A method and system for anomaly diagnosis and source tracing of battery test data

Abstract

This invention relates to the field of battery testing technology, and discloses a method and system for anomaly diagnosis and source tracing of battery testing data. The method includes: normalizing real-time battery testing data to obtain a preprocessed dataset; comparing the real-time data distribution of the preprocessed dataset with a pre-stored historical data baseline distribution and historical baseline threshold to obtain an adaptive detection threshold for the preprocessed dataset; using the adaptive detection threshold to determine anomalies in the preprocessed dataset to obtain anomalous data segments; tracing changes in related parameters from the battery's historical operating states based on the anomalous data segments, and constructing an anomaly propagation chain based on the temporal relationship of the changes. This invention can improve the accuracy of diagnostic results, the depth of source tracing analysis, and the operability of operation and maintenance guidance.

Description

Technical Field

[0001] This invention relates to the field of battery testing technology, and in particular to a method and system for anomaly diagnosis and tracing of battery testing data. Background Technology

[0002] With the rapid development of electric vehicles and large-scale energy storage systems, the safe operation and condition management of power sources such as lithium-ion batteries have become critical challenges. During use, the complex electrochemical system inside a battery may experience a series of parameter anomalies under the combined effects of external stress and aging. If these anomalies are not diagnosed and accurately traced in a timely manner, they can easily evolve from a single parameter deviation into catastrophic failures such as thermal runaway. Therefore, real-time or near-real-time anomaly diagnosis of battery testing data, and further tracing the root causes and evolution paths of these anomalies, are of paramount importance for achieving predictive maintenance and proactive safety control of battery systems.

[0003] Currently, existing methods for diagnosing and tracing battery data anomalies still have several limitations in practical applications. First, in the anomaly detection stage, most methods rely on fixed thresholds or simple statistical control limits preset based on historical experience, making it difficult to adapt to the dynamically fluctuating normal operating boundaries of batteries under different operating conditions and health states, leading to frequent false alarms and missed alarms. Second, after anomaly identification, existing technologies are often limited to describing and alarming the anomaly segment itself, or only performing simple parameter comparisons. Ultimately, this makes it difficult for maintenance personnel to pinpoint the initial source of the anomaly from a temporal perspective, and also prevents them from understanding the specific transmission mechanism of the anomaly between parameters within the battery system. This limits the depth and operability of diagnostic conclusions, making it difficult to support accurate maintenance decisions and risk warnings. Summary of the Invention

[0004] This invention provides a method and system for anomaly diagnosis and tracing of battery testing data to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for anomaly diagnosis and source tracing of battery testing data, the method comprising:

[0006] S1. Normalize the real-time detection data of the battery to obtain the preprocessed dataset of the real-time detection data;

[0007] S2. Based on the pre-stored historical data baseline distribution and historical baseline threshold, the real-time data distribution of the preprocessed dataset is compared to obtain the adaptive detection threshold of the preprocessed dataset.

[0008] S3. Use the adaptive detection threshold to determine the abnormal state of the preprocessed dataset to obtain abnormal data fragments of the preprocessed dataset.

[0009] S4. Based on the abnormal data fragments, trace the change events of related parameters from the battery's historical operating states, and construct an abnormal propagation chain according to the temporal relationship of the change events.

[0010] S5. Perform reverse traversal of the abnormal propagation chain to locate the time corresponding to the parameter event at the beginning of the abnormal propagation chain, and obtain the time of the abnormal origin.

[0011] S6. Integrate the battery state parameters, working environment data, and the propagation logic of the abnormality transmission chain corresponding to the time of the abnormality origin to obtain a diagnostic report of the battery detection data.

[0012] Preferably, comparing the real-time data distribution of the preprocessed dataset based on the pre-stored historical data benchmark distribution and historical benchmark threshold includes:

[0013] Real-time features of battery voltage, current, and temperature data are extracted from the preprocessed dataset.

[0014] From the pre-stored historical data benchmark distribution, obtain the benchmark features of the corresponding voltage, current and temperature data;

[0015] Calculate the volatility stability index of the data parameters in the historical data benchmark distribution, and assign weight factors to the data parameters according to the volatility stability index to obtain the weight vector of the data parameters;

[0016] The weight vector is used to weight and synthesize the real-time feature and the baseline feature respectively, to obtain a weighted real-time feature vector and a weighted baseline feature vector;

[0017] Distance quantization is performed between the weighted real-time feature vector and the weighted baseline feature vector to obtain the distribution difference between the two.

[0018] The historical baseline threshold and the distribution difference are linearly combined, and the combined result is used as the adaptive detection threshold.

