Battery short-circuit fault intelligent diagnosis method, system, equipment and medium

Abstract

The invention relates to the technical field of battery fault diagnosis, and discloses a battery short-circuit fault intelligent diagnosis method, system and device and a medium, and the method comprises the steps: obtaining a battery unit voltage data set with a timestamp, and generating a diagnosis sequence containing a detection node; based on the detection node set, the monitoring terminal is controlled to collect voltage data, capture current fluctuation signals and extract feature vectors, abnormal mark information is determined, and the diagnosis process is controlled according to the state of the detection node set and preset conditions; and after the diagnosis is completed, sending the feature vector to an analysis server to obtain a fault positioning identifier, storing abnormal mark information, analyzing the difference between a mark information log and a historical log library to determine a fault evolution path, and generating and filing a diagnosis report when a preset path model is matched. The system comprises an acquisition unit, a sequence construction unit and a diagnosis execution unit. The equipment comprises a processor and a storage device, and a medium stores corresponding programs. According to the invention, the accuracy, efficiency and intelligent level of fault diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery fault diagnosis, and in particular to a battery short circuit fault intelligent diagnosis method, system, equipment and medium. Background Art

[0002] With the rapid development of new energy technologies, battery packs are increasingly being used in electric vehicles, energy storage systems, and other fields. Battery packs typically consist of multiple battery cells connected in series or parallel, and their safety is directly related to the proper functioning of the equipment and the safety of users' lives and property. However, during use, batteries are prone to short-circuit failures due to factors such as internal structural aging, external mechanical damage, overcharging and over-discharging. Short-circuit failures not only degrade battery performance but, in severe cases, can even cause safety hazards such as fires and explosions. Therefore, timely and accurate diagnosis of battery short-circuit failures is crucial.

[0003] Traditional battery short-circuit fault diagnosis methods rely primarily on manual inspection or simple electrical measurements. Manual inspection is labor-intensive and time-consuming, and the results are significantly influenced by the operator's experience and subjective judgment, making it difficult to achieve a quick and accurate diagnosis. Simple electrical measurement methods, such as voltage and current measurements, while capable of providing some basic electrical parameters, often fail to accurately determine the location and severity of complex short-circuit faults.

[0004] With the development of intelligent sensing and data analysis technologies, several battery fault diagnosis methods based on sensor networks and data processing have been proposed. These methods deploy multiple sensors on the battery pack to collect real-time data such as battery cell voltage and current. These methods then use data analysis algorithms to process and analyze this data to diagnose battery short-circuit faults. However, existing diagnostic methods still have shortcomings in data processing efficiency, fault diagnosis accuracy, and real-time performance.

[0005] For example, in terms of data processing, existing methods usually perform simple threshold comparisons or statistical analysis on the collected voltage data, ignoring the time series relationship between the data and historical anomaly records, resulting in the inability to accurately capture the early characteristics of battery short-circuit faults. In terms of fault diagnosis accuracy, due to the randomness and complexity of the occurrence of battery short-circuit faults, existing methods are difficult to flexibly adjust according to different battery pack types and usage scenarios, resulting in limited accuracy of diagnostic results. In terms of real-time performance, existing methods usually require batch processing after collecting a large amount of data, which cannot achieve real-time diagnosis and early warning of short-circuit faults, and thus cannot take timely measures to avoid accidents.

[0006] In addition, the existing battery short circuit fault diagnosis system usually lacks analysis and recording of the fault evolution process, and cannot provide effective data support for the maintenance and management of the battery. Therefore, there is an urgent need for a battery short circuit fault intelligent diagnosis method and system that can improve data processing efficiency, enhance fault diagnosis accuracy and real-time performance, and analyze and record the fault evolution process. SUMMARY

[0007] The purpose of the present application is to provide a battery short circuit fault intelligent diagnosis method, system, device and medium to solve the problems raised in the background art.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical solutions: a battery short circuit fault intelligent diagnosis method, system, device and medium, the method comprising:

[0009] Obtaining a voltage data set of a plurality of battery cells corresponding to a target battery pack, wherein the cell voltage data in the voltage data set includes timestamp information;

[0010] Generating a diagnosis sequence according to the timestamp information included in the voltage data set, wherein the diagnosis sequence includes a plurality of detection nodes, and the number of detection nodes included in the diagnosis sequence is the same as the number of cell voltage data included in the voltage data set;

[0011] Based on the detection node set included in the diagnosis sequence, the following diagnosis steps are performed:

[0012] Controlling the monitoring terminal to collect voltage data corresponding to the timestamp information of the first detection node in the detection node set;

[0013] In response to capturing the current fluctuation signal, performing feature extraction processing on the current fluctuation signal to generate a feature vector;

[0014] According to the feature vector, determining the abnormal marker information of the cell voltage data corresponding to the detection node;

[0015] In response to the detection node set after deleting the first detection node being empty, controlling the monitoring terminal to output a diagnosis completion instruction;

[0016] In response to the detection node set after deleting the first detection node being not empty, and the monitoring terminal meeting the preset diagnosis condition, controlling the monitoring terminal to output a short circuit alarm instruction, and in response to detecting an alarm confirmation operation, taking the detection node set after deleting the first detection node as the detection node set, and executing the diagnosis step again.

[0017] Preferably, the method further comprises:

[0018] In response to the detection node set after deleting the first detection node not being empty and the monitoring terminal not satisfying the preset diagnosis condition, the detection node set after deleting the first detection node is taken as the detection node set, and the diagnosis step is executed again.

[0019] Preferably, the diagnosis sequence is generated according to the time stamp information included in the voltage data set, and the diagnosis sequence includes:

[0020] Obtaining reference time stamp information corresponding to the monitoring terminal;

[0021] According to the reference time stamp information, unit voltage data satisfying a first screening condition is screened from the voltage data set as reference voltage data, and time stamp information included in the reference voltage data is determined as a starting node.

[0022] The voltage data set after removing the reference voltage data is taken as a first data set, and the following sequence generation step is executed based on the first data set:

[0023] Obtaining a historical record group of each first data in the first data set, to obtain a historical record group set;

[0024] For each first data in the first data set, the mean value of each historical abnormal record included in the historical abnormal record set corresponding to the first data is determined as a risk prediction value;

[0025] The time interval difference between each first data in the first data set and the reference voltage data is determined to obtain a time interval set;

[0026] For each first data in the first data set, the data priority score is determined according to the risk prediction value and the time interval difference corresponding to the first data;

[0027] From the determined data priority scores, a priority score satisfying a second screening condition is selected as a target priority score, and the time stamp information included in the first data corresponding to the target priority score is determined as a target detection node;

[0028] In response to the first data set after deleting the first data corresponding to the target priority score being empty, the starting node and the target detection node are combined as a diagnosis sequence;

[0029] In response to the first data set after deleting the first data corresponding to the target priority score not being empty, the first data set after deleting the first data corresponding to the target priority score is taken as the first data set, the target detection node is determined as the starting node, and the sequence generation step is executed again.

