Battery voltage difference detection method, device, equipment, medium and product

Abstract

The invention discloses a battery voltage difference detection method, device and equipment, a medium and a product. The method comprises the following steps: acquiring current charging data of a to-be-detected battery pack, and determining a multi-cell voltage list at at least two charging process endpoint moments from the current charging data; performing feature extraction on the at least two multi-cell voltage lists to obtain voltage feature vectors; inputting the at least two multi-cell voltage lists and the voltage feature vectors into a trained large language model, and outputting an initial voltage difference detection result; and inputting the initial voltage difference detection result and the voltage feature vector into an error correction model, and outputting a target voltage difference detection result corresponding to the to-be-detected battery. According to the technical scheme provided by the invention, the effect of reducing the misjudgment rate is achieved while the detection precision of the complex voltage difference fault is improved and the classification accuracy of the battery fault in a complex scene is improved.

Description

Technical Field

[0001] This invention relates to the field of computer processing technology, and in particular to a battery differential pressure detection method, apparatus, equipment, medium, and product. Background Technology

[0002] With the widespread application of IoT technology in battery management systems (BMS), massive amounts of battery operation data can be uploaded to the cloud in real time, providing a data foundation for the condition monitoring and fault diagnosis of large-scale battery systems. The voltage difference between individual battery cells is a key indicator reflecting the consistency of the battery pack and potential faults. Abnormal changes in voltage often indicate safety hazards such as loose connections, sampling errors, and uneven aging, requiring efficient and accurate diagnostic methods.

[0003] In existing technologies, battery differential pressure detection mainly relies on two types of methods:

[0004] One approach is an expert rule system based on fixed thresholds, which makes judgments by setting thresholds for differential pressure or standard deviation. This method uses static, fixed rules, making it difficult to adapt to changes in operating conditions and battery aging, resulting in high false alarm and false negative rates.

[0005] Second, traditional machine learning models such as support vector machines and decision trees are used. This method relies on manual feature extraction, which makes it difficult to capture the deep dynamic features in voltage time series and the complex correlation between cells. It also has insufficient accuracy in identifying battery voltage difference faults in large-scale, multi-scenario applications and poor generalization ability. Summary of the Invention

[0006] This invention provides a battery differential pressure detection method, device, equipment, medium, and product to improve the detection accuracy of complex differential pressure faults, enhance the accuracy of battery fault classification in complex scenarios, and reduce the false judgment rate.

[0007] According to one aspect of the present invention, a battery differential pressure detection method is provided, the method comprising:

[0008] Obtain the current charging data of the battery pack to be tested, and determine a list of multiple cell voltages at at least two charging process endpoints from the current charging data;

[0009] Feature extraction is performed on at least two of the multi-unit voltage lists to obtain voltage feature vectors;

[0010] Input at least two of the multi-unit voltage lists and the voltage feature vectors into the trained large language model, and output the initial differential pressure detection result;

[0011] The initial differential pressure detection result and the voltage feature vector are input into the error correction model, and the target differential pressure detection result corresponding to the battery to be detected is output.

[0012] According to another aspect of the present invention, a battery differential pressure detection device is provided, the device comprising:

[0013] A multi-cell voltage list determination module is used to acquire the current charging data of the battery pack to be tested, and determine the multi-cell voltage list at at least two charging process endpoints from the current charging data.

[0014] A voltage feature vector determination module is used to extract features from at least two of the multi-unit voltage lists to obtain voltage feature vectors.

[0015] The initial differential pressure detection result determination module is used to input at least two of the multi-unit voltage lists and the voltage feature vectors into the trained large language model and output the initial differential pressure detection result;

[0016] The target differential pressure detection result determination module is used to input the initial differential pressure detection result and the voltage feature vector into the error correction model, and output the target differential pressure detection result corresponding to the battery to be tested.

[0017] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0018] At least one processor; and a memory communicatively connected to said at least one processor; wherein,

[0019] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the battery differential pressure detection method according to any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the battery differential pressure detection method according to any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the battery differential pressure detection method as described in any embodiment of the present invention.

[0022] The technical solution of this invention obtains the current charging data of the battery pack to be tested and determines a multi-cell voltage list at at least two charging process endpoints from the current charging data; performs feature extraction on the at least two multi-cell voltage lists to obtain voltage feature vectors; inputs the at least two multi-cell voltage lists and voltage feature vectors into a trained large language model to output an initial differential pressure detection result; inputs the initial differential pressure detection result and voltage feature vectors into an error correction model to output a target differential pressure detection result corresponding to the battery to be tested. This solves the problem of low accuracy in identifying battery differential pressure faults and the tendency for false alarms and missed alarms in the prior art based on expert rules with fixed thresholds and traditional machine learning models for differential pressure detection. By acquiring a list of multi-cell voltages of the battery pack under test at at least two charging process endpoints, and performing deep feature extraction on the multi-cell voltage list to generate physically meaningful voltage feature vectors, the multi-cell voltage list and the high-order voltage feature vectors are jointly input into a trained large language model. This model outputs an initial differential pressure detection result based on understanding the evolution of the charging process. Then, an error correction model is used to fuse the voltage feature vectors to perform logical consistency verification and correction on the initial differential pressure detection result, finally outputting a highly reliable target differential pressure detection result. This improves the detection accuracy of differential pressure faults in the battery pack, enhances the accuracy and robustness of differential pressure fault identification, effectively reduces false alarms and false negatives in differential pressure detection, and enhances the safety and reliability of battery system operation.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a battery differential pressure detection method provided according to an embodiment of the present invention;

[0026] Figure 2 This is a flowchart of a battery differential pressure detection method provided according to an embodiment of the present invention;

[0027] Figure 3 This is a flowchart of a battery differential pressure detection method provided according to an embodiment of the present invention;

[0028] Figure 4This is a schematic diagram of a battery differential pressure detection device according to an embodiment of the present invention;

[0029] Figure 5 This is a schematic diagram of the structure of an electronic device that implements the battery differential pressure detection method of this invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] It should be noted that the collection, gathering, updating, analysis, processing, use, transmission, and storage of user personal information involved in this disclosed technical solution all comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken to prevent unauthorized access to user personal information data and to safeguard user personal information security and network security.

[0033] It should also be noted that in the technical solutions disclosed herein, the collection, updating, analysis, processing, use, transmission, and storage of users' personal information are all carried out with the user's knowledge and permission, and comply with relevant privacy protection regulations.

[0034] Before introducing this technical solution, we can first introduce the application scenarios. The technical solution provided by the embodiments of the present invention can be applied to any scenario where battery differential pressure detection of battery pack is required.