[0019] The formula for calculating the volatility stability index is as follows:

[0020] ;

[0021] In the formula, Indicates the first The volatility stability index of each data parameter, the These correspond to voltage, current, and temperature, respectively. This represents the first [number] calculated based on the historical data baseline distribution. The standard deviation of each data parameter; It is a pre-defined, extremely small positive number.

[0022] Preferably, the step of using the adaptive detection threshold to determine the abnormal state of the preprocessed dataset includes:

[0023] Based on the adaptive detection threshold, continuous data points in the preprocessed dataset are labeled with their states, and data points exceeding the adaptive detection threshold are labeled as candidate outliers.

[0024] The candidate anomalies are continuously verified, and the set of candidate anomalies that appear continuously for a duration exceeding a preset window is determined as a valid anomaly state.

[0025] Based on the start and end points of the valid abnormal states on the time axis, the corresponding data sequences are extracted from the preprocessed dataset to obtain the abnormal data segments.

[0026] Preferably, the step of tracing changes in related parameters from the battery's historical operating states based on the abnormal data fragments, and constructing an anomaly propagation chain according to the temporal relationship of the changes, includes:

[0027] The abnormal data segment is decoupled from its parameters to obtain the set of key coupling parameters for the abnormal data segment.

[0028] Starting from the time when the abnormal data segment occurred, the time when each parameter in the key coupling parameter set first showed a statistically significant deviation is traced back in the battery's historical operating state to obtain a set of parameter event nodes;

[0029] A causal association analysis was performed on all the parameter event nodes to obtain the causal driving relationship between the parameter events;

[0030] By linking the parameter event nodes with the causal driving relationship, a causal event sequence of the abnormal data fragment is obtained;

[0031] The causal event sequence is integrated in chronological order to obtain the anomaly propagation chain.

[0032] Preferably, the parameter decoupling of the abnormal data fragment includes:

[0033] Extract standardized data sequences of battery voltage, current, and temperature parameters during the abnormal period from the abnormal data fragments;

[0034] Correlation analysis is performed on the parameters in the standardized data sequence to obtain information on the correlation strength between the parameters in the standardized data sequence;

[0035] Based on the association strength information, strongly associated parameter pairs exceeding a preset association threshold are identified;

[0036] The key coupling parameter set is obtained by combining the parameters involved in the strongly correlated parameters through a union.

[0037] Preferably, the causal correlation analysis of all the parameter event nodes includes:

[0038] Calculate the transition entropy between the data sequences corresponding to the parameter event nodes, and construct the causal strength matrix;

[0039] Each element in the causal intensity matrix is ​​compared with a preset causal threshold, and the transition entropy value that exceeds the causal threshold is recorded as the determined entropy value;

[0040] Establish the causal driving relationship between the parameter events based on the determined entropy value;

[0041] The formula for calculating the transfer entropy is as follows:

[0042] ;

[0043] in, Indicates the event from the parameter. arrive The transfer entropy value, and These represent two different parameter events. Indicates parameter event exist Before the moment Historical data sequences at various points in time. Indicates parameter event exist Before the moment Historical data sequences at various points in time. Indicates parameter event exist Future data values ​​at any given time This represents the joint probability distribution or conditional probability distribution of the corresponding event sequence.

[0044] Preferably, the step of reverse traversing the anomaly propagation chain to locate the time corresponding to the parameter event at the starting end of the anomaly propagation chain includes:

[0045] The anomaly propagation chain is parsed as a directed acyclic graph structure based on parameter events and causal driving relationships;

[0046] Perform a reverse topological sort on the directed acyclic graph structure to obtain the reverse event sequence of the directed acyclic graph structure;

[0047] Extract the parameter event node located at the end of the reverse event sequence and obtain its corresponding timestamp, and determine the timestamp as the time of the anomaly origin.

[0048] Preferably, the step of resolving the anomaly propagation chain into a directed acyclic graph structure based on parameter events and causal driving relationships includes:

[0049] Define each parameter event in the anomaly propagation chain as a graph node, and assign the graph node its corresponding parameter type and timestamp attribute;

[0050] Based on the causal driving relationship, a directed edge is established between two graph nodes that have a causal relationship, wherein the direction of the directed edge is from the parameter event node representing the cause to the parameter event node representing the result.

[0051] The directed edges are subjected to topological consistency verification to obtain the directed acyclic graph structure.

[0052] Preferably, the step of integrating the battery state parameters, operating environment data, and the propagation logic of the anomaly propagation chain corresponding to the anomaly origin time to obtain a diagnostic report of the battery detection data includes:

[0053] Extract the battery state parameters and operating environment data corresponding to the time of the anomaly origin;

[0054] Semantic parsing is performed on the anomaly propagation chain to obtain a description of the propagation path of the anomaly propagation chain;

[0055] The diagnostic report is obtained by structurally integrating the battery status parameters, operating environment data, and propagation path description.