[0030] Preferably, the diagnostic device further comprises an analysis server and a storage terminal;

[0031] In addition, after the monitoring terminal outputs the diagnosis completion instruction in response to the detection node set of the first detected node being empty, the method further comprises:

[0032] The monitoring terminal sends a set of feature vectors to the analysis server;

[0033] The analysis server returns a fault location identifier;

[0034] The storage terminal records all abnormal marker information corresponding to the current diagnosis cycle;

[0035] In response to the end of the diagnosis cycle, the analysis server extracts the marker information log of the storage terminal;

[0036] According to the difference value between the marker information log and the historical log library, a fault evolution path is determined;

[0037] In response to the fault evolution path matching a preset path model, at least one unit voltage data included in at least one feature vector corresponding to the marker information log is obtained;

[0038] Each unit voltage data in the at least one unit voltage data and the corresponding abnormal marker information are combined into a diagnostic report, and the diagnostic report is sent to the storage terminal for archiving.

[0039] Preferably, before the fault evolution path is determined according to the difference value between the marker information log and the historical log library, the method further comprises:

[0040] In response to the data amount of the marker information log being less than a data amount threshold of the historical log library, a data invalidity identifier is generated, and the data invalidity identifier is updated in the storage terminal.

[0041] Preferably, the method further comprises:

[0042] The analysis server transmits the recorded fault evolution path to a preset database for model optimization processing of the fault evolution path.

[0043] Preferably, the present application further comprises a battery short circuit fault intelligent diagnosis system, the system comprising:

[0044] The acquisition unit is configured to obtain a set of voltage data of a plurality of battery cells corresponding to a target battery pack, wherein the unit voltage data in the set of voltage data includes timestamp information;

[0045] a sequence construction unit configured to generate a diagnosis sequence according to the timestamp information included in the voltage data set, wherein the diagnosis sequence includes a plurality of detection nodes, and the number of detection nodes included in the diagnosis sequence is the same as the number of cell voltage data included in the voltage data set;

[0046] a diagnosis execution unit configured to execute the following diagnosis steps based on the detection node set included in the diagnosis sequence: control the monitoring terminal to collect voltage data of the timestamp information corresponding to the first detection node in the detection node set; in response to capturing the current fluctuation signal, perform feature extraction processing on the current fluctuation signal to generate a feature vector; determine abnormal marker information of the cell voltage data corresponding to the detection node according to the feature vector; in response to the detection node set after deleting the first detection node being empty, control the monitoring terminal to output a diagnosis completion instruction; in response to the detection node set after deleting the first detection node not being empty and the monitoring terminal satisfying a preset diagnosis condition, control the monitoring terminal to output a short-circuit alarm instruction, and in response to detecting an alarm confirmation operation, take the detection node set after deleting the first detection node as the detection node set and execute the diagnosis steps again.

[0047] Preferably, the sequence construction unit further includes:

[0048] configured to call different timestamp sorting algorithms according to the battery pack type code to adjust the node generation frequency of the diagnosis sequence.

[0049] Preferably, the application further includes a device, which includes:

[0050] one or more processors;

[0051] a storage device having one or more programs stored thereon;

[0052] when the one or more programs are executed by the one or more processors, the one or more processors implement the method of the battery short-circuit fault intelligent diagnosis as described above.

[0053] Preferably, the program is executed by the processor to implement the method of the battery short-circuit fault intelligent diagnosis as described above.

[0054] Compared with the prior art, the application has the following beneficial effects:

[0055] In terms of data processing, the application generates a diagnosis sequence according to the timestamp information in the voltage data set of the target battery pack. The generation process of the diagnosis sequence fully considers the time series relationship of the voltage data and the historical abnormal records. By filtering the reference voltage data, calculating the risk prediction value, time interval difference and data priority score, the voltage data can be effectively sorted and filtered, improving the efficiency and pertinence of data processing. This data processing method can highlight the key detection nodes and ensure that high-risk data is processed first during diagnosis, thereby improving the accuracy of fault diagnosis.

[0056] In the fault diagnosis process, the application controls the monitoring terminal to sequentially collect the voltage data of each detection node based on the detection node set of the diagnosis sequence, and performs feature extraction and abnormal marking when the current fluctuation signal is captured. This node-by-node diagnosis method can achieve fine detection of battery cells and accurately determine the abnormal condition of the cell voltage data corresponding to each detection node. At the same time, by setting a preset diagnosis condition, the short circuit alarm instruction can be output in time when the condition is met, realizing real-time warning of short circuit faults and improving the real-time performance of diagnosis. In addition, when the detection node set is not empty and the preset diagnosis condition is not met, the diagnosis steps can be automatically continued to ensure the integrity and continuity of the diagnosis process.

[0057] In terms of system architecture, the diagnosis device of the application includes monitoring terminals, analysis servers and storage terminals, and the components cooperate with each other to realize the integration of data acquisition, analysis, storage and fault positioning functions. The monitoring terminal is responsible for real-time acquisition of voltage data, the analysis server processes the feature vectors and locates faults, and the storage terminal records various information during diagnosis. Through this architecture design, comprehensive diagnosis and management of battery short circuit faults can be achieved. For example, after diagnosis is completed, the monitoring terminal sends the feature vector set to the analysis server, the analysis server returns the fault positioning identifier, and the storage terminal records the abnormal marking information. After the diagnosis period ends, the analysis server can determine the fault evolution path according to the difference value between the marking information log and the historical log library, and generate a diagnosis report for archiving when it matches the preset path model. This analysis and recording of the fault evolution path can provide valuable data support for the maintenance and management of the battery, helping to detect potential faults in advance, develop a reasonable maintenance plan and extend the service life of the battery.

[0058] In addition, the present application also considers the validity of the data and the optimization of the model. Before determining the fault evolution path, by judging whether the data amount of the mark information log is less than the data amount threshold of the historical log library, a data invalidity identifier is generated and updated, ensuring the validity and reliability of the data. At the same time, the recorded fault evolution path is transmitted to the preset database for model optimization processing, which can continuously improve the performance and accuracy of the diagnosis system, so that it can adapt to different types of battery packs and complex use scenarios.