[0035] Currently, in applications such as new energy vehicles, energy storage power stations, electric ships, and battery swapping facilities, the scale of battery systems is continuously expanding, and their operating conditions are becoming increasingly complex, leading to ever-increasing demands on battery safety and reliability. Against this backdrop, the voltage difference between individual battery cells (referred to as "voltage differential") serves as a crucial indicator reflecting battery pack consistency, state of health (SOH), and potential faults. Abnormal changes in voltage differential often foreshadow performance degradation or safety hazards in the battery system, such as loose connections, abnormal sampling, uneven cell aging, and precursors to thermal runaway. Therefore, accurate and efficient fault diagnosis of battery voltage differential has become a key aspect of ensuring the safe and stable operation of battery systems.

[0036] However, with the continuous growth in the deployment scale of battery systems and the diversification and dynamism of application scenarios (such as frequent start-stop of urban buses, continuous high load of long-distance logistics, and multi-mode charging and discharging of energy storage systems), the same model of battery system exhibits significantly different differential pressure evolution behaviors under different usage environments and operating conditions. For example, "voltage stratification" may occur in low-temperature environments, while complex fault modes such as "multi-point voltage drops" are prone to occur when discharging at high rates or when local heat dissipation is poor. These fault characteristics have blurred boundaries, are highly time-varying, and are coupled with multiple factors such as battery aging degree, historical usage trajectory, and ambient temperature. This makes existing differential pressure detection methods based on fixed threshold expert rules and traditional machine learning models insufficient in identifying battery differential pressure faults on a large scale and in multiple scenarios, easily leading to false alarms and missed alarms.

[0037] To address the aforementioned issues, battery differential pressure detection can be performed based on the technical solution provided in this embodiment, enabling automatic identification and accurate classification of various differential pressure fault types, thereby improving the intelligence level and safety assurance capabilities of battery system operation and maintenance.

[0038] Figure 1 This is a flowchart of a battery differential pressure detection method according to an embodiment of the present invention. This embodiment is applicable to the situation of battery differential pressure detection of a battery pack. The method can be executed by a battery differential pressure detection device, which can be implemented in hardware and / or software, and can be configured in a computing device. Figure 1 As shown, the method includes:

[0039] S110. Obtain the current charging data of the battery pack to be tested, and determine a list of multiple cell voltages at at least two charging process endpoints from the current charging data.

[0040] The battery pack under test can refer to a battery cell consisting of multiple individual cells connected in series or parallel, requiring differential voltage fault diagnosis. Current charging data refers to the set of operating parameters continuously or periodically collected by the BMS during the charging process of the battery pack under test. This current charging data may include information such as collection time, voltage, current, individual cell voltage, and battery temperature. The charging process endpoints refer to two time points related to the charging cycle, used to characterize significant changes in battery state. The multi-cell voltage list refers to an ordered sequence of values ​​formed by arranging the voltage measurements of all individual cells in the battery pack under test in a fixed order at the charging process endpoint. Each multi-cell voltage list reflects the voltage distribution state within the battery pack at that charging process endpoint.

[0041] In this embodiment, the current charging data of the battery pack under test can be retrieved from a cloud database or data platform based on its unique identifier (such as VIN code or device ID). Alternatively, the current charging data of all or part of the battery packs can be obtained from the cloud database or data platform, and these battery packs can be used as the battery packs to be tested. Furthermore, the voltage data of all individual cells at the end points of the charging process can be filtered from the current charging data of each battery pack to obtain a multi-cell voltage list corresponding to each end point of the charging process.

[0042] At each of the two charging endpoints, the voltage values ​​of all individual cells are extracted and arranged according to their physical or logical cell numbering, forming two independent multi-cell voltage lists. These voltage lists will then serve as input features for subsequent differential pressure fault identification and classification, characterizing the differential pressure evolution behavior of the battery pack at different stages of charging and ensuring the accuracy of differential pressure detection.

[0043] To ensure the accuracy of battery differential voltage detection, during the process of determining the list of multiple cell voltages at at least two charging process endpoints from the current charging data, the current charging data can be discarded based on preset invalid charging conditions to obtain valid charging data; the valid charging data can then be extracted to obtain the list of multiple cell voltages at at least two charging process endpoints.

[0044] The preset invalid charging conditions refer to a predefined set of rules or standards used to judge the low quality or lack of analytical value of data during a certain charging process. Valid charging data refers to the charging data retained after filtering for invalid charging conditions. This type of valid charging data represents a complete, stable process that conforms to normal charging characteristics. At least two endpoints of the charging process are included: the end of the resting period before charging begins, and the end of charging. The end of the resting period before charging begins can be the last valid data acquisition moment after the resting (no charging / discharging current) period before the battery pack officially begins charging. At the end of the resting period before charging begins, the battery voltage is in a relatively stable state, reflecting the true state before charging. The voltage difference between cells mainly reflects static factors such as self-discharge and uneven aging, and is the node for analyzing the initial voltage difference state. The end of charging can be the last valid data acquisition moment when the charging process terminates. At this time, the voltage is significantly affected by polarization, temperature rise, and consistency differences, making it a stage where voltage differences are easily apparent.

[0045] In this embodiment, the acquired current charging data can be filtered and eliminated based on preset invalid charging conditions. For example, invalid charging conditions may include: charging time too short (e.g., less than 10 minutes, considered an abnormal start / stop), charging depth too small (e.g., SOC increment below a threshold, insufficient information), prolonged interruption or drastic current fluctuations during charging, insufficient resting time (unable to reflect the true initial state), or a large number of missing or outlier data. By verifying each charging segment in the current charging data against these invalid charging conditions, invalid charging data meeting the preset conditions can be eliminated, retaining only valid charging data that meets the analysis requirements. Furthermore, a multi-unit voltage list can be extracted from the filtered valid charging data according to preset charging process endpoint times. Specifically, the starting boundary of the charging process for each charging segment in the valid charging data can be identified, and a current-free interval satisfying a minimum resting time (e.g., 5 minutes or more) can be traced back. The last valid data point of this resting phase is taken as the end time of the resting phase before charging begins, and the voltage values ​​of all individual cells at this end time of the resting phase before charging begins are extracted to form a voltage list. The termination point of the charging process can be located, that is, the moment when the charging current decays to the cutoff threshold and charging stops. The voltage values ​​of all individual cells corresponding to this charging end moment are taken to form another voltage list. By extracting the voltage distribution at these two moments with clear physical significance, the voltage difference evolution process of the battery pack from a resting steady state to a fully charged state can be effectively captured, ensuring the accuracy of battery voltage difference detection.

[0046] In this embodiment, the current charging data includes multiple charging segments, and the preset invalid charging conditions include at least one of the following: the idle time interval between two charging segments reaches a preset duration; the initial charging amount of the charging segment is greater than a preset first amount; and the final charging amount of the charging segment is less than a preset second amount.

