway early warning, State of Health (SOH) estimation, and Remaining Useful Life (RUL) prediction.

[0021] However, in the process of developing intelligent applications for the battery management system of new battery models, these new models have just rolled off the production line or have not yet been installed in vehicles on a large scale, lacking sufficient real-world operational data, such as data on extreme conditions like thermal runaway and battery aging. This makes it difficult to train an effective battery state prediction model based on limited data. Furthermore, the new battery models differ significantly from older models with existing data in terms of material systems, cell capacity, thermal characteristics, and battery management system control protocols, making it impossible to directly reuse data from older models.

[0022] Furthermore, existing methods for constructing training data for battery management systems (BMS) based on older batteries have serious flaws. While the resulting datasets are voluminous, they fail to fully conform to the actual specifications of BMS systems for new battery models. Battery state prediction models trained on datasets constructed using these methods often experience frequent false negatives and false positives after deployment due to inaccurate training data, severely impacting normal battery operation. Conversely, training models based solely on a small amount of operational data from new battery models can lead to overfitting, significantly affecting the model's generalization ability.

[0023] Based on the aforementioned technical problems, embodiments of this application provide a method, apparatus, device, storage medium, and program product for constructing a training dataset. The method includes: determining target battery parameters based on battery-related information of a target battery model; determining multiple historical battery models matching the target battery model based on the target battery parameters, and constructing a migration rule set between the target battery model and the multiple historical battery models; then, performing physical quantity migration and message timestamp migration on historical operating data of the multiple historical battery models according to the migration rule set to obtain migrated operating data; further, performing residual analysis on the migrated operating data and the target operating data of the target battery model, and correcting the residuals in the migrated operating data to obtain corrected operating data; and constructing a training dataset for training a battery state prediction model in a battery management system corresponding to the target battery model based on the corrected operating data and the target operating data.

[0024] The technical solution provided in this application utilizes existing historical operating data of historical battery models and a small amount of target operating data of the target battery model for precise transfer and combination. This constructs a high-quality training dataset that can be used to train a battery state prediction model for new battery models, effectively solving the problem that new batteries cannot train effective models due to a lack of sufficient operating data. Through dual transfer of physical quantities and message timestamps, as well as residual correction, this application ensures that the constructed training dataset not only matches the target battery model in terms of physical characteristics, but also that the message timestamps and decision specifications at the control logic level match the battery management system corresponding to the target battery model. The model trained on the dataset constructed based on this application's technical solution can accurately predict the battery state after actual deployment, effectively reducing false alarms or missed alarms, significantly improving model training efficiency and effectiveness, and providing a practical and reliable battery state prediction model for new battery models.

[0025] The execution entity used in the embodiments of this application can be a terminal device, such as a desktop computer or laptop computer, or a server. In addition, the execution entity in the embodiments of this application can also be a software entity, such as a client or software program installed on a terminal device. The specific type of execution entity corresponding to the training dataset construction method, apparatus, device, storage medium, and program product provided in the embodiments of this application is not strictly limited here; it can be flexibly selected and set according to the application scenario and actual needs.

[0026] It should be noted that the embodiments provided in this application do not limit the specific application scenarios of the training dataset construction method, apparatus, device, storage medium and program product provided above. The technical solutions provided in the embodiments of this application can be flexibly applied to various actual scenarios that require model training of battery management systems for new batteries, according to actual needs.

[0027] For example, if the battery state prediction model is a battery thermal runaway early warning model, the technical solution provided in this application can match multiple similar historical battery models based on the parameters of the target battery model, and confirm the migration rule set related to thermal runaway. Further, physical quantity migration and message timestamp migration are performed on the historical operating data to generate migration operating data. Then, using a small number of real thermal runaway samples from new battery models, residual correction is performed on the migration operating data to construct a training dataset that can be used to train the thermal runaway early warning model for the target battery model. This application can accurately construct a training dataset for the thermal runaway early warning model corresponding to new battery models, significantly improving the model training effect and efficiency. This enables the trained thermal runaway early warning model to accurately predict battery thermal runaway, significantly improving battery safety and lifespan.

[0028] It should be noted that the application scenarios described in the above embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will understand that with the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0029] Figure 1 This is a flowchart illustrating a training dataset construction method provided in some embodiments of this application.

[0030] like Figure 1 As shown, the training dataset construction method provided in this application embodiment includes steps S101 to S105.

[0031] S101: Based on the battery-related information of the target battery model, determine the target battery parameters corresponding to the target battery model.

[0032] As shown in step S101, in the embodiments provided in this application, the target battery model may be a newly developed battery model or a battery model with a small actual application range, resulting in a small amount of actual operating data. When training the battery state prediction model on the battery management system for such batteries, if the model is trained based on only a small amount of operating data, overfitting may occur, severely reducing the model's generalization ability. Therefore, it is necessary to reasonably adjust the large amount of historical operating data of older battery models to increase the amount of data when training the TV state prediction model for new battery models.

[0033] The aforementioned battery-related information can refer to specific data sources that fully record various dimensions of information such as the electrochemical characteristics, physical properties, design specifications, and usage constraints of the target battery model. In some embodiments, the battery-related information may specifically be the design documents of the target battery model, including but not limited to: cell design and specification documents, battery management system software and hardware product documents, and relevant academic literature.

[0034] Target battery parameters can refer to various specific parameters and indicators recorded in battery-related information that characterize the electrochemical and physical properties of the target battery model and the control logic specifications of the battery management system. In some embodiments, the specific parameter types included in the target battery parameters can be flexibly selected according to the specific functions of the battery state prediction model. For example, when the battery state prediction model is a thermal runaway early warning model, the target battery parameters may include, but are not limited to: the battery internal resistance, heat capacity, voltage plateau, convective heat transfer coefficient, capacity change curve, DC internal resistance (DCR) curve, open circuit voltage (OCV) curve, thermodynamic parameters (such as specific heat capacity, thermal runaway trigger temperature, etc.) corresponding to the target battery model, and control strategy parameters corresponding to the battery management system (such as control cycle, alarm threshold, de-jittering frame count, etc.).

[0035] Regarding the specific process of determining the target battery parameters, in the embodiments provided in this application, a preset parameter analysis model can be used to perform parameter parsing on battery-related information based on the parameter parsing template corresponding to the battery state prediction model, and extract the target battery parameters corresponding to the parameter parsing template.

[0036] In some embodiments, the parameter parsing template is used to indicate the specific types of parameters that the preset parameter analysis model needs to extract from battery-related information, similar to target battery parameters. The parameter parsing template can be flexibly set according to the specific functions of the battery state prediction model. In addition, the specific model type of the preset parameter analysis model can also be flexibly set according to actual needs and application scenarios. In some embodiments, the preset parameter analysis model can be a Large Language Model (LLM), while in other embodiments it can be other models or algorithms that extract target parameters from information and documents.

[0037] The above embodiments, through the combination of a preset parameter analysis model and a parameter parsing template, achieve the structured and standardized extraction of battery parameters, improving the processing efficiency and accuracy of parameter extraction for target battery models. This embodiment provides an accurate and reliable data foundation for the subsequent screening of historical battery models and the construction of migration rules, further enhancing the effectiveness of the training dataset for the subsequent construction of battery state prediction models in battery management systems.

[0038] S102: Based on the target battery parameters, determine multiple historical battery models that match the target battery model, and construct a set of migration rules between the target battery model and the multiple historical battery models.

[0039] As shown in step S102, in the embodiments provided in this application, battery models with more historical operating data than the target battery model can be used as older battery models. Furthermore, the large amount of historical operating data recorded for these older battery models, along with the corresponding historical battery parameters, can be stored in a database according to the battery model. This database can serve as a historical parameter library for historical battery models. Based on this, when constructing a training dataset for the target battery model, battery model matching can be performed based on the target battery parameters to determine multiple historical battery models that are close to the target battery model.

[0040] Specifically, in some embodiments, for each candidate historical battery model in a pre-built historical parameter library, the parameter difference between the target battery parameters and the historical battery parameters corresponding to that candidate historical battery model can be determined. The parameter types included in the historical battery parameters can correspond to the target battery parameters.

[0041] In some embodiments, when calculating the parameter difference, the difference between the target battery parameters and the historical battery parameters can be calculated separately for each parameter type, and this difference can be used as the difference for that parameter type. For example, in the example above, the differences between the target battery model and the candidate historical battery model in terms of battery internal resistance, thermal capacity, voltage platform, etc., can be calculated separately. Then, the absolute values ​​of the differences for multiple parameter types can be calculated and normalized, and then a weighted sum can be performed to calculate the overall parameter difference between the target battery model and the candidate historical battery model, with a specific value ranging from 0% to 100%.