[0059] In addition, the sequence construction unit in the diagnosis system of the present application can call different timestamp sorting algorithms according to the battery pack type code, and adjust the node generation frequency of the diagnosis sequence. This design enables the diagnosis system to be flexibly adjusted for different types of battery packs, improving the universality and adaptability of the system, and meeting the diagnosis requirements in different application scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 The working principle diagram of the battery short circuit fault intelligent diagnosis method of the present application;

[0061] Figure 2 The working flowchart of the diagnosis sequence generation method;

[0062] Figure 3 The working flowchart of the fault evolution analysis and report generation;

[0063] Figure 4 The working flowchart of the battery short circuit fault intelligent diagnosis system;

[0064] Figure 5 The working flowchart of the fault evolution path analysis and model optimization. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0066] Please refer to Figures 1-5 The battery short circuit fault intelligent diagnosis method of the present application is applied to a diagnosis device comprising a monitoring terminal, and the specific implementation steps are as follows:

[0067] The diagnosis device collects the voltage data of a plurality of battery cells in the target battery pack through the monitoring terminal in real time to form a voltage data set. Each unit voltage data carries timestamp information for marking the time point of data collection, ensuring the time sequence and traceability of the data.

[0068] Based on the timestamp information in the voltage data set, a diagnostic sequence is constructed. This sequence contains the same number of detection nodes as the unit voltage data, and each node corresponds to a timestamp for subsequent staged fault diagnosis.

[0069] Based on the detection node set of the diagnostic sequence, diagnose each node according to the following process:

[0070] The control monitoring terminal collects the voltage data corresponding to the timestamp of the first node in the detection node set and monitors the current signal synchronously.

[0071] If a current fluctuation signal is captured, feature extraction is immediately performed to generate a feature vector. This feature extraction can use signal processing algorithms such as Fourier transform and wavelet analysis to extract key characteristic parameters such as the frequency, amplitude, and phase of the current fluctuation.

[0072] According to the characteristic vector, a preset diagnostic model (such as a neural network, a support vector machine, etc.) is used to analyze and determine whether the unit voltage data corresponding to the detection node is abnormal, and generate abnormal marking information (such as "normal", "suspected short circuit", "confirmed short circuit", etc.).

[0073] If the detection node set is empty after the first detection node is deleted, it means that all nodes have been diagnosed, and the control monitoring terminal outputs a diagnosis completion instruction.

[0074] If the detection node set is not empty after deleting the first detection node, and the monitoring terminal meets the preset diagnostic conditions (such as multiple nodes detecting abnormal signals in succession, the degree of abnormality exceeds the threshold, etc.), the monitoring terminal is controlled to output a short-circuit alarm instruction, and after receiving the alarm confirmation operation, the remaining detection node set is used as the new node set and the above diagnostic steps are repeated; if the preset diagnostic conditions are not met, the remaining node set is directly used as the new node set and the diagnosis continues.

[0075] The present invention will be further described below in conjunction with Examples 1 to 5:

[0076] Example 1: Building on the above overall solution, this example further defines the branching logic of the diagnostic process: if the set of detection nodes after deleting the first detection node is not empty, the system determines whether the monitoring terminal meets the preset diagnostic conditions and executes different processing flows based on the judgment result. The preset diagnostic conditions are pre-set judgment criteria based on the typical characteristics of battery short-circuit faults and historical data, used to distinguish true fault signals from accidental interference signals.

[0077] If the monitoring terminal does not meet the preset diagnostic conditions, for example, the current fluctuation signal currently detected only appears once and does not show regularity or continuity, or the characteristic parameters such as the amplitude and frequency of the abnormal signal do not reach the preset alarm threshold, the system determines that the currently detected abnormality may be noise interference or occasional fluctuations, and does not constitute a reliable basis for a short-circuit fault. At this time, the system does not trigger the short-circuit alarm instruction, but directly uses the remaining detection node set after deleting the first detection node as a new detection node set and executes the diagnostic steps again. The design purpose of this mechanism is to avoid false alarms caused by single or atypical abnormal signals, and to accumulate more diagnostic basis by continuously monitoring the voltage data and current signals of subsequent detection nodes, thereby improving the accuracy and reliability of the diagnostic results.

[0078] During the specific implementation process, the preset diagnostic conditions can be configured and adjusted through system parameters to adapt to the diagnostic needs of battery packs of different types and usage scenarios. For example, for power battery packs with higher stability requirements, more stringent preset diagnostic conditions can be set, requiring that multiple consecutive detection nodes (such as 3) detect current fluctuation signals exceeding a specific threshold, and the characteristic parameters of each signal are consistent before the alarm condition is determined to be met; for small energy storage battery packs, the strictness of the preset conditions can be appropriately adjusted according to their operating characteristics. This flexibility enables the diagnostic system to adapt to a variety of application scenarios and improves the versatility of the system.

[0079] At the data processing level, each time a diagnostic step is executed, the system collects and analyzes the unit voltage data and current signals corresponding to each detection node in real time. When the diagnosis of the first detection node is completed, regardless of whether abnormality marking information is generated, the system will delete the node from the detection node set to ensure that each node is processed only once to avoid repeated diagnosis. If the remaining detection node set is not empty, the system will automatically enter the next round of diagnostic process and process subsequent detection nodes in sequence. During this process, the system will retain all abnormality marking information that has been generated, which will serve as an important basis for subsequent fault analysis and evolution path judgment.

[0080] It should be noted that the judgment of "not meeting the preset diagnostic conditions" is not only based on the diagnostic results of the current detection node, but may also be combined with historical diagnostic data for comprehensive analysis. For example, if a battery cell has never had an abnormal record in past diagnoses, and the current detection node only has a slight current fluctuation, the system may determine that the current abnormality is an occasional situation based on the historical stability of the cell and does not meet the alarm conditions. Conversely, if the cell frequently has abnormal signals in multiple recent diagnoses, even if the signal characteristics of the current single node do not fully reach the preset threshold, the system may adjust the trigger logic of the preset conditions by comprehensively evaluating historical data to detect potential faults earlier.

[0081] Furthermore, the system maintains the sequential order and continuity of the diagnostic sequence when processing the remaining detection node sets. Each node in the detection node set is arranged in chronological order by timestamp, ensuring that the diagnostic process proceeds sequentially according to the chronological order of data acquisition, accurately capturing the temporal trends of battery cell voltage and current. This chronological, node-by-node diagnostic approach helps the system track the development of faults and provides a complete chain of sequential data for subsequent fault location and evolution analysis.