[0047] The idle time interval between two charging segments reaching a preset duration can be understood as a long period of no charging activity (i.e., idle) between two consecutive charging processes in the continuous battery operation data stream. If this idle time interval exceeds the preset duration (e.g., 2 hours), it indicates that the consecutive charging sessions belong to independent usage cycles. This condition can be used to identify isolated, discontinuous charging events and avoid incorrectly associating data from different vehicle or energy usage cycles.

[0048] The fact that the initial charging capacity of a charging segment is greater than the preset first capacity can be understood as follows: at the start of charging, the battery's state of charge (SOC) is already higher than the set preset first capacity (e.g., 80%). This situation indicates that the charging is not a normal energy replenishment process that starts from a lower capacity, but may be a brief replenishment or abnormal start-up in the middle. Its initial state is unstable and is not suitable as the starting point for analyzing the evolution of the voltage difference.

[0049] When the charging segment ends with a charge level less than the preset second charge level, it can be understood that the battery's SOC (State of Charge) did not reach the set second charge threshold (e.g., 95%) at the end of the charging process. This indicates that the charging process was not completed or terminated prematurely, failing to enter the constant voltage stage or fully reflect the voltage difference characteristics under full charge conditions. Therefore, the data is insufficient to fully characterize the voltage difference behavior at the end of the charging process.

[0050] In other words, when processing the current charging data, it can be divided into several independent charging segments based on time and current changes, with each segment representing a complete charging attempt. For each charging segment, it can be checked whether it meets at least one of the aforementioned invalid conditions.

[0051] Specifically, by analyzing the time series between adjacent charging segments, the difference between the start time of the next segment and the end time of the previous segment can be calculated. If this idle time interval is greater than a preset duration, the two segments are determined to belong to different usage cycles, are evaluated independently, and marked as invalid. For the charging segment currently being evaluated, the battery SOC (i.e., the initial charging capacity) at its start time is extracted. If the initial charging capacity is higher than a preset first capacity, it is marked as invalid. Similarly, the battery SOC (i.e., the end charging capacity) at its end time can be extracted. If the end charging capacity is lower than a preset second capacity, it is also marked as invalid.

[0052] When a charging segment does not meet all the above preset invalid charging conditions, it can be considered valid charging data and used for subsequent feature extraction and differential pressure detection analysis. This filtering operation can effectively eliminate differential pressure detection interference caused by atypical user habits or data fragmentation, ensuring the accuracy of differential pressure detection.

[0053] S120. Extract features from at least two multi-unit voltage lists to obtain voltage feature vectors.

[0054] The voltage feature vector can be a feature vector used to characterize the voltage differential pattern, generated by mathematical transformation and statistical analysis of the voltage list. For example, the voltage feature vector may include the voltage mean, standard deviation, range (the difference between the maximum and minimum values), skewness, kurtosis, voltage standard deviation, maximum increase, decrease magnitude, voltage increase rate, etc.

[0055] In this embodiment, static statistical characteristics can be calculated based on each multi-cell voltage list. For example, static statistical characteristics such as voltage mean, standard deviation, range (difference between maximum and minimum values), skewness, and kurtosis can be calculated to describe the dispersion and distribution characteristics of the cell voltages at that moment. Indicators such as the coefficient of variation and the quantile difference after voltage sorting can also be introduced to enhance the representation of the degree of inconsistency. Dynamic evolution characteristics can also be extracted by combining multiple multi-cell voltage lists. For example, the change in voltage of each cell can be calculated from the end of the resting period before charging to the end of charging, and the dynamic statistical characteristics of these changes can be analyzed (such as the standard deviation of the change and the maximum increase / decrease). Or, the increase rate of the overall voltage difference (such as the standard deviation) during charging can be calculated to reflect the deterioration trend of the voltage difference as charging progresses. Cross-time-sequence correlation characteristics can also be constructed, such as comparing the changes in voltage sorting of different cells in the multi-cell voltage list before and after charging to identify whether there are abnormal voltage jumps or lags in individual cells. The above static statistical characteristics, dynamic statistical characteristics, and correlation characteristics can be combined to obtain a voltage feature vector. Based on voltage feature vectors and multi-cell voltage lists, differential voltage fault detection is performed to distinguish different battery differential voltage types such as "normal", "voltage stratification", and "multi-point voltage drop".

[0056] In this embodiment, feature extraction is performed on at least two multi-unit voltage lists to obtain voltage feature vectors, including: inputting at least two multi-unit voltage lists into a trained feature extraction model and outputting voltage feature vectors.

[0057] The feature extraction model can be used to learn and extract key numerical features that characterize the voltage difference pattern from the input multi-cell voltage list. For example, the feature extraction model can be a trained machine learning or deep learning model (such as a fully connected neural network, autoencoder, or temporal encoder). The voltage feature vector includes at least: the overall voltage standard deviation, overall voltage median, and overall voltage average of the battery pack to be tested, as well as the individual cell voltage standard deviation, individual cell voltage median, and individual cell voltage average of the multiple battery modules in the battery pack to be tested. The overall voltage standard deviation, overall voltage median, and overall voltage average can be statistical indicators calculated based on the voltage values ​​of all individual cells in the battery pack to be tested at the same time, reflecting the dispersion, median trend, and average level of the overall voltage, respectively, and used to characterize the overall consistency of the battery pack. The individual cell voltage standard deviation, individual cell voltage median, and individual cell voltage average of the battery module can be calculated based on the standard deviation, median, and average value of the individual cell voltage values ​​contained in each battery module within the battery pack.

[0058] In this embodiment, a list of multi-cell voltages obtained from two or more charging process endpoints can be used as input data for the feature extraction model. The feature extraction model can internally encode and transform the multi-cell voltage list at each time step to extract its deep features; combined with the deep features, it outputs a voltage feature vector.

[0059] For example, the output voltage feature vector contains at least the following features:

[0060] Packet-level characteristics: For the multi-cell voltage list at each time point, calculate the overall voltage standard deviation, overall voltage median, and overall voltage average of the battery pack under test to describe the macroscopic distribution characteristics of the overall pack voltage.

[0061] Module-level characteristics: For each battery module in the battery pack, calculate the standard deviation, median, and average value of its internal single-cell voltages. These three characteristics of all modules are arranged in a preset order to form module-level characteristics.

[0062] All package-level and module-level features are combined to form a voltage feature vector. This voltage feature vector not only includes the concentration and dispersion trends of voltage, but also integrates multi-level consistency information of the battery pack as a whole and its local (module) components, ensuring the accuracy of subsequent differential pressure fault classification by the large language model.

[0063] S130. Input at least two multi-unit voltage lists and voltage feature vectors into the trained large language model, and output the initial differential pressure detection results.