[0056] To address the aforementioned problems, this invention also provides a system for anomaly diagnosis and tracing of battery testing data, the system comprising:

[0057] The preprocessing module is used to normalize the real-time detection data of the battery to obtain the preprocessed dataset of the real-time detection data.

[0058] The data comparison module is used to compare the real-time data distribution of the preprocessed dataset with the pre-stored historical data benchmark distribution to obtain the adaptive detection threshold of the preprocessed dataset.

[0059] An anomaly detection module is used to determine the anomaly state of the preprocessed dataset using the adaptive detection threshold, and to obtain the abnormal data fragments of the preprocessed dataset.

[0060] An anomaly propagation chain generation module is used to trace the change events of related parameters from the battery's historical operating state based on the anomaly data fragments, and to construct an anomaly propagation chain according to the temporal relationship of the occurrence of the change events.

[0061] The source tracing analysis module is used to traverse the anomaly propagation chain in reverse, locate the time corresponding to the parameter event at the beginning of the anomaly propagation chain, and obtain the anomaly origin time.

[0062] The diagnostic report generation module is used to integrate the battery state parameters, operating environment data, and the propagation logic of the abnormality transmission chain corresponding to the time of the abnormality origin to obtain a diagnostic report of the battery detection data.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. By constructing a complete technology chain from real-time detection data perception to the generation of in-depth diagnostic reports, a systematic improvement in battery anomaly management efficiency has been achieved. Adaptively generating precise detection thresholds significantly enhances the sensitivity and accuracy of identifying anomalies under different operating conditions. By deeply tracing the correlations of anomaly events and constructing a clear transmission chain, the root cause of the anomaly can be precisely pinpointed, thus providing the battery system with a deep insight from the anomaly's symptoms to its underlying causes. The resulting diagnostic report comprehensively reflects the anomaly's origin, state background, and propagation logic, providing a comprehensive and reliable basis for operational and maintenance decisions.

[0065] 2. By introducing dynamic weight allocation and continuous verification mechanisms, the robustness and reliability of the anomaly detection process are further optimized. At the anomaly tracing level, the dynamic propagation mechanism of anomalies within the battery system is profoundly revealed through the analysis of multi-parameter coupling relationships and the quantitative analysis of causal driving relationships. Structured analysis and reverse traversal of the anomaly propagation chain ensure the rigor of the root cause localization process and the uniqueness of the conclusions. The entire technical solution is interconnected, collectively improving the accuracy of diagnostic results, the depth of tracing analysis, and the operability of operational guidance. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating a method for anomaly diagnosis and tracing of battery test data according to an embodiment of the present invention.

[0067] Figure 2 This is a functional block diagram of a battery testing data anomaly diagnosis and tracing system provided in an embodiment of the present invention;

[0068] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0069] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0070] This application provides a method for anomaly diagnosis and tracing of battery testing data. The execution subject of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for anomaly diagnosis and tracing of battery testing data can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0071] Reference Figure 1 The diagram shown is a flowchart illustrating an anomaly diagnosis and tracing method for battery testing data provided in an embodiment of the present invention.

[0072] In this embodiment, the method for anomaly diagnosis and tracing of battery test data includes:

[0073] S1. Normalize the real-time detection data of the battery to obtain the preprocessed dataset of the real-time detection data;

[0074] S2. Based on the pre-stored historical data baseline distribution and historical baseline threshold, compare the real-time data distribution of the preprocessed dataset to obtain the adaptive detection threshold of the preprocessed dataset.

[0075] S3. Use the adaptive detection threshold to determine the abnormal state of the preprocessed dataset and obtain the abnormal data fragments of the preprocessed dataset.

[0076] S4. Based on abnormal data fragments, trace the change events of related parameters from the battery's historical operating state, and construct the abnormal propagation chain according to the temporal relationship of the change events.

[0077] S5. Perform reverse traversal of the anomaly propagation chain, locate the time corresponding to the parameter event at the beginning of the anomaly propagation chain, and obtain the time of the anomaly origin.

[0078] S6. Integrate the battery state parameters, operating environment data, and the propagation logic of the anomaly propagation chain corresponding to the time of anomaly origin to obtain a diagnostic report of battery detection data.