[0082] In software implementation, this branching logic can be implemented through conditional judgment statements. For example, in the code flow of the diagnostic step, after completing the processing of the first detection node and deleting the node, the system will perform a conditional judgment: first, it checks whether the remaining detection node set is empty. If not, it further determines whether the monitoring terminal meets the preset diagnostic conditions. If so, the alarm process is triggered; if not, the remaining node set is directly used as the new input, recursively calling the processing function of the diagnostic step, and continuing to diagnose subsequent nodes. This modular code design makes the diagnostic process clear and controllable, facilitating subsequent maintenance and functional expansion.

[0083] From a hardware interaction perspective, the monitoring terminal performs data collection and command execution throughout the entire process. When the system needs to collect voltage data from a detection node, it sends a collection command to the monitoring terminal via the control bus. Upon receiving the command, the monitoring terminal immediately samples the voltage of the target battery cell and returns the timestamped voltage data to the system. Simultaneously, the monitoring terminal continuously monitors the battery pack's current signal and, if it detects current fluctuations, sends a trigger signal to the system, initiating feature extraction and diagnostic processes.

[0084] It is worth noting that the design of this embodiment does not completely exclude the importance of a single abnormal signal. Even if the current detection node does not meet the preset diagnostic conditions, the system will still record the abnormal mark information and continue to pay attention to the relevant battery cells in the subsequent diagnostic process. This "recording and trace, continuous monitoring" mechanism enables the system to accumulate sufficient historical data to provide support for long-term fault prediction and health management. For example, a battery cell may have slight current fluctuations in multiple diagnoses. Although no alarm is triggered each time because the preset conditions are not met, through analysis of these historical abnormal marks, the system can find that the failure tendency of the unit is gradually increasing, thereby issuing an early warning and realizing early identification and intervention of faults.

[0085] Example 2: This example describes in detail the diagnostic sequence generation process. This process dynamically generates and optimizes detection nodes through the screening of timestamp information, risk assessment of historical data, calculation of time intervals, and priority sorting, ensuring that the diagnostic sequence can prioritize high-risk data points and improve the pertinence and efficiency of fault detection.

[0086] The determination of the reference timestamp and the reference voltage data is the basic step for generating the diagnosis sequence. The system obtains the reference timestamp information corresponding to the monitoring terminal, which can be set according to the actual application scenario, for example, the system startup time, the last complete diagnosis end time or the user specified reference time. Taking the reference timestamp as the center, the unit voltage data in the voltage data set that meets the first screening condition is selected as the reference voltage data. The first screening condition is usually based on time proximity, for example, selecting the voltage data whose timestamp difference from the reference timestamp is within a preset time range, which can be determined according to the monitoring frequency and data update period of the battery pack, such as ±5 minutes, ±10 minutes, etc. Through this step, the starting node of the diagnosis sequence can be determined, that is, the timestamp corresponding to the reference voltage data, which provides a time reference point for subsequent sequence construction.

[0087] The voltage data set after removing the reference voltage data is taken as the first data set, and the analysis of the historical abnormal record is performed on each data item (i.e. the first data) in the data set. The historical abnormal record group contains the abnormal information of the battery unit corresponding to the first data that occurred in the past diagnosis, such as the time of abnormal occurrence, the type of abnormality (such as voltage drop, current fluctuation amplitude exceeding limit, etc.), the duration of abnormality, etc. The system traverses each first data in the first data set, extracts its corresponding historical abnormal record group, and calculates the mean value of each historical abnormal record in the group as the risk prediction value of the first data. The risk prediction value reflects the tendency of the battery unit to fail in the past, and the calculation method of the mean value can be arithmetic mean or weighted mean (for example, recent abnormal records are given higher weight) to reflect the time correlation of fault risk.

[0088] At the same time of obtaining the risk prediction value, the system needs to calculate the time interval difference between each first data and the reference voltage data to form a time interval set. The calculation of the time interval difference is based on the numerical difference of the timestamp, for example, in minutes, hours, etc., which reflects the time distance between the first data and the reference time point. The size of the time interval difference affects the diagnosis priority of the data to some extent, and in general, the closer to the reference time point, the more the data can reflect the current state of the battery pack, so it needs to be diagnosed first; but for the data with high historical risk, even if the time interval is far away, it may also need to be included in the diagnosis sequence in advance.

[0089] To comprehensively consider the influence of risk prediction value and time interval difference on the diagnostic priority, the system calculates a data priority score for each first data. The calculation method of the priority score is to quantitatively combine the risk prediction value and the time interval difference through a certain mapping relationship, for example, it is set that the higher the risk prediction value and the smaller the time interval difference (i.e. the data is closer to the reference time point), the higher the priority score. In specific implementation, a linear weighting method can be used to normalize the risk prediction value and the time interval difference to the same numerical interval (such as 0-100), and then calculate the weighted sum according to the preset weight (such as the risk prediction value accounts for 60% and the time interval difference accounts for 40%) to obtain the priority score of each first data. In this way, the qualitative risk assessment and quantitative time distance can be converted into comparable numerical indicators, which is convenient for subsequent priority sorting and screening.

[0090] Based on the calculated priority score, the system selects the priority score that meets the second screening condition from all the first data as the target priority score. The second screening condition is usually a preset threshold, for example, the data with a priority score higher than 60 points is regarded as high-priority data, and the timestamp information of the corresponding first data will be determined as the target detection node. After screening the target detection node, the system needs to judge whether the first data set after deleting the first data corresponding to the target priority score is empty: if it is empty, it means that all high-priority data has been processed, at this time the starting node (the timestamp of the reference voltage data) and the target detection node are combined in time sequence to form a complete diagnostic sequence; if it is not empty, the remaining first data set is taken as the new input, the current target detection node is reset as the starting node, and the sequence generation step is executed again. This iterative process will continue until the first data set is empty, ensuring that all data points are included in the diagnostic sequence and high-priority data points are processed first.

[0091] In the iterative process, each time the starting node is reset, the system recalculates the time interval difference between each data item in the remaining first data set and the new starting node, and updates the risk prediction value (if the latest diagnostic result needs to be considered to affect the historical abnormal record). This dynamic adjustment mechanism enables the diagnostic sequence to continuously optimize the generation order of subsequent nodes according to the currently processed data points, ensuring that each iteration can prioritize high-priority points in the current remaining data. For example, if the target detection node selected in a certain iteration corresponds to a battery unit that is found to be abnormal in diagnosis, the neighboring data points of this node in the subsequent iteration process may be assigned a higher priority to further confirm the development trend of the fault.