[0064] The large language model refers to a language model architecture that possesses the ability to recognize differential pressure patterns after being trained on a large amount of labeled battery differential pressure data (including voltage lists, voltage feature vectors, and their corresponding fault type labels and fault handling methods). The initial differential pressure detection result refers to the preliminary judgment result of the differential pressure state output by the large language model based on the analysis of the input data, along with the corresponding fault handling method. For example, the initial differential pressure detection result includes whether the differential pressure is abnormal, the possible differential pressure types (such as "normal," "voltage stratification," "multi-point voltage drop," etc.), and the corresponding fault handling method for each differential pressure type.

[0065] In this embodiment, a multi-cell voltage list and voltage feature vectors can be input together into a pre-trained large language model. The large language model can utilize its sequence modeling capabilities to perform deep feature understanding on the input multi-time-stack multi-cell voltage list, such as capturing spatial difference patterns in voltage between individual cells and analyzing the dynamic evolution trend of voltage distribution during charging; or it can identify the changing behavior of key cells (such as abnormal lag or sudden drop in voltage of individual cells) through an internal attention mechanism, and combine this with the voltage feature vectors to comprehensively judge the differential voltage status. The large language model outputs the initial differential voltage detection result, clearly indicating whether the differential voltage of the battery pack under test is normal during charging, or what type of fault differential voltage it belongs to.

[0066] S140. Input the initial differential pressure detection result and voltage feature vector into the error correction model, and output the target differential pressure detection result corresponding to the battery to be detected.

[0067] In this embodiment, although the initial differential pressure detection result has high accuracy, it may still misjudge when facing ambiguous boundaries or rare patterns. Further verification and correction can be performed based on an error correction model. For example, the error correction model can be a pre-trained language model (or a large model-driven intelligent module) with reasoning and judgment capabilities, used to verify and correct the initial differential pressure detection result. This error correction model can understand the input data and built-in logic to output the final target differential pressure detection result. The target differential pressure detection result may include the fault type and fault handling method. For example, the error correction model can combine voltage feature vectors to evaluate the rationality of the initial differential pressure detection result, identify potential misjudgments, and output a more robust and accurate target differential pressure detection result.

[0068] Specifically, the initial differential voltage detection result output by the large language model and the voltage feature vector can be input together into a pre-trained error correction model. The error correction model can parse the category label in the initial differential voltage detection result and perform consistency verification in conjunction with the input voltage feature vector. For example, if the initial differential voltage detection result is "voltage stratification," the model will focus on verifying whether the voltage feature vector exhibits typical characteristics such as moderate differential voltage and a stepped voltage distribution; if the initial differential voltage detection result is "multi-point voltage drop," the model will focus on checking whether there are indicators such as sudden voltage drops in multiple cells and rapid increases in differential voltage in the voltage feature vector. The error correction model can identify the judgment bias of the initial differential voltage detection result under specific feature combinations by learning from a large number of historical misjudgment cases and feature distribution patterns. When the initial differential pressure detection result is found to be mismatched with the actual pattern reflected by the voltage feature vector (e.g., the feature vector shows an extremely small differential pressure, but the initial differential pressure detection result is judged as abnormal), the error correction model can activate a correction mechanism to reassess the confidence level of each differential pressure fault type and output the differential pressure fault type with the highest confidence level as the target differential pressure detection result; or, by combining the voltage feature vector, the initial differential pressure detection result can be corrected to obtain the corrected target differential pressure detection result. At this point, the target differential pressure detection result is the final differential pressure diagnosis result after correction by the error correction model. It can accurately represent the differential pressure status of the battery pack under test and can be used for actual operation and maintenance decisions, such as triggering alarms and generating maintenance suggestions.

[0069] In this embodiment, the initial differential pressure detection result and voltage feature vector are input into the error correction model, and the target differential pressure detection result corresponding to the battery to be detected is output. This includes: correcting the initial differential pressure detection result based on the verification rules pre-written in the error correction model to obtain the target differential pressure detection result.

[0070] The verification rules can refer to pre-set logical judgment conditions based on battery electrochemical characteristics and actual operation and maintenance experience, used to identify typical misjudgments in the initial differential pressure detection results. Optionally, the verification rules may include at least the following: if the initial differential pressure detection result is of the multi-point drop type, and the standard deviation of the individual cell voltage in the voltage feature vector is less than a preset threshold, then the initial differential pressure detection result is corrected to the voltage stratification type. The voltage standard deviation can be used to measure the dispersion of the voltage of each cell at a certain moment or process. The larger the standard deviation value, the more significant the voltage difference between cells; the smaller the value, the more concentrated the voltage distribution. The preset threshold can refer to the critical value of the voltage standard deviation determined by statistical analysis of a large amount of measured data, used to distinguish the boundary between "significant voltage difference" and "slight voltage difference". Optionally, the preset threshold is 0.01V. This preset threshold can reflect the voltage difference intensity necessary for "multi-point drop" type faults. The multi-point drop type can be manifested as a significant and synchronous abnormal drop in the voltage of multiple cells during charging, accompanied by a significant increase in the overall voltage difference (voltage standard deviation). Voltage stratification can be characterized by a stepped distribution of cell voltage, with moderate voltage differential and small standard deviation.

[0071] In practical applications, domain-knowledge-based verification rules can be pre-embedded in the error correction model. When the error correction model receives the initial differential pressure detection result and voltage feature vector, it can analyze whether the initial differential pressure detection result is of the "multi-point voltage drop type." If the initial differential pressure detection result is of the "multi-point voltage drop type," it further extracts the standard deviation of the individual cell voltage from the voltage feature vector and compares it with a pre-set threshold. If the standard deviation of the individual cell voltage is less than the preset threshold, it indicates that although the large language model judges it as a multi-point voltage drop type, the actual voltage difference between cells is relatively weak, not conforming to the strong discrete characteristics that this fault should have, and there is a possibility of misjudgment. At this time, the rule correction mechanism is triggered, adjusting the initial differential pressure detection result from the multi-point voltage drop type to the voltage stratification type, and outputting the corrected target differential pressure detection result.

[0072] Through this verification rule, the error correction model can verify the logical consistency of the output results of the large language model, prevent misclassification caused by the oversensitivity of the large language model or feature confusion, and thus output more accurate target pressure difference detection results that conform to physical reality.

[0073] In this embodiment, verification rules can be written into the prompts in the error correction model. A prompt is a guiding text instruction provided to the error correction model when it is invoked, used to clarify the task objective, input format, inference logic, and output requirements. Prompts are key features controlling the behavior of the error correction model, and may contain task descriptions, contextual information, rule constraints, and the expected response format.