[0079] In this embodiment of the invention, the real-time data distribution of the preprocessed dataset is compared based on a pre-stored historical data baseline distribution and a historical baseline threshold, including:

[0080] Real-time features of battery voltage, current, and temperature data are extracted from the preprocessed dataset;

[0081] The baseline characteristics of the corresponding voltage, current and temperature data are obtained from the pre-stored historical data baseline distribution;

[0082] Calculate the volatility stability index of data parameters in the historical data benchmark distribution, and assign weight factors to the data parameters according to the volatility stability index to obtain the weight vector of the data parameters.

[0083] The real-time features and the baseline features are weighted and synthesized using weight vectors to obtain weighted real-time feature vectors and weighted baseline feature vectors.

[0084] Distance quantization is performed between the weighted real-time feature vector and the weighted baseline feature vector to obtain the distribution difference between the two.

[0085] The historical baseline threshold and the distribution difference are linearly combined, and the combined result is used as the adaptive detection threshold.

[0086] The formula for calculating the volatility stability index is as follows:

[0087] ;

[0088] In the formula, Indicates the first The volatility stability index of each data parameter These correspond to voltage, current, and temperature, respectively. This represents the first value calculated based on the historical data baseline distribution. The standard deviation of each data parameter; It is a pre-defined, extremely small positive number.

[0089] Specifically, battery voltage, current, and temperature data are extracted from the preprocessed dataset according to a preset time window. The mean, maximum, and minimum values ​​of each parameter within the window are calculated, and these statistical results are used as the corresponding real-time features.

[0090] Furthermore, voltage, current, and temperature data in the same scenario as the real-time data are selected from the pre-stored historical data benchmark distribution. The mean, maximum, and minimum values ​​of these historical data are also calculated, and these results are used as the corresponding benchmark features.

[0091] Furthermore, the fluctuation ranges of voltage, current, and temperature in the historical data benchmark distribution are statistically analyzed, that is, the difference between the maximum and minimum values ​​of each parameter. The smaller the fluctuation range, the higher the fluctuation stability index of the parameter. Weight factors are assigned to each parameter according to the index size, and the parameter with the highest fluctuation stability index has the largest weight factor, thus forming a weight vector containing the weights of the three parameters.

[0092] Furthermore, each weight in the weight vector is multiplied by the mean, maximum, and minimum values ​​of the corresponding parameters in the real-time features, and then summed to obtain the weighted real-time feature vector; in the same way, the weight vector is multiplied by the statistical results of the corresponding parameters in the benchmark features, and then summed to obtain the weighted benchmark feature vector.

[0093] Furthermore, the difference between the corresponding values ​​in the weighted real-time feature vector and the weighted baseline feature vector is calculated, and the absolute values ​​of all differences are added together. The sum obtained is the distribution difference between the two.

[0094] Finally, the historical baseline threshold and the distribution variability are added together in the same proportion, and the result is used as the adaptive detection threshold.

[0095] It should be noted that the fluctuation stability index is derived from calculations of three data parameters: voltage, current, and temperature. The index is based on the historical data baseline distribution. The standard deviation of a data parameter is calculated by first calculating the mean of all data in the historical data baseline distribution, then calculating the difference between each data point and the mean, squaring all the differences and then taking the average, and finally taking the square root of this average. The preset minimum positive number is a fixed value set in advance to avoid the denominator being zero during the calculation.

[0096] Specifically, the meaning of this formula is to quantify the first... The stability of a data parameter's fluctuations within a historical data baseline distribution is calculated by combining the parameter's historical standard deviation with a preset minimum positive number. This directly yields a fluctuation stability index that reflects the parameter's stable fluctuation state, providing a basis for subsequently assigning weighting factors to the data parameter.

[0097] Specifically, when the first The larger the historical standard deviation of a data parameter, the more drastic the fluctuation of that parameter in the historical data baseline distribution. The larger the denominator after substituting it into the formula, the smaller the value of the calculated volatility stability index. When the historical standard deviation is smaller, the more gradual the fluctuation of that parameter in the historical data baseline distribution. The smaller the denominator, the larger the value of the volatility stability index. The preset minimum positive number will not change this overall trend of change with the standard deviation.

[0098] In summary, traditional methods rely on fixed thresholds, which are difficult to adapt to the normal boundaries of batteries under different operating conditions and health states, and are prone to false alarms and missed alarms. This solution combines historical data benchmark distribution with real-time distribution comparison, and the generated adaptive detection threshold can be dynamically adjusted to fit the actual operating conditions of the battery and reduce false alarms and missed alarms. In addition, by allocating weights through fluctuation stability index, more weight is given to key parameters with stable historical fluctuations, so that the weighted feature vector can better reflect the true state, the quantified distribution difference is more valuable for reference, and interference from secondary parameters is avoided.