[0092] It should be noted that the acquisition and updating of historical anomaly records rely on the historical diagnostic data recorded in the storage terminal. After each diagnosis is completed, the system will store the newly generated anomaly marker information to the storage terminal for subsequent sequence generation process calls. Therefore, as the number of diagnoses increases, the historical anomaly records will continue to accumulate, and the calculation of risk prediction values will be more accurate, thereby making the generation of diagnostic sequences more in line with the actual fault risk distribution of the battery pack.

[0093] In specific implementations, the processing of timestamp information needs to consider time synchronization issues. Components such as monitoring terminals and analysis servers need to maintain time synchronization to ensure that the timestamps of various data points are consistent and comparable. Time synchronization can be achieved through the Network Time Protocol (NTP) or other time synchronization mechanisms to avoid errors in the generation of diagnostic sequences caused by time deviations.

[0094] In addition, the generation process of the diagnostic sequence needs to balance computational efficiency and accuracy. For battery packs with large amounts of data, analyzing historical records and calculating priorities for each data item one by one can result in high computational overhead. To optimize performance, the system can use batch processing or utilize a distributed computing framework to parallelize the priority calculation of multiple data items, ensuring that the diagnostic sequence can be generated within a reasonable time without affecting the needs of real-time diagnosis.

[0095] Embodiment 3: This embodiment extends the hardware architecture of the diagnostic device and the processing flow after diagnosis. The diagnostic device includes three core components: a monitoring terminal, an analysis server, and a storage terminal. The components are connected through a data bus or a network to exchange information. After completing the diagnosis step (i.e., the detection node set is empty and the monitoring terminal outputs a diagnosis completion instruction), the system enters the subsequent data transmission, fault location, information storage, and fault evolution analysis flow. The specific implementation is as follows:

[0096] After diagnosis is completed, the monitoring terminal needs to transmit the key data generated during the diagnosis process to the analysis server. Specifically, the monitoring terminal will collect the feature vectors generated by extracting the current fluctuation signal features of each detection node, forming a feature vector set. The feature vector is a multi-dimensional data vector obtained by quantitatively extracting the time domain, frequency domain, or time-frequency domain features of the current fluctuation signal, such as including signal amplitude mean, frequency peak, rising edge time, etc. The monitoring terminal sends the feature vector set to the analysis server through wired or wireless communication links (such as CAN bus, Ethernet, Wi-Fi, etc.), providing data basis for subsequent fault location.

[0097] After receiving the feature vector set, the analysis server performs the fault location operation. The analysis server has a built-in fault location model, which is trained based on historical fault data. It can identify the characteristic patterns related to short-circuit faults from the feature vectors through pattern matching or algorithm calculation, and determine the specific location of the fault or the associated battery cell. For example, the fault location model can be a classifier based on machine learning, which learns the difference in feature vectors when different battery cells are short-circuited through training data, and outputs the corresponding fault location identifier when a new feature vector is input. The fault location identifier can be a unique number of the battery cell, location coordinates, or other information that can identify the fault location. After the analysis server generates the fault location identifier, it returns it to the monitoring terminal or directly transmits it to the storage terminal for recording.

[0098] At the same time, the storage terminal is responsible for recording all abnormal marking information in the current diagnostic cycle. The abnormal marking information corresponds to each detection node one by one, including the timestamp of the node, the corresponding unit voltage data, the generated abnormal marking (such as "normal", "abnormal", "suspected short circuit", etc.) and the corresponding feature vector summary. The storage terminal stores this information in a structured data format (such as database tables, CSV files, etc.) to ensure the queryability and traceability of the data. The storage medium of the storage terminal can be a local hard disk, solid-state drive or cloud storage device, which can be flexibly selected according to the hardware configuration and data management requirements of the diagnostic device.

[0099] When a complete diagnostic cycle is completed (for example, the diagnosis of all preset detection nodes of the target battery pack is completed), the system triggers the fault evolution path analysis process. First, the analysis server extracts the tag information log of the current diagnostic cycle from the storage terminal. The tag information log contains the abnormal tag information and related time series data of all detection nodes in the cycle. Subsequently, the analysis server compares and analyzes the tag information log with the data in the historical log library. The historical log library stores the tag information of multiple past diagnostic cycles of the battery pack and records the changing trend of the battery cell status over time.

[0100] The analysis server determines the fault evolution path by calculating the difference between the marked information log and the historical log library. The difference value calculation can be carried out from multiple dimensions, such as:

[0101] Changes in the number of abnormal nodes: Compare the number of detected nodes marked as abnormal in the current cycle with that in the previous cycle. If the number increases significantly, it may indicate that the scope of the fault has expanded.

[0102] Abnormality type evolution: Analyzes the changes in abnormality marks of the same battery cell in different cycles. For example, the change from "suspected short circuit" to "confirmed short circuit" reflects the increase in fault severity.

[0103] Abnormal node distribution migration: Observe the change of the location distribution of abnormal nodes in the battery pack, and judge whether the fault shows a regional diffusion trend.

[0104] To quantify the difference value, the system can use the following formula:

[0105]

[0106] Where D represents the comprehensive difference value, n represents the number of dimensions participating in comparison (such as the number of abnormal nodes, abnormal type, distribution migration, etc.), ω i represents the weight coefficient of the i-th dimension (the value range is 0-1, and satisfies ), δ i represents the standardized difference value of the i-th dimension (through normalization processing, the difference value of different dimensions is mapped to the interval 0-1). The weight coefficient ω i may be pre-set according to the influence degree of each dimension on fault evolution, for example, the weight of abnormal type evolution is higher than that of abnormal node number change, in order to highlight the importance of fault severity change.

[0107] Based on the comprehensive difference value D and the difference analysis of each dimension, the analysis server constructs the fault evolution path. The fault evolution path describes the process of the battery pack from normal state to short circuit fault occurrence and development, including fault starting position, expansion direction, severity change and other information. For example, if a certain battery unit is marked as abnormal in three consecutive diagnosis cycles, and the abnormal type gradually changes from voltage fluctuation to continuous low voltage, combined with the change of current fluctuation characteristics, it can be inferred that the unit has a gradually deteriorating internal short circuit fault, and the fault evolution path shows the development from local poor electrical contact to substantial short circuit.