[0042] Furthermore, the parameter difference can be compared with a preset difference threshold (e.g., 15%, 10%). When the parameter difference between the candidate historical battery model and the target battery parameters is less than or equal to the preset difference threshold, i.e., the difference between the parameters is small, the candidate historical battery model can be used as the historical battery model corresponding to the target battery model. Based on this, each candidate historical battery model in the historical parameter library can be tested to filter out multiple historical battery models that are similar to the target battery model. In some embodiments, the number of historical battery models filtered out should be equal to or greater than two. This allows for effective suppression of errors introduced by a single historical battery model due to its own data bias or individual characteristics through cross-validation between multiple migration data from different historical battery models during subsequent residual analysis and residual correction. This reduces the over-correction of migration data in the wrong direction due to relying solely on single battery data, thus preventing overfitting.

[0043] In the above embodiments, based on the quantitative calculation and threshold screening of parameter differences, multiple historical models that are closest to the target battery are automatically matched from the historical parameter library as the data source for subsequent migration operation data. The screening of multiple historical battery models enables the subsequent migration operation data to cover a wider range of operating conditions, significantly improving the data diversity and construction efficiency of the constructed training dataset.

[0044] After identifying multiple historical battery models that match the target battery model, a set of migration rules can be further determined based on the historical battery parameters of the historical battery models to migrate the historical operating data of the historical battery models to the target battery model.

[0045] Specifically, in the embodiments provided in this application, target battery parameters can be categorized according to their actual meaning into physical characteristic parameters representing the specific physical characteristics of the target battery model, such as battery internal resistance, heat capacity, voltage platform, thermodynamic parameters, etc., and control strategy parameters corresponding to the battery management system's battery control strategy for the target battery model, such as battery control cycle, parameter alarm threshold, de-jittering frame rate, etc. Historical battery parameters corresponding to historical battery models can also be categorized into these two types of battery parameters.

[0046] During the construction of the migration rule set, for each historical battery model, the conversion relationship between the physical characteristic parameters in the target battery parameters and the historical physical characteristic parameters corresponding to that historical battery model can be determined, thus obtaining the physical characteristic transformation parameters corresponding to that historical battery. Simultaneously, the conversion relationship between the control strategy parameters in the target battery parameters and the historical control strategy parameters corresponding to that historical battery model can be determined, thus determining the control strategy transformation parameters corresponding to that historical battery.

[0047] The physical characteristic transformation parameters and control strategy transformation parameters can specifically represent the parameter processing required to transform historical battery parameters to target battery parameters. For example, physical characteristic transformation parameters can include the amplification / reduction of the temperature rise slope by a factor of n and the increase / reduction of the voltage drop slope by a factor of m in historical battery parameters. Control strategy transformation parameters can include specific period adjustment values ​​for adjusting the control period in historical battery parameters to the control period corresponding to the target battery model, and specific frame number adjustment values ​​for setting the number of frames to be jitter-free to the target battery model.

[0048] Furthermore, at least one physical feature transformation rule can be constructed based on the physical feature transformation parameters to perform migration processing on the historical operating data of historical battery models. Simultaneously, at least one control strategy transformation rule can be constructed based on the control strategy transformation parameters to perform migration processing on the historical operating data of historical battery models. Then, the physical feature transformation rules and control strategy transformation rules are integrated to obtain a migration rule set used to guide the migration of operating data between the target battery model and historical battery models.

[0049] Through the above processing, multidimensional physical feature transformation parameters and control strategy transformation parameters can be extracted sequentially from various battery parameters and then decoupled and executed in the subsequent migration process. This ensures that multidimensional temporal coupling changes caused by differences in physical feature parameters such as internal resistance and thermal capacity are transferred collaboratively. Furthermore, multiple migration rule sets corresponding to different historical battery models provide a sufficient data foundation for the subsequent construction of training datasets.

[0050] The physical feature transformation rules and control strategy transformation rules in the migration rule set correspond to the physical feature transformation parameters and control strategy transformation parameters determined above, respectively, and are specifically descriptions of the actual rules for parameter transformation. In some embodiments, the migration rule set can be represented as a configuration file of a JSON object.

[0051] For example, if the target battery type is a high-nickel semi-solid-state battery (NCM811), the target battery parameters could include: a charge capacity of 55 Ah, a cell weight of 0.82 kg, and a specific heat capacity of 1.05 J / (g•K). The corresponding historical battery type could include a conventional ternary liquid battery (NCM523), and historical battery parameters could include: a charge capacity of 50 Ah, a cell weight of 0.75 kg, and a specific heat capacity of 1.10 J / (g•K).

[0052] Based on the above embodiments, the migration rule set can be determined as follows: { "rule_id":"R001_NCM523_to_NCM811_Migration", / / Rule ID: The set of migration rules from NCM523 to NCM811; "source_bms_model":"NCM523_50Ah", / / Source battery management system model: NCM523, charge capacity 50Ah; "target_bms_model":"NCM811_55Ah_Solid", / / Target battery management system model: NCM811, charge capacity 55Ah, semi-solid; "physical_transformations": { / / Rules for physical feature transformations; "temperature": { / / temperature; "operator":"scale_and_shift", / / Operator: scaling and translation; "params": { / / parameters; "rate_scale_factor": 1.3, / / Temperature rise rate scaling factor: The cell temperature rise rate of the target battery model is 1.3 times that of the historical battery models; "threshold_shift": -5.0 / / Temperature threshold shift: Temperature alarm threshold is reduced by 5℃ (degrees Celsius); } }, "voltage": { / / voltage "operator":"scale_drop_rate", / / Operator: scales the rate of voltage drop; "params": { / / parameters; "drop_rate_scale_factor": 0.85, / / Voltage drop rate scaling factor: The cell voltage drop rate of the target battery model is 0.85 times that of the historical battery models; "platform_shift_v": -0.02 / / Voltage platform offset: The operating voltage platform of the target battery model is 0.02V lower than that of the historical battery model; } }, "soc": { / / State of charge; "operator":"remap_ocv", / / Operator: Remaps the open-circuit voltage curve; "params": { / / parameters; "ocv_curve_id":"NCM811_OCV_v1.2" / / Calculate the state of charge using the open-circuit voltage curve table of version 1.2 corresponding to the target battery model; } } }, "control_logic_override": { / / Control logic override (control strategy transformation rules); "thermal_alarm_threshold_celsius": 55.0, / / Thermal alarm threshold: The overheat protection trigger temperature for the target battery model is set to 55℃; "overvoltage_threshold_v": 4.25, / / Overvoltage protection threshold: The overvoltage protection voltage for the target battery model is set to 4.25V; "debounce_frames": 3, / / Debounce frame count: The fault signal needs to be confirmed for 3 consecutive frames before an alarm is triggered; "control_period_hz": 1 / / Control cycle: The main cycle frequency of the battery management system is 1Hz (Hertz); } } As shown in the example above, `rule_id` serves as a unique rule identifier, strongly linking the physical feature transformation rule with the control strategy transformation rule. The migration rule set constructed through the above embodiments can include multi-dimensional physical feature transformation rules and control strategy transformation parameters that migrate parameters of historical battery models to the corresponding parameters of the target battery model, providing accurate guidance for subsequent migration processing of historical operational data. In some embodiments, the extraction of physical feature transformation parameters and control strategy transformation parameters, as well as the construction of the migration rule set, can be performed by a preset parser. If the parameters or migration rule set output by the parser do not conform to the preset format, a retry mechanism can be automatically triggered. If the number of retries exceeds a preset number (e.g., 3 times, 5 times, etc.), an anomaly can be confirmed, and a prompt message can be generated to alert the user or relevant personnel for timely handling, ensuring the stability of the training dataset construction process.

[0053] In the above embodiments, two types of parameters—physical characteristics and control strategies—are each constructed with their own independent transformation rules. This ensures that subsequent physical quantity migration and message timestamp migration can be decoupled and executed, preventing interference between the migration operations of the two types of data. The rule construction in this embodiment provides accurate migration instructions for subsequent data migration, and the decoupled execution of different parameter types effectively reduces repetitive configuration costs, significantly improves data migration efficiency, and enhances the usability of the migrated data.

[0054] S103: Based on the migration rule set, perform physical quantity migration and message timestamp migration on the historical operation data of multiple historical battery models to obtain migration operation data.