[0074] Specifically, when constructing the reasoning process of the error correction model, the verification rules do not need to be hard-coded into the model as programming logic. Instead, the verification rules can be transformed into clear and rigorous natural language descriptions and integrated into the prompts sent to the error correction model. For example, the prompts could explicitly state: "You are a battery differential pressure fault diagnosis model. Based on the provided initial differential pressure detection results and voltage feature vector, please determine and output the final target differential pressure detection result. The following rules must be followed: If the initial differential pressure detection result is 'multi-point plunge,' but the voltage standard deviation in the voltage feature vector is less than 5mV, then the judgment is unreasonable and should be corrected to 'voltage stratification.' In other cases, please make a comprehensive judgment based on the feature vector to ensure that the result conforms to the battery's physical characteristics."

[0075] In actual operation, the initial differential pressure detection result and voltage feature vector of the battery pack under test (described in text form, such as "voltage standard deviation: 3.2mV") can be concatenated with the above-mentioned prompt words containing verification rules and input together into the error correction model. After understanding the entire context, the error correction model can actively apply the verification rules declared in the prompt words to perform logical reasoning: when it identifies that the initial differential pressure detection result is "multi-point drop" and the standard deviation is lower than the preset threshold, it automatically triggers the correction mechanism and outputs "voltage stratification" as the target differential pressure detection result.

[0076] The advantage of this setup is that it leverages the error correction model's ability to understand and execute natural language rules, enabling flexible embedding and dynamic application of verification logic, thereby improving the interpretability and ease of maintenance of battery differential pressure fault detection.

[0077] The technical solution provided in this embodiment obtains the current charging data of the battery pack to be tested and determines a list of multiple individual cell voltages at at least two charging process endpoints from the current charging data; performs feature extraction on the at least two multiple individual cell voltage lists to obtain voltage feature vectors; inputs the at least two multiple individual cell voltage lists and voltage feature vectors into a trained large language model to output an initial differential pressure detection result; inputs the initial differential pressure detection result and voltage feature vectors into an error correction model to output a target differential pressure detection result corresponding to the battery to be tested. This solves the problem in existing technologies that use expert rules based on fixed thresholds and traditional machine learning models for differential pressure detection, which have low accuracy in identifying battery differential pressure faults and are prone to false alarms and false negatives. This solution achieves comprehensive... By acquiring a list of multi-cell voltages of the battery pack under test at at least two charging process endpoints, and performing deep feature extraction on the multi-cell voltage list to generate physically meaningful voltage feature vectors, the multi-cell voltage list and the high-order voltage feature vectors are jointly input into a trained large language model. This model outputs an initial differential pressure detection result based on understanding the evolution of the charging process. Then, an error correction model is used to fuse the voltage feature vectors to perform logical consistency verification and correction on the initial differential pressure detection result, finally outputting a highly reliable target differential pressure detection result. This improves the detection accuracy of differential pressure faults in the battery pack, enhances the accuracy and robustness of differential pressure fault identification, effectively reduces false alarms and false negatives in differential pressure detection, and enhances the safety and reliability of battery system operation.

[0078] Figure 3 This is a flowchart of a battery differential pressure detection method according to an embodiment of the present invention. Based on the foregoing embodiments, a large language model can also be pre-trained. Specific implementation details can be found in the technical solution of this embodiment. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0079] like Figure 3 As shown, the method specifically includes the following steps:

[0080] S210: Obtain multiple differential pressure fault detection command corpora.

[0081] The differential pressure fault detection command corpus refers to a specialized mixed sample of text data used to train models to identify faults caused by excessive differences in cell voltage within a battery system. The differential pressure fault detection command corpus includes historical multi-cell voltage lists at at least two charging process endpoints, a corresponding voltage feature vector to be used, the fault type, and the fault handling method. In other words, each differential pressure fault detection command corpus contains a list of voltages of all individual cells within the battery pack at multiple charging process endpoints, clearly labeled fault types (such as "single-point voltage drop," "voltage stratification," "differential pressure exceeding limits," etc.), and corresponding fault handling methods (such as "replace module," "perform active balancing," "replace battery pack," etc.). The historical multi-cell voltage list refers to the numerical sequence of cell voltage values ​​recorded sequentially at charging process endpoints (such as the end of the rest period before charging begins, or the end of charging) during the battery pack's past operation. Each historical multi-cell voltage list reflects the voltage distribution state of the battery pack at that moment. The fault type can refer to a classification label for abnormal differential pressure patterns in the battery pack, used to identify the fault category represented by the differential pressure fault detection command corpus. Fault types can include "normal", "voltage stratification" (cell voltage is distributed in a stepped manner), and "multi-point voltage drop" (multiple cells experience abnormal voltage drops). Fault handling methods can refer to standardized measures taken in actual operation and maintenance for the actual fault type, such as "recommending equalization charging", "checking the tightness of connectors", "replacing the abnormal cell", and "strengthening thermal management monitoring".

[0082] In this embodiment, a large number of battery pack charging data segments that have undergone diagnosis can be filtered from the cloud-based historical database of the battery management system as the original data source. For each charging data segment, invalid charging conditions can be preset to eliminate the segments and obtain valid charging segments. From the valid charging segments, a list of historical multi-cell voltages at each charging endpoint is extracted. Then, based on expert experience, on-site inspection reports, or historical maintenance records, the voltage difference status of each data segment can be manually labeled or determined through backtracking using a high-confidence algorithm to identify its fault type. Simultaneously, the corresponding fault handling method can be associated with the standard maintenance procedures or actual handling records for this fault type. At least two historical multi-cell voltage lists can be input into a trained feature extraction model to output a voltage feature vector to be used. The voltage feature vector to be used includes at least: the overall voltage standard deviation, overall voltage median, and overall voltage average of the battery pack, as well as the individual cell voltage standard deviation, individual cell voltage median, and individual cell voltage average of multiple battery modules in the battery pack. Each set of "historical multi-cell voltage list, corresponding voltage feature vector, fault type, and fault handling method" can be organized into a question-and-answer format and packaged into a complete differential pressure fault detection command corpus. By collecting a large amount of charging data covering various vehicle models, operating conditions, aging stages, and fault modes, a differential pressure fault detection command corpus can be constructed for training a large language model.