[0108] The system needs to judge whether the fault evolution path matches the preset path model. The preset path model is a typical fault development mode induced based on historical fault data, such as "single point fault → regional diffusion" and "gradual deterioration → sudden short circuit". If the match is successful, it means that the current fault development conforms to the known typical mode, and the system can further extract the corresponding feature vector and unit voltage data in the marker information log. Specifically, the analysis server retrieves all feature vectors related to the fault evolution path from the storage terminal, as well as the unit voltage data (including time stamp, voltage value, etc.) corresponding to these feature vectors.

[0109] The system combines the voltage data of each unit with its corresponding abnormality mark information to generate a diagnostic report. The diagnostic report contains information such as the timeline of the fault, the voltage and current characteristics of each stage, the evolution process of the abnormality type and the fault location results, and is presented in the form of a structured document (such as PDF, Word report). The generated diagnostic report is sent to the storage terminal through the analysis server for archiving, so that operation and maintenance personnel can review, analyze and trace the fault. The storage terminal can classify and store the diagnostic report according to index conditions such as diagnostic cycle and battery pack number to facilitate subsequent retrieval and statistical analysis.

[0110] At the hardware interaction level, communication between monitoring terminals, analysis servers, and storage terminals must meet real-time and reliability requirements. For example, feature vector sets can be transmitted using TCP / IP to ensure data integrity, while archiving diagnostic reports can be handled asynchronously to avoid impacting the real-time diagnostic process. The analysis server's computing resources must be configured based on the size of the battery pack and the amount of data, ensuring rapid processing of large-scale labeled information logs and historical data comparison tasks.

[0111] Example 4: This example defines the preconditions and data verification mechanism for fault evolution path analysis to ensure that the data volume of the marking information log meets the reliability requirements before performing fault evolution path analysis, thereby avoiding erroneous analysis due to data loss. The specific implementation method is as follows:

[0112] After completing the processing of all detection nodes in the current diagnostic cycle, the system enters the fault evolution path analysis process. The analysis server needs to extract the tag information log of the current cycle from the storage terminal. The tag information log contains the timestamp of each detection node in the cycle, the corresponding unit voltage data, abnormal tag information and feature vector summary, etc., which is the basic data for analyzing the fault evolution trend. However, if the data volume of the current tag information log is too small, it may not reflect the actual state changes of the battery pack, resulting in unreliable evolution path analysis results. Therefore, before performing difference value calculation and path construction, the system must first verify the validity of the data volume of the tag information log.

[0113] The specific steps for data volume verification are as follows: the system obtains the data volume threshold of the historical log library in the storage terminal. The threshold is usually set based on the statistical characteristics of the historical log library, such as taking the average, median or a certain percentile value (such as the 75% percentile) of the data volume of the marked information logs of multiple diagnostic cycles in the past. The data volume can be measured in the form of the number of log records, data storage capacity (such as the number of bytes), or the number of detection nodes included, etc., which is determined according to the data storage format. For example, if the marked information log of each diagnostic cycle in the historical log library contains an average of 1,000 records, the data volume threshold can be set to 800, that is, the current log data volume must reach at least 80% of the historical average level.

[0114] The system counts the data volume of the current marker information log and compares it with the data volume threshold of the historical log library. If the data volume of the current log is less than the threshold, it indicates that the data volume is insufficient, and the following situations may exist: one is that the monitoring terminal fails to collect data completely within the diagnosis period (such as device failure or communication interruption); two is that the diagnosis period is set too short, resulting in insufficient number of detection nodes collected; three is that data loss or error occurs in the data storage process. Regardless of the reason, insufficient data volume will lead to insufficient sample support for subsequent difference value calculation and fault evolution path analysis, which may lead to misleading conclusions.

[0115] When it is detected that the data volume of the marker information log is less than the threshold, the system generates a data invalidity identifier. The data invalidity identifier is a state marker used to indicate that the current log data is unreliable and needs special processing. This identifier can be included in the metadata of the storage terminal or recorded as an independent field in the marker information log. For example, a "data validity" field is added to the database table, with values of "valid" or "invalid", which is automatically set to "invalid" when the data volume is insufficient.

[0116] After generating the data invalidity identifier, the system updates the identifier in the storage terminal to ensure that the subsequent data query and analysis process can identify the unreliability of the current log. At the same time, the system can trigger an alarm mechanism to send a data anomaly prompt to the operator, such as through an interface popup window, email or SMS notification, to remind the operator to check the working state of the monitoring terminal, whether the communication link is normal, or to adjust the diagnosis period to obtain more complete data.

[0117] In the case of insufficient data volume, the system will not perform the step of determining the fault evolution path based on the difference value between the marker information log and the historical log library. This is because there is insufficient data support, and the difference value calculation may not reflect the true fault trend changes. For example, if only 200 detection node data is collected in the current period, and the average data volume of the historical log library is 1000, the number of abnormal nodes calculated at this time may have a large deviation due to the small sample size, and it cannot be reliably determined whether the fault has spread.

[0118] It should be noted that the setting of the data volume threshold needs to be combined with the monitoring characteristics and diagnosis needs of the battery pack. For high-frequency monitoring of battery packs (such as electric vehicle power batteries, which collect data every minute), the data volume of the historical log library is usually large, and the threshold can be set to a higher absolute value (such as 5000 records); while for low-frequency monitoring of battery packs (such as energy storage station batteries, which collect data every hour), the threshold can be correspondingly reduced (such as 500 records). In addition, the threshold can also be dynamically adjusted according to the number of battery units, for example, each battery unit needs at least N detection data, and the total data volume threshold is the number of units x N, to ensure that the state change of each unit is supported by sufficient data points for analysis.

[0119] At the data storage level, the storage terminal needs to retain the marker information log of all diagnostic cycles, including the log of invalid data, for subsequent tracing and problem troubleshooting. For the log of invalid data, the system can separately establish a storage partition or add a label to distinguish it from valid logs and avoid confusion. At the same time, the storage terminal needs to have data backup and recovery functions to prevent data loss due to hardware failure and affect subsequent diagnosis and analysis.

[0120] When the system detects that the data volume of the marker information log has recovered to above the threshold in the subsequent diagnostic cycle, the previous data invalidity identifier can be automatically cleared, and the normal fault evolution path analysis process is resumed. This dynamic verification mechanism ensures that the system only performs key analysis when there is sufficient data, improving the credibility of the diagnostic results.