[0055] As shown in step S103, in the embodiments provided in this application, based on the physical feature transformation rules and control strategy transformation rules in the migration rule set, the historical operating data corresponding to each historical battery model can be migrated in terms of physical quantity and message timestamp in the control strategy in turn, thereby migrating the historical operating data into migration operating data that can be used to train the battery state prediction model corresponding to the target battery model.

[0056] Specifically, based on physical characteristic change rules, migration processing can be performed on the data corresponding to each physical quantity in historical operational data at the original sampling frequency. These physical characteristic change rules can correspond to different transformation types; in some embodiments, the transformation types can be divided into linear transformations and nonlinear transformations. Linear transformations include, for example, "scale_and_shift" in the aforementioned migration rule set, while nonlinear transformations include, for example, "polynomial" and "piecewise_nonlinear".

[0057] During the physical quantity transfer process, based on the transformation type corresponding to the physical characteristic change rule, a corresponding computational algorithm can be selected to transfer the physical quantities corresponding to the physical characteristic change rule in the historical operating data, thus obtaining physical quantity transfer data. It should be noted that the sampling frequency corresponding to the determined physical quantity transfer data should be consistent with the sampling frequency of the historical operating data. This avoids the accidental loss of rapid nonlinear jump characteristics within seconds before battery failure or abnormal problems (such as thermal runaway) due to sampling frequency adjustments, thereby improving the accuracy and practicality of the physical quantity transfer data.

[0058] Furthermore, for physical quantity migration data, based on the control strategy transformation rules, the timestamp migration parameters corresponding to the training data of the target battery model can be determined from the historical operation data. Furthermore, based on the timestamp migration parameters, the timestamps of each message in the physical quantity migration data can be migrated and adjusted accordingly, thereby obtaining migration operation data in which both physical quantities and message timestamps have achieved corresponding migration.

[0059] In the embodiments provided in this application, the timestamp migration parameters determined based on the control strategy transformation rules can correspond to the control strategy parameters determined during the construction of the migration rule set. Specifically, they can represent the adjustment values ​​of each control parameter when migrating the battery management system control strategy of a historical battery model to the target battery model. In some embodiments, the timestamp migration parameters may include, but are not limited to: the parameter fault threshold corresponding to the target battery model, the target sampling frequency, and the number of de-jittering operations (e.g., the number of de-jittering frames in the example above). These parameters can be flexibly set according to the specific model function of the battery state prediction model.

[0060] Regarding the specific execution process of the message timestamp migration process, in some embodiments, it can be determined first, based on the target sampling frequency in the timestamp migration parameters, whether the sampling frequency of the physical quantity migration data is greater than the target sampling frequency.

[0061] If the sampling frequency of the physical quantity transfer data is greater than the target sampling frequency, it indicates that the physical quantity transfer data is higher frequency data compared to the target battery model. In order to make the transferred data conform to the target sampling frequency, data can be extracted from the physical quantity transfer data according to the target sampling frequency, thereby obtaining the sampling operation data corresponding to the target sampling frequency.

[0062] For example, if the sampling frequency of a historical battery model is 10Hz, and the target sampling frequency of the target battery model is 1Hz, then the physical quantity migration data with a sampling frequency of 10Hz can be processed by extracting data points at a frequency of 1Hz to obtain the down-frequency sampling operation data.

[0063] Then, based on one or more fault threshold parameters in the timestamp migration parameters, fault detection can be performed on the corresponding parameter data in the sampled operation data to determine the fault time corresponding to the fault data. Further, a de-jittering judgment can be performed on the consecutive counts of the determined fault timestamps. When a fault time exists in the sampled operation data, and the consecutive counts of the fault timestamps are greater than or equal to the de-jittering count in the timestamp migration parameters, the fault time whose consecutive counts reach the de-jittering count can be used as the fault message timestamp corresponding to the target battery model in the sampled operation data.

[0064] Furthermore, based on the determined fault message timestamps, the timestamps of the sampled operation data can be updated, that is, the original fault message timestamps corresponding to the historical battery models can be replaced with the fault message timestamps corresponding to the target battery models, so as to obtain migration operation data in which both physical quantity values ​​and fault timestamps are accurately migrated.

[0065] For example, taking the battery state prediction model in the above example as a thermal runaway early warning model, if the timestamp migration parameters include the temperature fault threshold corresponding to the target battery model, such as 55℃, the target sampling frequency is 1Hz, and the number of jitter removals is 3 frames.

[0066] After extracting data points from the 10Hz physical quantity migration data, the sampled operation data was obtained. Based on the temperature fault threshold, the sampled operation data was determined to contain the following multiple consecutive fault moments: Frame 1 (corresponding to Ns (seconds)): Temperature value ≥ 55℃, de-jitter count = 1; Frame 2 (corresponding to N+1s): Temperature value ≥55℃, de-jitter count = 2; Frame 3 (corresponding to N+2s): Temperature value ≥ 55℃, dejitter count = 3, alarm triggered. This moment is used as the fault message timestamp for the target battery model in the temperature fault dimension. It can be seen that in this example, when the number of consecutive fault moments reaches the dejitter count in the timestamp migration parameters, the fault message timestamp corresponding to the target battery model can be accurately determined, thereby achieving precise timestamp offset for physical quantity migration data.

[0067] In some embodiments, when the target sampling frequency is equal to or greater than the sampling frequency of the physical quantity migration data (e.g., both are 10Hz), the above data extraction stage can be skipped, and the physical quantity migration data of equal frequency or low frequency can be directly used as the sampling operation data. This allows for further fault time detection and jitter reduction count detection as described in the above embodiments, and timestamp migration based on the fault message timestamp to obtain the migration operation data. In some embodiments, the data extraction based on the target sampling frequency in the above embodiments can be directly replaced by frame-by-frame jitter reduction judgment of the physical quantity migration data, improving the timestamp migration rate while fully ensuring the accuracy of the fault timestamp after migration.

[0068] The aforementioned timestamp migration process, through data extraction, fault moment detection, and debouncing, ensures that the fault message timestamps in the migration operation data fully conform to the actual control cycle and judgment specifications of the battery management system corresponding to the target battery model. This avoids the error problem of misaligned fault timestamps caused by blindly using historical data timestamps. This embodiment achieves strict alignment between the migration operation data and the true characteristics of the target battery model in terms of physical features and fault moments, providing sufficient data support for subsequent training data construction and model training. This effectively reduces the problem of missed or false alarms in battery state prediction and significantly improves the model's predictive capability.

[0069] In the above embodiments, physical quantity migration is first performed using different transformation methods for different physical quantities, significantly improving migration accuracy. Furthermore, the physical quantity migration process completely preserves the original sampling frequency, retaining complete high-frequency details for the subsequent message timestamp migration stage and avoiding the loss of key features in the data due to frequency reduction or interpolation. In addition, in this embodiment, the message timestamp migration and physical quantity migration of the running data are executed independently without interference, ensuring that the migration running data is precisely aligned with the target battery model in both physical features and control logic dimensions, providing effective data for subsequent training dataset construction and model training.

[0070] In addition, in the embodiments provided in this application, before performing runtime data migration based on the migration rule set, this application considers that formulating migration rules solely based on the differences between parameters may result in some constructed rules that clearly violate physical principles. Data obtained after migration based on migration rules that clearly violate physical principles is obviously anomalous data. During the model training phase, such data will seriously mislead the model, affecting the actual training effect and efficiency.

[0071] Based on this, in some embodiments, before determining the migration operation data based on the migration rule set, physical constraint checks can be performed on each rule in the migration rule set, and migration rules that do not conform to physical principles can be eliminated in advance to prevent abnormal data from affecting the accuracy of training dataset construction during the subsequent migration process.

[0072] Specifically, we can first calculate the average parameter values ​​for multiple historical battery models based on their corresponding historical operating data. Since the historical operating data is collected from the actual operation of the historical battery models, the actual distribution and changes of each parameter conform to the corresponding physical principles.

[0073] Then, based on the transformation scaling factor corresponding to the parameter type of the average parameter value in the migration rule set, the average parameter value can be mapped to calculate the predicted parameter value for the target battery model corresponding to this parameter type. Specifically, the transformation scaling factor can represent the degree of difference between historical battery models and the target battery model for the same parameter type. Simultaneously, based on a small amount of pre-collected target operating data, the actual parameter value for the target battery model for the same parameter type can be calculated.