[0083] For example, the differential pressure fault detection command corpus can be represented as:

[0084] {"instruction":"Please identify the fault mode and output maintenance measures based on the information of the lithium iron phosphate battery system before and after charging. The voltage list of the battery system before and after charging is [v1, v2, ..., v96]. The calculated standard deviation, median, and mean of the individual cell voltage of the battery system are std, v_mod, and v_mean, respectively. The overall voltage standard deviation, median, and mean of each battery pack are [pack_std1, pack_std2, ..., pack_stdn], [pack_mod1, pack_mod2, ..., pack_modn], and [pack_mean1, pack_mean2, ..., pack_meann], respectively. The individual cell voltage standard deviation, median, and mean of each module are [module_std1, module_std2, ..., module_stdm], [module_mod1, module_mod2, ..., module_modm], and [module_mean1, module_mean2, ..., module_meanm], respectively.";

[0085] The list of voltages at the end of battery system charging is [v1, v2, ..., v96]. The calculated standard deviation, median, and mean of the individual battery system voltages are std, v_mod, and v_mean, respectively. The overall voltage standard deviation, median, and mean of each battery pack are [pack_std1, pack_std2, ..., pack_stdn], [pack_mod1, pack_mod2, ..., pack_modn], and [pack_mean1, pack_mean2, ..., pack_meann], respectively. The individual voltage standard deviation, median, and mean of each module are [module_std1, module_std2, ..., module_stdm], [module_mod1, module_mod2, ..., module_modm], and [module_mean1, module_mean2, ..., module_meanm], respectively.

[0086] Please identify the fault type (single-point voltage drop, voltage stratification, z3, ..., z12) based on the above information and provide the fault handling method (replace module, t2, ..., t9).

[0087] "input":"",

[0088] "output": "Fault type: Single-point voltage drop, voltage stratification;

[0089] Troubleshooting method: Replace the module and then perform equalization.

[0090] S220. For each differential pressure fault detection command corpus, input the differential pressure fault detection command corpus into the large language model to be trained to obtain the predicted voltage detection result.

[0091] The large language model to be trained can refer to a large language model undergoing fine-tuning in the differential pressure detection domain, possessing natural language understanding and generation capabilities. In this task, the large language model not only receives historical multi-unit voltage lists from the differential pressure fault detection command corpus, but also integrates the feature vectors of the voltages to be used from the historical multi-unit voltage lists, combining them with the command context to predict possible fault types and fault handling methods. The predicted voltage detection result can refer to the preliminary differential pressure state detection result output by the large language model based on the input historical multi-unit voltage lists and the feature vectors of the voltages to be used, after internal inference. The predicted voltage detection result includes the predicted fault type and fault handling method.

[0092] In this embodiment, during the model training phase, each differential pressure fault detection command corpus is input into the large language model to be trained. The large language model utilizes its sequence modeling capabilities to parse the input historical multi-cell voltage list, identifying the relative relationships and trends of voltage changes between cells. Simultaneously, it combines the voltage feature vectors to enhance the understanding of differential pressure patterns, focusing on changes in cell behavior (such as abnormal lag or sudden drops in voltage of individual cells), and comprehensively judging the evolution of differential pressure by integrating voltage lists from multiple time points. During forward propagation, the large language model outputs the fault category and fault handling method with the highest probability as the predicted voltage detection result. This predicted voltage detection result is compared with the manually labeled actual fault types and fault handling methods in the differential pressure fault detection command corpus, the loss function is calculated, and the model parameters of the large language model are updated through the backpropagation algorithm, gradually improving its accuracy and generalization ability in battery differential pressure detection tasks.

[0093] S230. Based on the predicted voltage detection results and the fault type and fault handling method in the differential pressure fault detection instruction corpus, determine the loss value.

[0094] The loss value can be a numerical metric used to measure the degree of difference between the prediction results of a large language model and the actual annotations. For example, the loss value can be calculated based on the degree of matching between the predicted voltage detection results and the actual annotated fault types and handling methods in the differential pressure fault detection instruction corpus. The smaller the loss value, the closer the model prediction is to the reality. This loss value is used for backpropagation to update the parameters of the large language model.

[0095] In this embodiment, a loss function can be used to compare the predicted voltage detection result with the preset fault type and fault handling method in the differential pressure fault detection command corpus to calculate the loss value. For example, the loss function can be the cross-entropy loss function. For example, if the fault category in the predicted voltage detection result is consistent with the actual fault type, the loss value is lower; if they are inconsistent, the loss value is higher. If the fault handling method in the predicted voltage detection result is consistent with the actual fault handling method, the loss value is reduced; if they are inconsistent, the loss value is increased. Alternatively, the loss value between the fault category and the actual fault type, and the loss value between the fault handling method in the predicted voltage detection result and the actual fault handling method can be calculated separately, and the two loss values ​​can be weighted and fused to obtain a comprehensive loss value. This loss value can comprehensively reflect the performance of the large language model in terms of both fault identification accuracy and the rationality of operation and maintenance suggestions. It serves as an optimization target for backpropagation, driving the model parameter update of the large language model and improving the accuracy and stability of differential pressure fault detection.

[0096] S240. Based on the loss value, the model parameters in the large language model are corrected to obtain a well-trained large language model.

[0097] In this context, model parameters in a large language model refer to adjustable variables that constitute the internal structure of the model. These parameters determine the model's reasoning ability on input data (such as historical multi-cell voltage lists and voltage feature vectors), and their values ​​are continuously optimized through the training process. A well-trained large language model refers to a model whose internal parameters have converged to a relatively optimal state after multiple rounds of data iteration and parameter updates. This large language model possesses the ability to accurately identify battery differential voltage fault types and can stably output reliable initial differential voltage detection results.

[0098] In this embodiment, the model parameters in the large language model can be corrected based on the calculated loss value using a gradient backpropagation algorithm. Specifically, the gradient of the loss value relative to the model parameters of each layer of the large language model can be calculated, i.e., the partial derivative of the loss function with respect to the current model parameter values. This gradient indicates the direction and magnitude of the model parameter adjustment to reduce prediction error. Optimization algorithms (such as Adam, SGD, etc.) can be used to update the model parameters in the large language model based on the gradient information. The model parameters are fine-tuned along the gradient in the opposite direction, so that the loss value generated by the next prediction of the same or similar corpus is reduced. This process is iterated repeatedly on a large corpus of differential pressure fault detection instructions. Each time, one or a batch of corpus is input, the loss value is calculated, the gradient is calculated through backpropagation, and the parameters are updated. As the number of training rounds increases, the large language model's ability to recognize complex patterns such as "voltage stratification" and "multi-point plunge" continuously improves, and the loss value tends to stabilize and reach a low level. When the performance of the large language model on the validation corpus no longer improves significantly or reaches the preset iteration period, the model parameter update can be stopped. At this point, the large language model is considered well-trained.

[0099] It should be noted that the large language model provided in this embodiment is applicable to battery differential pressure fault detection in all types of lithium iron phosphate battery systems without requiring additional transfer learning or retraining. This expert large model learns the underlying logic for identifying differential pressure faults, unaffected by the number of battery packs or modules in the battery system. Traditional machine learning or deep learning models have weak generalization capabilities or lack scalability: for example, a model trained on data from a battery system consisting of 360 cells may not be applicable to a battery system consisting of 576 cells. However, this technical solution, based on the characteristic that the input data of the large language model is natural language, is applicable to battery differential pressure fault detection in battery systems composed of any number of modules and battery packs.