[0121] Taking a certain energy storage battery pack as an example, it contains 100 battery units, and 10 data of each unit are collected per diagnostic cycle, with a total data volume of 1000 records. If a communication module fails in a certain cycle and only 300 records are successfully collected, which is significantly lower than the threshold of 800 historical average data volume. At this time, the system generates a data invalidity identifier and records it, does not perform fault evolution path analysis, and prompts the operation and maintenance personnel to check the communication module. After the operation and maintenance personnel repair the module, 1000 complete records are collected in the next cycle, the system verifies that the data volume meets the threshold, clears the invalid identifier, and normally performs difference value calculation and path analysis, thereby avoiding false diagnosis caused by data loss.

[0122] Embodiment 5: This embodiment describes the model optimization mechanism of the diagnostic system. By transmitting the fault evolution path data to the preset database, the historical data is used to train and optimize the diagnostic model, forming a closed-loop feedback mechanism, and improving the recognition ability and prediction accuracy of the system for short circuit faults. The specific implementation is as follows:

[0123] After the analysis server completes the fault evolution path determination, the system triggers the model optimization process. First, the analysis server formats the fault evolution path data of the current record to form a standard data format. The fault evolution path data includes marker information logs, difference value calculation results, fault positioning identifiers, fault evolution path descriptions, and other contents in the diagnostic cycle. For example, a fault evolution path data may record that a battery unit numbered B001 in a certain battery pack gradually developed from "voltage fluctuation anomaly" in the 1st cycle to "confirmed short circuit" in the 3rd cycle, during which the current fluctuation characteristics changed from low frequency and small amplitude to high frequency and large amplitude, and the path type matched the preset "progressive deterioration" mode.

[0124] The formatted path data is transmitted to a preset database through a network or a data bus. The preset database is a structured storage system specially used for storing fault evolution data, which can be deployed in a cloud server or a local data center, and has the functions of large-capacity storage, efficient query and data security protection. The table structure design of the database needs to include the fields of battery pack number, diagnosis cycle time, fault positioning information, evolution path characteristics, marker information log index, etc., so as to facilitate subsequent data retrieval and model training.

[0125] After the preset database receives the path data, the data is cleaned and preprocessed. The preprocessing steps include: removing duplicate records, filling missing fields (such as supplementing part of the missing feature parameters by interpolation method), standardizing feature values (such as converting different dimensional current amplitude and voltage values into normalized values), etc. For example, for the missing voltage value of a certain detection node in a certain path data, the system can perform linear interpolation according to the voltage data at the previous and subsequent time points to ensure the integrity of the data.

[0126] After the preprocessing is completed, the system starts the model optimization algorithm. The model optimization algorithm is based on machine learning techniques, such as using neural networks, random forests or clustering algorithms, to learn from historical path data in the preset database. Taking the fault positioning model as an example, the optimization process of the model is as follows: first, extract the feature vector set and its corresponding actual fault positioning result (such as the known fault battery unit number) in the historical path data from the database as the training data set; then, adjust the model parameters through iterative training, so that the model can more accurately output the fault positioning identifier when inputting new feature vectors. For the optimization of the preset path model, the mode distribution of the historical path data is analyzed, and the types and feature parameters of the existing path model are added or adjusted, such as identifying a new "multi-point synchronous deterioration" fault mode and including it in the preset model library.

[0127] The triggering mechanism of model optimization can be divided into timing triggering and data volume triggering. Timing triggering refers to that the system automatically starts the optimization process according to the preset time interval (such as every day at 24:00, every Sunday morning), which is suitable for scenarios where data accumulation speed is relatively uniform; data volume triggering refers to that when the newly added path data in the preset database reaches a certain number (such as 100, 500), the optimization is automatically triggered, which is suitable for scenarios where data accumulates rapidly. For example, a battery monitoring system of an electric vehicle generates a large amount of diagnosis data every day, and the model optimization can be triggered once every 200 newly added path data, so as to ensure that the model absorbs the latest data features in time.

[0128] During the model optimization process, efficient data interaction between the analysis server and the preset database is required. The analysis server reads the training data set from the database, performs algorithm calculation, and returns the optimized model parameters to the database for storage. At the same time, the system needs to retain the old version of the model parameters to perform rollback when the performance of the new version of the model does not meet the standards, ensuring the stability of the diagnostic system.

[0129] Taking a large-scale energy storage battery pack containing 1000 battery units as an example, it may generate multiple failure modes during operation. Assuming that the preset database has stored 1000 historical failure evolution paths, 70% of which are single-point progressive deterioration failures, 20% are regional diffusion failures, and 10% are sudden short-circuit failures. When the system detects a new path data showing that 3 adjacent battery units in a certain region simultaneously exhibit abnormalities, and the abnormal characteristics have similarities, the analysis server will mark this path as a "multi-point synchronous abnormality" type and store it in the preset database. After a certain amount of similar data accumulation, the model optimization algorithm can identify the common characteristics of this type of data (such as the proximity of abnormal node spatial location, the same frequency component in the feature vector), thereby adding a "multi-point synchronous deterioration" mode in the preset path model, and adjusting the input features of the fault location model to include spatial location correlation as a calculation factor, improving the prediction ability of similar faults.

[0130] After the model optimization is completed, the analysis server pushes the updated model parameters to each component of the diagnostic device. For example, the monitoring terminal can adjust the key parameters for feature extraction in real time based on the optimized fault location model when collecting data; the analysis server uses the new preset path model for matching when performing the diagnosis step, improving the accuracy of fault evolution path judgment. This closed-loop mechanism enables the diagnostic system to continuously accumulate fault data as the running time increases, automatically optimize the model performance, and gradually improve the recognition ability of new types of faults and complex faults.

[0131] It should be noted that the model optimization process needs to follow the principles of data security and privacy protection. The fault data in the preset database may contain sensitive information such as the running state and location information of the battery pack, and the system needs to ensure data security through encryption transmission, access control, data desensitization, and other technical means. For example, when transmitting path data, use SSL / TLS encryption protocol, and hash the battery pack location information in the database to avoid sensitive information leakage.

[0132] In addition, the effect of model optimization can be evaluated by an offline verification mechanism. The system periodically extracts a part of historical data that does not participate in training from the preset database, uses the optimized model for fault positioning and path matching, calculates the accuracy, recall rate and other indicators to ensure that the improvement of model performance has actual effectiveness. If the evaluation finds that the model performance has decreased, the system automatically loads the last version of the model parameters and investigates the current optimization process to find out the cause of the performance decline (such as data noise, improper algorithm parameter setting, etc.) for targeted adjustment.