[0074] Furthermore, based on the actual and predicted parameter values, the migration rules in the migration rule set can be physically constrained, and migration rules that clearly violate physical constraints can be removed, thus obtaining a validated rule set. Specifically, this can be determined by comparing the magnitudes of the actual and predicted parameter values. When the predicted parameter value is greater than the actual parameter value, it indicates that the data migrated based on the relevant migration rules does not match the actual performance of the target battery model, clearly violating actual physical principles. To avoid affecting the determination of subsequent migration operation data and the construction of the training dataset, such rules should be removed promptly.

[0075] Taking the battery state prediction model as a thermal runaway early warning model in the above example as an example, in some embodiments, physical constraint verification can be performed on the rules related to battery heat generation in the migration rule set. The specific battery parameters targeted can be the average heat generation power of the target battery model and the historical battery models.

[0076] In this example, for a historical battery model, the corresponding historical average heat output can be calculated based on historical operating data. Specifically, this can be done using the following formula (1): (1) in, This represents the historical average heat output power corresponding to multiple historical battery models. This indicates the cell quality, specifically the average cell quality across multiple historical battery models. This indicates the specific heat capacity of the battery cell, which can be the average specific heat capacity of cells from multiple historical battery models. This represents the average rate of temperature rise. Indicates the temperature change value. This indicates the time difference.

[0077] Then, the historical average heat production power calculated based on formula (1) can be used. Multiplying by the transformation parameters corresponding to the heat generation power of the target battery model and historical battery models in the migration rule set, i.e., the aforementioned transformation scaling factor, the predicted average heat generation power of the target battery model in the migration operation data is calculated. Furthermore, based on a small amount of target operating data corresponding to the target battery model, the actual average heat generation power of the target battery model can be calculated using a formula similar to formula (1). In some embodiments, the actual average heat production power is calculated. When, the average temperature rise rate in formula (1) can be used. Replace with the maximum temperature rise rate of the target battery model in the target operating data. This allows for accurate detection and prediction of average heat production power. Does it satisfy physical principles under all operating conditions?

[0078] In this example, if the calculated predicted average heat production power Greater than based on the maximum temperature rise rate The calculated actual average heat production power This indicates that the corresponding migration rules clearly violate the physical principles of heat generation during normal use of the actual target battery model. The data migrated based on these rules also clearly fails to reflect the true operating conditions of the target battery model. Therefore, the related migration rules should be removed, or the previous step should be revisited to re-determine them, thereby ensuring the accurate and stable execution of subsequent migration processes. Then, the migration process described in the above embodiment can be executed again on the verified rule set to determine the migration operation data corresponding to the target battery model.

[0079] In the above embodiments, physical constraint checks are performed on the migration rules before data migration, eliminating migration rules that may violate physical constraints at the source and avoiding contamination of the migration execution data. Through this rule verification process, the migration execution data obtained based on the verified migration rule set all meet physical feasibility constraints, ensuring the physical rationality of the training dataset from the data source, reducing the risk of subsequent model training deviations due to data distortion, and significantly improving model training efficiency and the actual capabilities of the trained model.

[0080] S104: Perform residual analysis on the migration operation data and the target operation data of the target battery model, and based on the residual analysis results, perform residual correction on the migration operation data to obtain corrected operation data.

[0081] As shown in step S104, in the embodiments provided in this application, it is considered that although the migration operation data is obtained by migrating the historical operation data of the historical battery model based on the target battery parameters, it is ultimately not the actual operation data of the target battery model that is collected. Therefore, there may be a certain degree of deviation between the migration operation data and the target operation data.

[0082] Based on this, before constructing the training dataset, residual analysis and correction can be performed on the migration running data based on a small amount of target running data, so that the corrected running data can be closer to the target running data and conform to the actual running conditions of the target battery model.

[0083] Specifically, in some embodiments, based on multiple preset feature dimensions, the feature vectors corresponding to the migration operation data and target operation data for each historical battery model are determined for each feature dimension, thereby obtaining multiple migration feature vectors corresponding to the migration operation data and multiple target feature vectors corresponding to the target operation data.

[0084] In some embodiments, the feature dimensions can be flexibly set according to the actual model function of the battery state prediction model corresponding to the battery management system. For example, if the battery state prediction model is the thermal runaway early warning model mentioned above, the corresponding feature dimensions may include, but are not limited to: peak temperature rise rate, fault trigger time, voltage drop inflection point slope, etc.

[0085] Then, residual analysis can be performed on the migration feature vector and the target feature vector under the same feature dimension to determine the difference between the vectors, that is, the single-dimensional residual vector between the migration running data and the target running data under this feature dimension. By combining the single-dimensional residual vectors of multiple feature dimensions, an overall residual vector between the migration running data and the target running data can be constructed as the result of the residual analysis.

[0086] For example, if the average fault trigger time in the migration operation data is 1622 seconds, while the average fault trigger time in the target operation data is 1580 seconds, the single-dimensional residual vector in this feature dimension would be 42 seconds. In some embodiments, the residual analysis result can be specifically expressed as Δ=[ , ,..., ], to These are the single-dimensional residual vectors between the migration running data and the target running data, representing n feature dimensions.

[0087] Furthermore, by pre-setting a correction detection model, based on the residual analysis results and migration rule set determined above, the corresponding correction intensity can be determined for the data corresponding to each feature dimension of the migration run data. The correction intensity represents the adjustment strength to the data corresponding to the feature dimension. Then, based on the correction intensity, residual correction can be performed on the feature dimensions in the migration run data that show significant differences compared to the target run data, resulting in corrected run data.

[0088] In some embodiments, the preset correction detection model may have a similar model structure or function to the preset parameter analysis model in the above embodiments, or may be the same model, such as a large language model, or other models that can determine the existence of rules that need to be corrected in the migration rule set based on the residual analysis results. The specific model can be flexibly selected according to actual needs and application scenarios.

[0089] Regarding the specific process for determining the correction intensity, in the embodiments provided in this application, the mean difference between the migration run data and the target run data in each feature dimension can be calculated first based on the single-dimensional residual vector corresponding to each feature dimension in the residual vector (residual analysis result). Simultaneously, the data dispersion of the migration run data and the target run data in each feature dimension can be calculated. In some embodiments, the data dispersion can specifically be the variance of the data in each feature dimension of the migration run data or the target run data, which can be calculated based on the mean of the data in different feature dimensions.

[0090] Then, the calculated mean differences and data dispersion for each feature dimension can be input into the preset correction detection model. This allows the preset correction detection model to perform importance analysis on the rules corresponding to multiple feature dimensions in the migration rule set constructed in the above steps based on the received data, and determine the importance weight corresponding to each feature dimension. The importance weight is used to represent the basic adjustment range of the data corresponding to that feature dimension in the subsequent correction process.

[0091] In some embodiments, the preset correction detection model can generate a diagnostic report containing importance weights corresponding to multiple feature dimensions based on the mean difference and data dispersion. For example, in a scenario where the battery state prediction model is a thermal runaway early warning model, the diagnostic report output by the preset correction detection model may include: the temperature rise amplification of 1.3 times in the migration rule set is too conservative and fails to fully reflect the effect of local heat accumulation caused by the low thermal conductivity of the semi-solid battery (target battery model); it is recommended to increase the importance weight of the peak temperature rise rate, for example, to 0.9; the voltage decay inflection point deviation in the migration rule set is small; it is recommended to reduce the importance weight, for example, to 0.2, etc. This example is only used to illustrate the diagnostic report; the specific report content and format can be flexibly selected according to actual needs and application scenarios.

[0092] Furthermore, for each feature dimension, the adjustment coefficient corresponding to the importance weight of the above prediction correction detection model output can be calculated based on the data dispersion of the migration running data and the target running data in that feature dimension, as well as the number of samples of the target running data.

[0093] In some embodiments, the adjustment coefficient can characterize the distribution quality of the target running data and the migration running data, and its magnitude can be used to evaluate the confidence level of the importance weights output by the preset correction detection model. The larger the amount of target running data or the smaller the data dispersion, the larger the adjustment coefficient, indicating that the corresponding importance weights are more stable and reliable. Conversely, it indicates that the importance weights may have low confidence, and a conservative correction should be adopted when making subsequent corrections.

[0094] Furthermore, for each feature dimension, the product of the corresponding adjustment coefficient and the importance weight can be calculated as the correction strength of the transfer running data for that feature dimension. In some embodiments, the upper limit of the correction strength is a preset multiple of the importance weight, such as 0.5 times or 0.6 times the importance weight. The purpose of this design is to prevent the transfer running data from being forced to be completely consistent with the target running data even if the amount of target running data is large enough. This effectively prevents overfitting problems in the battery state prediction model after building the training dataset, and improves the actual training effect and generalization ability of the model.