[0100] The technical solution provided in this embodiment constructs a differential pressure fault detection command corpus containing a historical multi-cell voltage list at multiple charging endpoints, corresponding available voltage feature vectors, fault types, and fault handling methods. This corpus deeply integrates structured voltage data with semantic fault knowledge and serves as training samples for a large language model. During model training, the original multi-cell voltage list and high-order voltage feature vectors are simultaneously introduced, enabling the large language model to not only understand the spatiotemporal variation patterns of cell voltages but also enhance its perception of complex differential pressure faults with the help of pre-extracted features. Thus, during training, it learns the precise mapping relationship between multi-dimensional data and fault types and handling methods, overcoming the problems of false alarms and missed alarms caused by the reliance on manually set thresholds and poor generalization ability in traditional methods. This achieves end-to-end automated learning of differential pressure fault identification and fault handling methods, improving the accuracy, interpretability, and applicability of the large language model in battery system fault diagnosis tasks, and enhancing the safety and reliability of battery system operation.

[0101] As an optional embodiment of the above embodiments, specific application scenario examples are provided to enable those skilled in the art to further understand the technical solutions of the embodiments of the present invention. Specifically, please refer to the following detailed content.

[0102] See Figure 3 The system filters charging data from the vehicle network backend, removing discontinuous charging segments (idle intervals greater than 20 minutes, i.e., preset duration) or invalid SOC intervals (initial SOC greater than 30%, i.e., preset first charge level or ending SOC greater than 95%, i.e., preset second charge level) that do not meet preset invalid charging conditions, thus ensuring data quality. For segments that meet the charging / discharging criteria (i.e., valid charging data), the system extracts voltage lists for all individual cells before charging begins (i.e., at the end of the resting period) and at the end of charging, obtaining multi-cell voltage lists at the two endpoints of the charging process. Then, for these two multi-cell voltage lists, the system calculates and extracts their statistical and morphological characteristics to construct a multi-dimensional voltage feature vector. Voltage feature vectors include, but are not limited to: the standard deviation, median, and mean voltage of the battery system; the standard deviation, median, and mean voltage of each battery pack; the standard deviation, median, and mean voltage of each individual cell in each module; voltage fluctuation frequency, voltage trend slope, or other statistical features (such as skewness, kurtosis, interquartile range), morphological features (such as the number of peaks and troughs, curve length), and trend features (such as the slope of piecewise linear fitting, the slope of moving averages), etc. The advantage of calculating voltage feature vectors is that they can transform invisible patterns into quantitative indicators that are easier for large language models to understand, reducing the learning difficulty of large language models.

[0103] The list of multi-cell voltages at the two charging endpoints, along with the constructed voltage feature vector, serves as input data to the large language model. The model outputs an initial voltage difference detection result, such as "voltage stratification" or "multi-point voltage drop." This large language model, optimized for voltage data in the battery domain, possesses a deep understanding of battery voltage difference patterns. It should be noted that when adjusting the model parameters, efficient fine-tuning techniques based on low-rank adaptation (LoRA) can be used. Alternatively, traditional full-parameter fine-tuning, freezing partial network layers, or other efficient fine-tuning techniques such as cue word fine-tuning can be employed, including directly utilizing the model's zero-shot inference capabilities.

[0104] The initial differential pressure detection result output by the large language model is input into an error correction model based on verification rules. This error correction model performs secondary verification and correction on the initial differential pressure detection result according to preset verification rules. For example, the verification rule can be defined as: "If the initial differential pressure detection result is of the multi-point plunge type, but its corresponding voltage standard deviation is less than a preset threshold (e.g., 0.01V), then the initial differential pressure detection result is forcibly corrected to the voltage stratification type." Based on the verification rules, considering the fuzzy and unstable characteristics of the output result of the large language model, an additional verification step is needed to effectively filter out misjudgments of differential pressure faults in fuzzy boundary scenarios.

[0105] Specifically, verification rules can be written into the prompt words of the error correction model (such as a general large language model), allowing the error correction model to re-verify the output of the fine-tuned language model (i.e., the initial differential pressure detection result). This includes distinguishing or merging multiple fault handling methods (such as maintenance suggestions) when the large language model identifies multiple fault types and their corresponding fault handling methods, ensuring that the fault handling methods do not conflict or overlap. It should be noted that the error correction model can use verification rules as prompt words to guide the model, or adopt mechanisms that integrate multiple models.

[0106] Furthermore, the target differential pressure detection results, after error correction, can be integrated to generate a structured battery health diagnostic report. This report can include not only specific fault types (such as "voltage stratification"), but also corresponding fault handling methods (such as "battery balancing"), thus forming a complete diagnostic recommendation report.

[0107] The technical solution of this embodiment combines three steps in sequence: voltage feature extraction, large language model prediction, and error correction model correction. Specifically, it first enhances the input data of the large language model, then uses the large language model for analysis, and finally corrects the output results of the large language model according to rules. This solves the problems of low accuracy or inability to identify mixed fault modes in the prior art when identifying differential pressure faults. It can improve the detection accuracy of complex differential pressure faults, improve the accuracy of battery fault classification in complex scenarios, and at the same time reduce the false judgment rate of differential pressure faults.

[0108] Figure 4 This is a schematic diagram of a battery differential pressure detection device according to an embodiment of the present invention. Figure 4 As shown, the device includes: a multi-unit voltage list determination module 310, a voltage feature vector determination module 320, an initial differential pressure detection result determination module 330, and a target differential pressure detection result determination module 340.

[0109] The module includes a multi-cell voltage list determination module 310, which acquires the current charging data of the battery pack to be tested and determines the multi-cell voltage list at at least two charging process endpoints from the current charging data; a voltage feature vector determination module 320, which extracts features from at least two of the multi-cell voltage lists to obtain voltage feature vectors; an initial differential pressure detection result determination module 330, which inputs at least two of the multi-cell voltage lists and the voltage feature vectors into a trained large language model and outputs an initial differential pressure detection result; and a target differential pressure detection result determination module 340, which inputs the initial differential pressure detection result and the voltage feature vectors into an error correction model and outputs a target differential pressure detection result corresponding to the battery to be tested.