[0133] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0134] While embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, combinations, and variations can be undertaken by those skilled in the art without departing from the spirit and scope of the present application, which is defined by the following claims and their equivalents.

Claims

1. A battery short circuit fault intelligent diagnosis method, applied to a diagnostic device, wherein the diagnostic device includes a monitoring terminal, characterized in that: The method comprises: Acquire a voltage data set of a plurality of battery cells corresponding to a target battery pack, wherein the cell voltage data in the voltage data set includes timestamp information; generating a diagnostic sequence according to each time stamp information included in the voltage data set, wherein the diagnostic sequence includes a plurality of detection nodes, and the number of detection nodes included in the diagnostic sequence is the same as the number of unit voltage data included in the voltage data set; Based on the set of detection nodes included in the diagnostic sequence, the following diagnostic steps are performed: Controlling the monitoring terminal to collect voltage data corresponding to timestamp information of a first detection node in the detection node set; In response to capturing the current fluctuation signal, performing feature extraction processing on the current fluctuation signal to generate a feature vector; determining abnormality marking information of the unit voltage data corresponding to the detection node according to the characteristic vector; In response to the detection node set after the first detection node is deleted being empty, controlling the monitoring terminal to output a diagnosis completion instruction; In response to the detection node set of deleting the first detection node being not empty and the monitoring terminal meeting the preset diagnostic conditions, the monitoring terminal is controlled to output a short circuit alarm instruction, and in response to detecting an alarm confirmation operation, the detection node set of deleting the first detection node is used as the detection node set, and the diagnostic step is executed again.

2. The battery short circuit fault intelligent diagnosis method according to claim 1, characterized in that: The method further comprises: In response to the detection node set from which the first detection node is deleted being not empty and the monitoring terminal not satisfying the preset diagnosis condition, the detection node set from which the first detection node is deleted is used as the detection node set, and the diagnosis step is performed again.

3. The battery short circuit fault intelligent diagnosis method according to claim 1, characterized in that: Generating a diagnostic sequence according to each time stamp information included in the voltage data set includes: Obtaining reference timestamp information corresponding to the monitoring terminal; According to the reference timestamp information, filter out unit voltage data that meets a first filter condition from the voltage data set as reference voltage data, and determine the timestamp information included in the reference voltage data as a starting node; The voltage data set without the reference voltage data is used as a first data set, and the following sequence generation steps are performed based on the first data set: Obtaining a historical abnormality record group of each first data in the first data set to obtain a historical record group set; For each first data in the first data set, determining the average of each historical abnormal record included in the historical abnormal record set corresponding to the first data as a risk prediction value; determining a time interval difference between each first data in the first data set and the reference voltage data to obtain a time interval set; For each first data in the first data set, determining a data priority score according to a risk prediction value and a time interval difference corresponding to the first data; Selecting a priority score that meets a second screening condition from the determined data priority scores as a target priority score, and determining timestamp information included in the first data corresponding to the target priority score as a target detection node; In response to the first data set from which the first data corresponding to the target priority score is deleted being empty, combining the start node and the target detection node into a diagnosis sequence; In response to the first data set from which the first data corresponding to the target priority score is deleted being not empty, the first data set from which the first data corresponding to the target priority score is deleted is used as the first data set, the target detection node is determined as the starting node, and the sequence generation step is executed again.

4. The battery short circuit fault intelligent diagnosis method according to claim 1, characterized in that: The diagnostic device also includes an analysis server and a storage terminal; Furthermore, after controlling the monitoring terminal to output a diagnosis completion instruction in response to the detection node set being empty after deleting the first detection node, the method further includes: Controlling the monitoring terminal to send a feature vector set to the analysis server; receiving a fault location identifier returned by the analysis server; Controlling the storage terminal to record all abnormal marking information corresponding to the current diagnosis cycle; In response to the end of the diagnosis cycle, controlling the analysis server to extract the tag information log of the storage terminal; Determining a fault evolution path based on a difference between the tag information log and the historical log library; In response to the fault evolution path matching a preset path model, obtaining at least one unit voltage data included in at least one feature vector corresponding to the mark information log; Each unit voltage data in the at least one unit voltage data and the corresponding abnormality mark information are combined into a diagnosis report, and the diagnosis report is sent to the storage terminal for archiving.

5. The battery short circuit fault intelligent diagnosis method according to claim 4, characterized in that: Before determining the fault evolution path based on the difference between the mark information log and the historical log library, the method further includes: In response to the data volume of the mark information log being smaller than the data volume threshold of the history log library, a data invalidation flag is generated, and the data invalidation flag is updated at the storage terminal.

6. The intelligent diagnosis method for battery short circuit fault according to claim 4, characterized in that: The method further comprises: The analysis server is controlled to transmit the recorded fault evolution path to a preset database to perform model optimization processing on the fault evolution path.

7. A battery short circuit fault intelligent diagnosis system, comprising: a collection unit configured to obtain a voltage data set of a plurality of battery cells corresponding to a target battery pack, wherein the cell voltage data in the voltage data set includes timestamp information; a sequence construction unit configured to generate a diagnostic sequence according to each time stamp information included in the voltage data set, wherein the diagnostic sequence includes a plurality of detection nodes, and the number of detection nodes included in the diagnostic sequence is the same as the number of unit voltage data included in the voltage data set; The diagnostic execution unit is configured to perform the following diagnostic steps based on the detection node set included in the diagnostic sequence: controlling the monitoring terminal to collect voltage data corresponding to the timestamp information of the first detection node in the detection node set; in response to capturing a current fluctuation signal, performing feature extraction processing on the current fluctuation signal to generate a feature vector; determining the abnormal mark information of the unit voltage data corresponding to the detection node based on the feature vector; in response to the detection node set from which the first detection node is deleted being empty, controlling the monitoring terminal to output a diagnostic completion instruction; in response to the detection node set from which the first detection node is deleted being not empty and the monitoring terminal meeting a preset diagnostic condition, controlling the monitoring terminal to output a short-circuit alarm instruction; and in response to detecting an alarm confirmation operation, using the detection node set from which the first detection node is deleted as the detection node set, and performing the diagnostic steps again.

8. The battery short circuit fault intelligent diagnosis system according to claim 7, wherein: The sequence building unit also includes: The system is configured to call different time stamp sorting algorithms according to the battery pack type code to adjust the node generation frequency of the diagnostic sequence.

9. A device comprising: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

10. A medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images