[0095] In the embodiments provided in this application, the specific calculation process for determining the correction strength based on importance weight can be referred to the following formula (2): (2) in, Represents the first of multiple feature dimensions The correction strength corresponding to each feature dimension This represents the importance weight determined by the pre-defined modified detection model for the j-th feature dimension. This indicates the number of data points in the target runtime data. The preset smooth equivalent sample number is used to adjust the target running data volume for the correction of intensity gain rate or attenuation rate. In some embodiments, the smooth equivalent sample number... The value can be 5, 6, etc.

[0096] This indicates the confidence level of the target operational data affected by the volume of the target operational data. When the data volume... When it is extremely small (e.g., 2), the value is small (e.g., Take 5, that is =2 / 7≈0.29), which can effectively limit the correction strength to prevent overfitting during subsequent model training. And when the amount of data... When the value is large (e.g., 40), it approaches 1 (e.g., Take 5, that is =40 / 45≈0.89), which can effectively improve the correction strength, thereby effectively correcting the migration operation data.

[0097] and These represent the target runtime data and migration runtime data at the [number]th [year]. Data dispersion across each feature dimension This is a preset, relatively small constant used to prevent the denominator from being zero. This represents the dispersion penalty term affected by data dispersion. When the dispersion of the migration running data is much greater than that of the target running data, it indicates that the distribution of the migration running data is too scattered. Directly correcting based on importance weights may lead to further data distortion. This term will approach 0, thus reducing the correction strength. When the dispersion of the migration running data and the target running data are close, this term will approach 0.5. That is, the aforementioned importance weights The corresponding adjustment factor.

[0098] Based on the above formula (2), the correction strength corresponding to multiple adjustment dimensions can be accurately calculated. Furthermore, as in the above embodiment, even with a large amount of data... The data is sufficient, and the data dispersion between the migration operation data and the target operation data is close, so the upper limit of the correction strength is also 0.5. The purpose of this design is to prevent the migration run data from being forcibly corrected to be consistent with or too similar to the target run data, thereby avoiding overfitting problems during subsequent model training and fully ensuring the generalization ability of the battery state prediction model after training. In other embodiments, when the amount of data... If the data size is too small, or if the data dispersion between the migration run data and the target run data is too large, the overall correction strength will be close to zero. In this case, you can choose not to correct the migration run data to prevent forced correction from causing overfitting problems in model training.

[0099] Furthermore, all terms in the above formula (2) are very important; without the importance weights determined by the pre-defined correction detection model, the overall importance of the formula would be significantly reduced. In the actual data correction process, if the same correction is applied to feature dimensions of different importance, it is very likely that less important dimensions will be over-corrected, or additional noise will be introduced. Furthermore, if there is a lack of target operational data confidence levels related to the target operational data volume, the correction may be ineffective. When only a small number of samples are used for correction, the direction of correction is easily influenced by extreme samples, leading to severe overfitting during model training. Furthermore, the lack of a discrete penalty term... When the migration data has a high degree of dispersion in some feature dimensions, forcibly correcting the migration data to a small number of target samples will cause the data distribution to collapse after correction, resulting in the loss of coverage of extreme working conditions, and thus affecting the model training process.

[0100] The process of determining the correction intensity described above introduces an adjustment coefficient, allowing the correction intensity to adaptively adjust based on the number of samples and data dispersion. When the sample size is too small or the variance of the transfer data is too large, the correction magnitude is automatically reduced to prevent data distribution collapse due to overfitting a small number of real samples. Simultaneously, the upper limit of the correction intensity is limited to a preset multiple of the importance weight. Even with sufficient samples and low data dispersion, the correction will not be indiscriminately amplified. This effectively corrects bias while preserving the original diversity of the transfer data, significantly improving the generalization ability of the model after training.

[0101] The above embodiments can accurately determine the correction intensity of migration operation data in different feature dimensions, thereby performing precise data correction and obtaining corrected operation data. It should be noted that, regarding the specific correction process, in the embodiments provided in this application, for each feature dimension, the migration feature vector of the migration operation data in that feature dimension can be adjusted based on the corresponding correction intensity. Furthermore, based on the adjusted migration feature vector, data inverse mapping is performed to obtain initial corrected data with the same sampling frequency as the historical operation data.

[0102] In some embodiments, the adjusted migration feature vector can be back-mapped using nonlinear interpolation. In some embodiments, piecewise cubic spline interpolation can be selected to perform the back-mapping. The reason why linear interpolation is not chosen in this application is that linear interpolation may generate broken lines between data points, which will destroy the smooth transition characteristics of battery parameters (such as physical quantities such as temperature and voltage), and for models such as thermal runaway warning, it may produce non-physical slope abrupt changes in the rapid temperature rise stage before thermal runaway. Nonlinear interpolation, such as piecewise cubic spline interpolation, can maintain the continuity of the second derivative of the data, thereby better preserving the physical authenticity of the high-frequency nonlinear jump characteristics in the migration operation data.

[0103] Furthermore, for the initial correction data with the same sampling frequency as the historical operating data, the values ​​of each physical quantity in the bus message can be updated, and the message timestamp migration process in the above embodiment can be re-executed to obtain the fault timestamp and the correction operating data with the sampling frequency matching the target battery model, effectively ensuring the consistency between the control logic of the correction operating data and the target operating data.

[0104] In the above embodiments, multi-feature dimension residual analysis can accurately identify the deviation distribution between the migration operation data and the target operation data in each physical feature dimension. By using a preset correction detection model to determine the correction intensity corresponding to each dimension based on the residual analysis results and the migration rule set, differentiated rather than completely uniform correction intensities are applied to different feature dimensions. This effectively avoids the problems of irrelevant dimensions introducing noise due to equal-weighted correction or insufficient correction in key dimensions leading to residual deviations. Residual correction based on correction intensity ensures that the corrected operation data retains the original diversity of the migration data while aligning the feature distribution of each dimension towards the target operation data. This ensures a reasonable proportion of data correction across dimensions, significantly improving the effectiveness of subsequent battery state prediction model training and the generalization ability of the trained model.

[0105] In addition to the above, in the embodiments provided in this application, considering that the migration operation data before residual correction may cause some physical quantities after migration to violate the law of conservation of energy due to data migration processing, if subsequent residual correction is carried out based on the migration data that violates the law of conservation of energy, it will seriously affect the accuracy and effectiveness of the training dataset corresponding to the subsequent battery state prediction model, thereby introducing erroneous data in the model training process.

[0106] Based on this, in the embodiments provided in this application, energy conservation verification can be performed on the migration operation data to ensure that the actual migration operation data for residual correction conforms to the law of energy conservation. Specifically, the expected energy deviation threshold corresponding to the target battery model can be determined based on the historical battery parameters corresponding to the historical battery model and the target battery parameters. The expected energy deviation threshold is used to represent the energy conservation benchmark that the data after migration should meet, that is, the energy deviation limit corresponding to the data after migration.

[0107] Then, for each migrated data point in the migration operation data, the corresponding output energy and input energy are determined, specifically electrical energy. Further, based on the output and input energy, the actual energy deviation corresponding to each migrated data point is calculated. When the calculated energy deviation exceeds the previously determined expected energy deviation threshold, it indicates that the migrated data does not satisfy energy conservation after the previous migration processing, the energy deviation exceeds expectations, and it is considered erroneous data.

[0108] Therefore, migrated data with energy deviations exceeding the expected energy deviation threshold should be removed to prevent contamination of subsequent corrected operational data, resulting in verified operational data. Then, during the residual correction process described in the above embodiments, residual correction can be applied to the verified operational data.

[0109] In some embodiments, taking the battery state prediction model as a thermal runaway early warning model as an example, the calculation process of the expected energy deviation threshold corresponding to the thermal runaway early warning model can refer to the following formula (3): (3) in, This indicates the expected energy deviation threshold corresponding to the target battery model. This indicates the preset default energy deviation threshold that satisfies the law of conservation of energy, such as 5%, 3%, etc. and These represent the specific heat capacity of the target battery model and the historical battery model, respectively. and These represent the cell quality of the target battery model and the historical battery model, respectively.

[0110] The specific principle corresponding to the above formula (3) is: In the field of battery temperature control, batteries with larger heat capacity allow for a larger energy deviation tolerance because their thermal inertia has a stronger buffering capacity for temperature distribution. Formula (3) can accurately calculate the expected energy deviation threshold corresponding to the target battery model, for example: =5%, =1.05, =0.82, =1.10, =0.75, corresponding to the expected energy deviation threshold .