[0110] The technical solution of this embodiment obtains the current charging data of the battery pack to be tested and determines a multi-cell voltage list at at least two charging process endpoints from the current charging data; performs feature extraction on the at least two multi-cell voltage lists to obtain voltage feature vectors; inputs the at least two multi-cell voltage lists and voltage feature vectors into a trained large language model to output an initial differential pressure detection result; inputs the initial differential pressure detection result and voltage feature vectors into an error correction model to output a target differential pressure detection result corresponding to the battery to be tested. This solves the problem in the prior art of differential pressure detection based on expert rules with fixed thresholds and traditional machine learning models, which has low accuracy in identifying battery differential pressure faults and is prone to false alarms and false negatives. This solution achieves a solution through... The system acquires a list of multi-cell voltages of the battery pack under test at at least two charging endpoints, performs deep feature extraction on the multi-cell voltage list to generate physically meaningful voltage feature vectors, and then inputs the multi-cell voltage list and high-order voltage feature vectors into a trained large language model. This model outputs an initial differential pressure detection result based on its understanding of the charging process evolution. Then, an error correction model fuses the voltage feature vectors to perform logical consistency verification and correction on the initial differential pressure detection result, finally outputting a highly reliable target differential pressure detection result. This improves the detection accuracy of differential pressure faults in the battery pack, enhances the accuracy and robustness of differential pressure fault identification, effectively reduces false alarms and false negatives in differential pressure detection, and enhances the safety and reliability of the battery system operation.

[0111] Based on the above-described device, optionally, the multi-unit voltage list determination module 310 includes:

[0112] The valid charging data determination unit is used to remove the current charging data based on preset invalid charging conditions to obtain valid charging data;

[0113] A multi-cell voltage list determination unit is used to extract the effective charging data to obtain a multi-cell voltage list at at least two charging process endpoints.

[0114] Among them, at least two of the charging process endpoint times include the end time of rest before charging begins and the end time of charging.

[0115] Based on the above device, optionally, the current charging data includes multiple charging segments, and the preset invalid charging condition includes at least one of the following:

[0116] The idle time interval between the two charging segments reaches a preset duration;

[0117] The initial charging capacity of the charging segment is greater than a preset first capacity;

[0118] The charging segment ends with a charge level less than a preset second charge level.

[0119] Based on the above device, optionally, a voltage feature vector determination module 320 is used to input at least two of the multi-unit voltage lists into the trained feature extraction model and output voltage feature vectors.

[0120] The voltage feature vector includes at least: the overall voltage standard deviation, overall voltage median, and overall voltage average of the battery pack to be tested, as well as the individual voltage standard deviation, individual voltage median, and individual voltage average of the multiple battery modules in the battery pack to be tested.

[0121] Based on the above device, optionally, a target differential pressure detection result determination module 340 is used to correct the initial differential pressure detection result based on the verification rules pre-written in the error correction model to obtain the target differential pressure detection result;

[0122] The verification rules include at least the following: if the initial differential pressure detection result is of the multi-point drop type and the standard deviation of the individual voltage in the voltage feature vector is less than a preset threshold, then the initial differential pressure detection result is corrected to the voltage stratification type; the preset threshold is 0.01V.

[0123] Optionally, based on the above-described apparatus, the apparatus may further include:

[0124] The large language model training module is used to train and obtain a large language model;

[0125] The large language model training module includes:

[0126] The corpus determination unit is used to acquire multiple differential pressure fault detection instruction corpora, wherein the differential pressure fault detection instruction corpora include at least two historical multi-cell voltage lists at the end points of the charging process, a voltage feature vector to be used corresponding to the historical multi-cell voltage lists, fault type, and fault handling method;

[0127] The predicted voltage detection result determination unit is used to input the differential pressure fault detection command corpus into the large language model to be trained for each differential pressure fault detection command corpus to obtain the predicted voltage detection result.

[0128] The loss value determination unit is used to determine the loss value based on the predicted voltage detection result and the fault type and fault handling method in the differential pressure fault detection instruction corpus.

[0129] The parameter correction unit is used to correct the model parameters in the large language model based on the loss value to obtain the trained large language model.

[0130] Optionally, based on the above-described apparatus, the apparatus may further include:

[0131] The verification rule writing unit is used to write the verification rules into the prompt words in the error correction model.

[0132] The battery differential pressure detection device provided in this embodiment of the invention can execute the battery differential pressure detection method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0133] Figure 5 This is a schematic diagram of an electronic device implementing the battery differential pressure detection method of this invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0134] like Figure 5 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory 12 or a random access memory 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the read-only memory 12 or a computer program loaded from storage unit 18 into the random access memory 13. The random access memory 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, read-only memory 12, and random access memory 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0135] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0136] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the battery differential pressure detection method.

[0137] In some embodiments, the battery differential pressure detection method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via read-only memory 12 and / or communication unit 19. When the computer program is loaded into random access memory 13 and executed by processor 11, one or more steps of the battery differential pressure detection method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the battery differential pressure detection method by any other suitable means (e.g., by means of firmware).

[0138] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0139] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0140] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0141] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0142] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0143] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0144] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from read-only memory 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.

[0145] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the battery differential pressure detection method provided in any embodiment of this invention.

[0146] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0147] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0148] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method of detecting a pressure difference in a battery, the method comprising: The method comprises: obtaining current charging data of a battery pack to be detected, and determining a plurality of multi-cell voltage lists at at least two charging process endpoint moments from the current charging data; performing feature extraction on the at least two multi-cell voltage lists to obtain a voltage feature vector; inputting the at least two multi-cell voltage lists and the voltage feature vector into a trained large language model to output an initial pressure difference detection result; inputting the initial pressure difference detection result and the voltage feature vector into a correction model to output a target pressure difference detection result corresponding to the battery to be detected.

2. The method of claim 1, wherein, The method further comprises: training a large language model; the training of the large language model comprises: obtaining a plurality of pressure difference fault detection instruction corpora, wherein the pressure difference fault detection instruction corpora comprise at least two historical multi-cell voltage lists at charging process endpoint moments, voltage feature vectors corresponding to the historical multi-cell voltage lists, fault types, and fault handling methods; 3. The method of claim 2, wherein, for each pressure difference fault detection instruction corpus, inputting the pressure difference fault detection instruction corpus into a large language model to be trained to obtain a predicted voltage detection result; and training the large language model based on the predicted voltage detection result and the voltage feature vector corresponding to the historical multi-cell voltage list to obtain the large language model. ​ ​ 4. The method of claim 1, wherein, ​ ​ ​ 5. The method of claim 1, wherein, ​ ​ ​ 6. The method of claim 1, wherein, ​ ​ ​ ​ ​ determine a loss value based on the predicted voltage detection result and the fault type and fault handling mode in the differential pressure fault detection instruction corpus; correct the model parameters in the large language model based on the loss value to obtain a trained large language model.

7. The method according to any one of claims 1 to 6, characterized in that, The method further comprises: writing the review rules into the prompt words in the error correction model.

8. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected in communication with the at least one processor; wherein The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the battery differential pressure detection method in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing the processor to implement the battery differential pressure detection method in any one of claims 1-7 when executed.

Images