[0111] In this example, if the energy deviation of each post-migration data determined based on the migration operation data corresponding to the thermal runaway prediction model is greater than the expected energy deviation threshold calculated by the above formula (3), then... If the data violates the law of energy conservation, it should be removed. By performing corresponding calculations and processing on each transferred data point, we can obtain verified running data that conforms to the law of energy conservation for the thermal runaway prediction model, providing accurate data support for subsequent residual correction and training dataset construction.

[0112] In the above embodiments, before residual analysis, the expected energy deviation threshold corresponding to the target battery model is calculated first. Then, migration data that violates the energy conservation rule is removed in advance, ensuring that all migration data entering the subsequent residual analysis stage has physical rationality. This avoids interference with the correction direction and subsequent dataset construction caused by erroneous data participating in the correction calculation. The energy conservation verification in this embodiment allows subsequent correction processing to focus on migration data that maintains energy conservation, significantly improving the accuracy of residual correction and ensuring the rationality of the corrected data and the accuracy of the training dataset construction.

[0113] S105: Based on the corrected running data and the target running data, construct a training dataset. The training dataset is used to train the battery state prediction model in the battery management system corresponding to the target battery model.

[0114] As shown in step S105, in the embodiments provided in this application, the corrected running data obtained by the above steps can be merged with the target running data to obtain a training dataset for training the battery state prediction model corresponding to the target battery model.

[0115] In some embodiments, the encapsulated training dataset may contain multi-dimensional data such as time-series data groups, bus message subgroups, metadata and traceability groups. Among them, the traceability group binds the preset correction detection model and the processing logs generated by the preset correction detection model (including rule identifiers, historical battery model identifiers, transformation parameters, correction strengths, etc.) in the above processing with the corresponding training data, so that the generation source of each piece of data in the training dataset can be accurately traced and audited.

[0116] To further understand the technical solution provided in this application, the following uses lithium iron phosphate (LFP) prismatic batteries as the historical battery model, ternary cylindrical batteries (Lithium Nickel Cobalt Manganese Oxide NCM622 (Ni:Co:Mn=6:2:2)) as the target battery model, and the battery state prediction model as a thermal runaway early warning model. The training dataset construction method provided in this application is illustrated by combining the above multiple embodiments.

[0117] In some embodiments, the lithium iron phosphate prismatic battery has a cell capacity of 100Ah, a cell weight of 1.5kg, and a specific heat capacity of 1.20J / (g•K), with a corresponding historical operating data sampling frequency of 10Hz. The ternary cylindrical battery has a cell capacity of 5Ah, a cell weight of 0.07kg, and a specific heat capacity of 1.05J / (g•K), with a target sampling frequency of 10Hz.

[0118] In this example, the target battery parameters can be determined based on the battery-related information of the ternary cylindrical battery. Further, based on the historical battery parameters of the lithium iron phosphate prismatic battery, a migration rule set between the ternary cylindrical battery and the lithium iron phosphate prismatic battery can be constructed. For the thermal runaway early warning model, the migration rule set can include, for example: an increase in internal resistance of approximately 30% (the internal resistance of the ternary cylindrical battery cell is higher), an increase in thermal conductivity of approximately 40% (the heat dissipation path of the cylindrical structure of the ternary cylindrical battery is shorter), a decrease in the thermal runaway trigger temperature (e.g., from 160℃ to 130℃), and adjustment of the voltage plateau, etc. A specific example of the migration rule set can be found below: { "rule_id":"R010_LFP_to_NCM622_Migration", / / Rule identifier: The set of migration rules for migrating from lithium iron phosphate prismatic batteries to ternary cylindrical batteries; "source_bms_model":"LFP_100Ah_Prismatic", / / Historical battery management system model: Lithium iron phosphate, 100Ah charge capacity, prismatic structure; "target_bms_model":"NCM622_5Ah_Cylindrical", / / Target battery management system model: ternary cylindrical battery, charge capacity 5Ah, cylindrical structure; "physical_transformations": { / / Rules for physical feature transformations; "temperature": { / / Temperature, a physical quantity;} "operator":"piecewise_nonlinear", / / Operator: piecewise nonlinear transformation, using different transformation rules in different temperature ranges; "params": { / / Transformation parameters; "segments": [ / / Segment configuration, defining the transformation rules for different temperature ranges; {"range": [0, 80], / / Temperature range: 0℃ to 80℃; "type":"linear", / / Transformation type: linear transformation, using a constant scaling ratio within the range of 0℃ to 80℃; "scale": 0.95}, / / Scaling factor: Within the temperature range of 0℃ to 80℃, the temperature rise of the target battery model is 0.95 times that of the historical battery models; {"range": [80, 160], / / Temperature range: 80℃ to 160℃; "type":"exponential", / / Transformation type: Exponential transformation, using an exponential function to transform within the range of 80℃ to 160℃; "base": 1.02, / / Exponential base: The rate of temperature rise of the target battery model in the range of 80℃ to 160℃ increases exponentially with increasing temperature, with a base of 1.02; "offset": -30 / / Exponential offset: The exponential curve of the target battery model in the 80℃ to 160℃ range is offset by -30℃ along the vertical axis; } } }, "voltage": { / / Voltage, a physical quantity;} "operator":"scale_drop_rate", / / Operator: scales the rate of voltage drop; "params": { / / Transformation parameters; "drop_rate_scale_factor": 1.2, / / Voltage drop rate scaling factor: The voltage drop rate of the target battery model is 1.2 times that of the historical battery models; "platform_shift_v": 0.15 / / Voltage platform offset: The operating voltage platform of the target battery model is 0.15V higher than that of the historical battery model; } }, "soc": { / / State of charge; "operator":"remap_ocv", / / Operator: Remaps the open-circuit voltage curve; "params": { / / Transformation parameters; "ocv_curve_id":"NCM622_OCV_v1.0" / / Calculate the state of charge using the open-circuit voltage curve table of version 1.0 corresponding to the target battery model; } } }, "control_logic_override": { / / Control logic override, i.e., control strategy transformation rules; "thermal_alarm_threshold_celsius": 130.0, / / Thermal alarm threshold: The overheat protection trigger temperature for the target battery model is set to 130℃; "debounce_frames": 5, / / Debounce frame count: Fault signals must be confirmed for 5 consecutive frames before an alarm is triggered; "control_period_hz": 10 / / Control cycle: The main loop frequency for determining the state of the target battery management system is 10Hz; } } Then, based on the migration rule set, the historical operating data physical quantities and message timestamps of the lithium iron phosphate prismatic battery can be migrated. In some embodiments, physical rule verification can be performed before migration processing to eliminate migration rules that violate physical constraints. Specifically, during message timestamp migration in the migration process, the control cycle of the ternary cylindrical battery is 10Hz, which is equal to the sampling frequency of the lithium iron phosphate prismatic battery. Therefore, during timestamp migration, it is not necessary to sample and extract historical operating data. The fault time and de-jitter timestamp can be determined frame by frame. For example, the temperature can be checked frame by frame at a frequency of 10Hz to see if it exceeds the 130°C threshold. After de-jittering 5 frames, an alarm is triggered.

[0119] Furthermore, residual calculations and corrections can be performed on the migration operation data. In some embodiments, energy conservation checks can be performed on the migration operation data before correction, thereby accurately removing post-migration data that violates energy conservation.

[0120] Specifically, the expected energy deviation threshold corresponding to the ternary cylindrical battery can be calculated using formula (3) in the above embodiment. Because the heat capacity of ternary cylindrical batteries is much smaller than that of lithium iron phosphate prismatic batteries, the corresponding energy deviation threshold is significantly tightened, reflecting the physical reality that small heat capacity systems are more sensitive to energy deviation. Although the determined energy deviation threshold is extremely stringent, the high-precision migration in the preceding step and the nonlinear interpolation of the subsequent reverse mapping can effectively ensure the high fidelity of the verified data.

[0121] In the residual correction process of this example, the importance weight of each feature dimension in the migration run data can be determined by performing weight analysis on the data of each feature dimension through the preset correction detection model for the large language model. For example, the preset correction detection model can detect that the plateau period feature before thermal runaway of lithium iron phosphate prismatic batteries does not exist in ternary cylindrical batteries, so it is recommended to set the importance weight of the temperature rise acceleration dimension to 0.95. If the sample size of the target run data is K=3, the confidence level of the target run data calculated by the above formula (3) is only 3 / 8=0.375, so that the correction intensity of this feature dimension can be effectively suppressed.

[0122] Furthermore, a training dataset for training the thermal runaway prediction model can be constructed based on the corrected operating data and a small amount of target operating data corresponding to the ternary cylindrical battery. Simultaneously, in some embodiments, a source tracing group or source tracing log corresponding to the training dataset can be set up to record the migration process corresponding to different migration data.

[0123] The above describes the specific implementation of the training dataset construction method provided in this application. The technical solution provided in this application utilizes existing historical operating data of historical battery models and a small amount of target operating data of the target battery model for precise transfer and combination, constructing a high-quality training dataset that can be used to train a battery state prediction model for new battery models. This effectively solves the problem that new batteries cannot train effective models due to a lack of sufficient operating data. In this application, through dual transfer of physical quantities and message timestamps, as well as residual correction, the constructed training dataset not only matches the target battery model in terms of physical characteristics, but also matches the message timestamps and decision specifications at the control logic level with the battery management system corresponding to the target battery model. The model trained based on the dataset constructed using the technical solution of this application can accurately predict the battery state after actual deployment, effectively reducing the problems of missed or false alarms, significantly improving model training efficiency and training effect, and providing a practical and reliable battery state prediction model for new battery models.

[0124] Based on the same technical concept, this application also provides a training dataset construction apparatus, such as... Figure 2 As shown.

[0125] Figure 2 This is a schematic diagram of the structure of a training dataset construction device provided in some embodiments of this application.

[0126] like Figure 2 As shown in the illustration, this application also provides a training dataset construction apparatus 200, applied to an electronic device. The training dataset construction apparatus 200 includes: The parameter determination module 201 is used to determine the target battery parameters corresponding to the target battery model based on battery-related information of the target battery model. The rule construction module 202 is used to determine multiple historical battery models that match the target battery model based on the target battery parameters, and to construct a migration rule set between the target battery model and the multiple historical battery models. The migration rule set includes physical feature transformation rules and control strategy transformation rules. Data migration module 203 is used to perform physical quantity migration and message timestamp migration on historical operating data of multiple historical battery models according to the migration rule set to obtain migration operating data; The residual correction module 204 is used to perform residual analysis on the migration operation data and the target operation data of the target battery model, and to perform residual correction on the migration operation data based on the residual analysis results to obtain corrected operation data. The dataset construction module 205 is used to construct a training dataset based on the corrected running data and the target running data. The training dataset is used to train the battery state prediction model in the battery management system corresponding to the target battery model.

[0127] In some embodiments, the target battery parameters include physical characteristic parameters and control strategy parameters corresponding to the target battery model; The rule construction module 202 mentioned above is used for: For each historical battery model, the physical characteristic transformation parameters corresponding to the historical battery model are determined based on the physical characteristic parameters and the historical physical characteristic parameters corresponding to the historical battery model. The control strategy transformation parameters corresponding to the historical battery model are determined based on the control strategy parameters and the historical control strategy parameters corresponding to the historical battery model. Based on the physical characteristic transformation parameters, at least one physical characteristic transformation rule corresponding to the historical battery model is constructed, and based on the control strategy transformation parameters, at least one control strategy transformation rule corresponding to the historical battery model is constructed. Based on at least one physical characteristic transformation rule and at least one control strategy transformation rule, a migration rule set between the target battery model and the historical battery model is constructed.

[0128] In some embodiments, the data migration module 203 described above is used for: Based on the transformation type corresponding to at least one physical feature transformation rule, physical quantity migration is performed on the physical quantity data corresponding to at least one physical feature transformation rule in the historical operation data to obtain physical quantity migration data. The sampling frequency of the physical quantity migration data is consistent with that of the historical operation data. The transformation type includes linear transformation and nonlinear transformation. Based on at least one control strategy transformation rule, determine the timestamp migration parameters corresponding to the historical operation data, and perform message timestamp migration on the physical quantity migration data according to the timestamp migration parameters to obtain the migration operation data.

[0129] In some embodiments, the timestamp migration parameters include at least the parameter fault threshold corresponding to the target battery model, the target sampling frequency, and the number of dejittering attempts; The aforementioned data migration module 203 is used for: When the target sampling frequency is less than the sampling frequency, data is extracted from the physical quantity migration data based on the target sampling frequency to obtain the sampling operation data corresponding to the target sampling frequency. Based on the parameter fault threshold, fault moment detection is performed on the sampled operation data; If there is a fault in the sampled running data, and the number of consecutive fault times is greater than or equal to the number of de-jittering times, the fault time with the number of consecutive fault times equal to the number of de-jittering times will be used as the fault message timestamp corresponding to the target battery model. Based on the fault message timestamp, the sampled running data is timestamped to obtain the migration running data.

[0130] In some embodiments, the residual correction module 204 described above is used for: Based on multiple preset feature dimensions, determine the migration feature vector corresponding to the migration running data in each feature dimension, and determine the target feature vector corresponding to the target running data in each feature dimension; Residual analysis is performed on the migration feature vector and target feature vector for each feature dimension to determine the residual vector between the migration running data and the target running data, which is used as the residual analysis result; By using a pre-defined correction detection model, and based on the residual analysis results and migration rule set, the correction intensity corresponding to each feature dimension of the migration operation data is determined. Based on the correction intensity, residual correction is performed on the migration operation data to obtain corrected operation data.

[0131] In some embodiments, the residual correction module 204 described above is used for: Based on the residual vector, calculate the mean difference between the migration run data and the target run data in each feature dimension, and determine the data dispersion of the migration run data and the target run data in each feature dimension. By using a pre-defined correction detection model, based on the mean difference and data dispersion, the importance of the rules corresponding to multiple feature dimensions in the migration rule set is analyzed to determine the importance weight of each feature dimension. For each feature dimension, the adjustment coefficient corresponding to the importance weight is calculated based on the data dispersion of the migration run data and the target run data in the feature dimension, and the number of samples in the target run data. The product of the adjustment coefficient and the importance weight is calculated as the correction strength of the migration run data in the feature dimension. The upper limit of the correction strength is the importance weight multiplied by a preset multiple.

[0132] In some embodiments, the rule construction module 202 described above is further used for: Calculate the average parameter values ​​for multiple historical battery models based on historical operating data; Based on the migration rule set and average parameter values, the predicted parameter values ​​corresponding to the target battery model are determined, and based on the target operating data, the actual parameter values ​​of the target battery model are determined. Based on the actual and predicted parameter values, the migration rule set is subjected to physical constraint verification. Rules that violate physical constraints are removed from the migration rule set to obtain the verified rule set. The rule construction module 202 mentioned above is used for: Based on the verified rule set, physical quantity migration and message timestamp migration are performed on the historical operation data to obtain the migrated operation data.

[0133] In some embodiments, the data migration module 203 is further configured to: Based on the target battery parameters and the historical battery parameters corresponding to multiple historical battery models, calculate the expected energy deviation threshold corresponding to the target battery model. Determine the output energy and input energy corresponding to multiple migrated data in the migration operation data, and calculate the energy deviation corresponding to multiple migrated data based on the output energy and input energy; After removing the migration data whose energy deviation exceeds the expected energy deviation threshold, the verified operation data is obtained. The aforementioned data migration module 203 is used for: Residual analysis was performed on the verified operating data and the target operating data of the target battery model.

[0134] In some embodiments, the parameter determination module 201 described above is used for: By using a preset parameter analysis model, the battery-related information is analyzed based on the parameter parsing template corresponding to the battery state prediction model to determine the target battery parameters.

[0135] In some embodiments, the rule construction module 202 described above is used for: Determine the parameter difference between the target battery parameters and the historical battery parameters corresponding to each candidate historical battery model in the historical parameter library; Based on parameter differences and preset difference thresholds, multiple historical battery models are selected from multiple candidate historical battery models.

[0136] Figure 3 This is a schematic diagram of the structure of a terminal device provided in some embodiments of this application.

[0137] The terminal device may include a processor 301 and a memory 302 storing computer program instructions.

[0138] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0139] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to an end device. In a particular embodiment, memory 302 is a non-volatile solid-state memory.

[0140] In a particular embodiment, memory 302 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Thus, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to any of the training dataset construction methods disclosed in this application.

[0141] The processor 301 reads and executes computer program instructions stored in the memory 302 to implement any of the training dataset construction methods in the above embodiments.

[0142] In one example, the terminal device may also include a communication interface 303 and a bus 310. Wherein, for example... Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 310 and complete communication with each other.

[0143] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0144] Bus 310 includes hardware, software, or both, that couples components of an end device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 310 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0145] Furthermore, in conjunction with the data processing methods in the above embodiments, this application embodiment can provide a computer storage medium for implementation. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the training dataset construction methods in the above embodiments.

[0146] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the training dataset construction methods described in the above embodiments.

[0147] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0148] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0149] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0150] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in 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, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0151] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for constructing a training dataset, characterized in that, include: Based on battery-related information of the target battery model, determine the target battery parameters corresponding to the target battery model; Based on the target battery parameters, multiple historical battery models that match the target battery model are determined, and a migration rule set between the target battery model and the multiple historical battery models is constructed. The migration rule set includes physical feature transformation rules and control strategy transformation rules. Based on the migration rule set, physical quantity migration and message timestamp migration are performed on the historical operating data of the multiple historical battery models to obtain migration operating data; Residual analysis is performed on the migration operation data and the target operation data of the target battery model, and residual correction is performed on the migration operation data based on the residual analysis results to obtain corrected operation data; Based on the corrected operating data and the target operating data, a training dataset is constructed. The training dataset is used to train the battery state prediction model in the battery management system corresponding to the target battery model.

2. The method according to claim 1, characterized in that, The target battery parameters include physical characteristic parameters and control strategy parameters corresponding to the target battery model; Constructing a migration rule set between the target battery model and multiple historical battery models, including: For each historical battery model, the physical characteristic transformation parameters corresponding to the historical battery model are determined based on the physical characteristic parameters and the historical physical characteristic parameters corresponding to the historical battery model, and the control strategy transformation parameters corresponding to the historical battery model are determined based on the control strategy parameters and the historical control strategy parameters corresponding to the historical battery model. Based on the physical feature transformation parameters, at least one physical feature transformation rule corresponding to the historical battery model is constructed, and based on the control strategy transformation parameters, at least one control strategy transformation rule corresponding to the historical battery model is constructed. Based on at least one of the physical feature transformation rules and at least one of the control strategy transformation rules, a migration rule set between the target battery model and the historical battery model is constructed.

3. The method according to claim 2, characterized in that, Based on the migration rule set, physical quantity migration and message timestamp migration are performed on the historical operating data of the multiple historical battery models to obtain migration operating data, including: According to the transformation type corresponding to the at least one physical feature transformation rule, physical quantity migration is performed on the physical quantity data corresponding to the at least one physical feature transformation rule in the historical operation data to obtain physical quantity migration data. The physical quantity migration data has the same sampling frequency as the historical operation data. The transformation type includes linear transformation and nonlinear transformation. Based on the at least one control strategy transformation rule, the timestamp migration parameters corresponding to the historical operation data are determined, and the physical quantity migration data is subjected to message timestamp migration according to the timestamp migration parameters to obtain the migration operation data.

4. The method according to claim 3, characterized in that, The timestamp migration parameters include at least the parameter fault threshold, target sampling frequency, and number of de-jitter counts corresponding to the target battery model; Based on the timestamp migration parameters, the physical quantity migration data is subjected to message timestamp migration to obtain the migration operation data, including: When the target sampling frequency is less than the sampling frequency, data is extracted from the physical quantity migration data based on the target sampling frequency to obtain the sampling operation data corresponding to the target sampling frequency; Based on the aforementioned fault threshold parameters, fault moment detection is performed on the sampled operational data. If the sampling operation data contains the fault time, and the number of consecutive fault times is greater than or equal to the number of de-jittering times, the fault time with the number of consecutive fault times equal to the number of de-jittering times shall be used as the fault message timestamp corresponding to the target battery model. Based on the timestamp of the fault message, the timestamp of the sampled running data is updated to obtain the migration running data.

5. The method according to claim 1, characterized in that, Residual analysis is performed on the migration operation data and the target operation data of the target battery model. Based on the residual analysis results, residual correction is performed on the migration operation data to obtain corrected operation data, including: Based on multiple preset feature dimensions, the migration feature vector corresponding to the migration operation data in each feature dimension is determined, and the target feature vector corresponding to the target operation data in each feature dimension is determined. Perform residual analysis on the migration feature vector and target feature vector for each feature dimension to determine the residual vector between the migration running data and the target running data, which is used as the residual analysis result; By using a pre-defined correction detection model, based on the residual analysis results and the migration rule set, the correction intensity corresponding to each feature dimension of the migration operation data is determined; Based on the correction intensity, residual correction is performed on the migration operation data to obtain the corrected operation data.

6. The method according to claim 5, characterized in that, Based on the residual analysis results and the migration rule set, the correction strength corresponding to each feature dimension of the migration operation data is determined, including: Based on the residual vector, calculate the mean difference between the migration running data and the target running data for each feature dimension, and determine the data dispersion of the migration running data and the target running data for each feature dimension. By using a pre-defined correction detection model, based on the mean difference and the data dispersion, an importance analysis is performed on the rules corresponding to the multiple feature dimensions in the migration rule set, and the importance weight corresponding to each feature dimension is determined. For each of the aforementioned feature dimensions, an adjustment coefficient corresponding to the importance weight is calculated based on the data dispersion of the migration operation data and the target operation data corresponding to the respective feature dimension, and the number of samples of the target operation data. The product of the adjustment coefficient and the importance weight is calculated as the correction intensity corresponding to the feature dimension of the migration running data, and the upper limit of the correction intensity is the importance weight by a preset multiple.

7. The method according to claim 1, characterized in that, Before performing physical quantity migration and message timestamp migration on the historical operating data of the multiple historical battery models according to the migration rule set to obtain the migrated operating data, the method further includes: Based on the historical operating data, calculate the average parameter values ​​of the multiple historical battery models; Based on the migration rule set and the average parameter value, the predicted parameter value corresponding to the target battery model is determined, and based on the target operating data, the actual parameter value of the target battery model is determined. Based on the actual parameter values ​​and the predicted parameter values, the migration rule set is subjected to physical constraint verification, and rules that violate physical constraints are removed from the migration rule set to obtain the verified rule set. Based on the migration rule set, physical quantity migration and message timestamp migration are performed on the historical operating data of the multiple historical battery models to obtain migration operating data, including: Based on the verified rule set, the historical running data is subjected to physical quantity migration and message timestamp migration to obtain the migrated running data.

8. The method according to claim 1, characterized in that, Before performing residual analysis on the migration operation data and the target operation data of the target battery model, the method further includes: Based on the target battery parameters and the historical battery parameters corresponding to the multiple historical battery models, calculate the expected energy deviation threshold corresponding to the target battery model; Determine the output energy and input energy corresponding to multiple migrated data in the migration operation data, and calculate the energy deviation corresponding to the multiple migrated data based on the output energy and the input energy; After removing the post-migration data whose energy deviation is greater than the expected energy deviation threshold from the migration operation data, the verified operation data is obtained. Residual analysis is performed on the migration operation data and the target operation data of the target battery model, including: Residual analysis is performed on the verified operating data and the target operating data of the target battery model.

9. The method according to claim 1, characterized in that, Based on battery-related information of the target battery model, determine the target battery parameters corresponding to the target battery model, including: By using a preset parameter analysis model, the battery-related information is analyzed based on the parameter parsing template corresponding to the battery state prediction model to determine the target battery parameters.

10. The method according to claim 1, characterized in that, Based on the target battery parameters, multiple historical battery models matching the target battery model are determined, including: Determine the parameter difference between the target battery parameters and the historical battery parameters corresponding to each candidate historical battery model in the historical parameter library; Based on the parameter difference degree and the preset difference degree threshold, the multiple historical battery models are selected from the multiple candidate historical battery models.

11. A training dataset construction apparatus, characterized in that, include: The parameter determination module is used to determine the target battery parameters corresponding to the target battery model based on battery-related information of the target battery model. The rule building module is used to determine multiple historical battery models that match the target battery model based on the target battery parameters, and to build a migration rule set between the target battery model and the multiple historical battery models. The migration rule set includes physical feature transformation rules and control strategy transformation rules. The data migration module is used to perform physical quantity migration and message timestamp migration on the historical operating data of the multiple historical battery models according to the migration rule set, so as to obtain the migrated operating data; The residual correction module is used to perform residual analysis on the migration operation data and the target operation data of the target battery model, and to perform residual correction on the migration operation data based on the residual analysis results to obtain corrected operation data. The dataset construction module is used to construct a training dataset based on the corrected running data and the target running data. The training dataset is used to train the battery state prediction model in the battery management system corresponding to the target battery model.

12. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the training dataset construction method as described in any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the training dataset construction method as described in any one of claims 1-10.

14. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device causes the electronic device to perform the training dataset construction method as described in any one of claims 1-10.

Images