[0099] In summary, by integrating historical benchmark thresholds with real-time distribution differences, the basic safety boundary is anchored by historical experience while responding to real-time state deviations. This avoids excessive threshold fluctuations or inability to cope with special operating conditions, providing an accurate benchmark for subsequent anomaly judgment, ensuring that real anomalies can be identified, and guaranteeing the reliability of the "preprocessing-anomaly identification" chain.

[0100] In this embodiment of the invention, anomaly detection is performed on the preprocessed dataset using an adaptive detection threshold, including:

[0101] Based on the adaptive detection threshold, continuous data points in the preprocessed dataset are labeled with their states, and data points that exceed the adaptive detection threshold are labeled as candidate outliers.

[0102] Continuous verification of candidate anomalies is performed, and the set of candidate anomalies that appear continuously for a duration exceeding a preset window is determined as a valid anomaly state;

[0103] Based on the start and end points of valid abnormal states on the time axis, the corresponding data sequences are extracted from the preprocessed dataset to obtain abnormal data fragments.

[0104] Specifically, the continuous data points in the preprocessed dataset are read one by one, and the value of each data point is directly compared with the adaptive detection threshold. If the value of a data point exceeds the adaptive detection threshold, the data point is directly marked as a candidate outlier.

[0105] Furthermore, the order of appearance and duration of all candidate anomalies on the timeline are recorded. Using a preset fixed time window as the standard, it is determined whether the candidate anomalies appear continuously. If the continuous duration of a certain segment of candidate anomalies exceeds the preset time window, then the set of candidate anomalies in that segment is determined to be a valid anomaly.

[0106] Furthermore, the start and end times of valid abnormal states on the timeline are clearly defined. Based on these two times, all data within the time interval are precisely extracted from the preprocessed dataset to form a complete data sequence, which is the abnormal data segment.

[0107] In summary, this solution continuously verifies candidate anomalies, only identifying points that continuously exceed a preset window as valid anomalies, thus eliminating false signals from instantaneous data fluctuations and avoiding the misjudgment problem of the traditional "single point exceeding the threshold is considered an anomaly"; furthermore, by extracting data sequences according to the start and end points of the timeline of valid anomalies, complete anomaly data segments are obtained, and irrelevant data is removed, providing clear targets for subsequent tracing of correlation parameters and construction of anomaly propagation chains, and reducing interference from invalid data.

[0108] In this embodiment of the invention, the method of tracing changes in associated parameters from the historical operating states of the battery based on abnormal data fragments, and constructing an anomaly propagation chain according to the temporal relationship of the changes, includes:

[0109] Decouple the abnormal data segments by parameters to obtain the key coupling parameter set of the abnormal data segments;

[0110] Starting from the moment when the abnormal data segment occurs, the moment when each parameter in the key coupling parameter set first shows a statistically significant deviation is traced back in the battery's historical operating state to obtain a set of parameter event nodes;

[0111] Perform causal correlation analysis on all parameter event nodes to obtain the causal driving relationship between parameter events;

[0112] By linking parameter event nodes with causal driving relationships, a causal event sequence of abnormal data fragments is obtained;

[0113] By integrating the causal event sequence in chronological order, an anomaly propagation chain is obtained.

[0114] In this embodiment of the invention, parameter decoupling of abnormal data segments includes:

[0115] Extract standardized data sequences of battery voltage, current, and temperature parameters during abnormal periods from abnormal data fragments;

[0116] Correlation analysis is performed on parameters in a standardized data sequence to obtain information on the strength of the correlation between parameters in the standardized data sequence;

[0117] Strongly correlated parameter pairs exceeding a preset correlation threshold are identified based on correlation strength information;

[0118] The set of key coupling parameters is obtained by combining the strongly correlated parameters into a union.

[0119] In this embodiment of the invention, causal correlation analysis is performed on all parameter event nodes, including:

[0120] Calculate the transition entropy between the data sequences corresponding to the parameter event nodes, and construct the causal strength matrix;

[0121] Each element in the causal intensity matrix is ​​compared with a preset causal threshold, and the transition entropy value that exceeds the causal threshold is recorded as the determined entropy value.

[0122] Establish causal driving relationships between parameter events based on determined entropy values;

[0123] The formula for calculating the transfer entropy is:

[0124] ;

[0125] in, Indicates the event from the parameter. arrive The transfer entropy value, and These represent two different parameter events. Indicates parameter event exist Before the moment Historical data sequences at various points in time. Indicates parameter event Before the moment Historical data sequences at various points in time. Indicates parameter event exist Future data values ​​at any given time This represents the joint probability distribution or conditional probability distribution of the corresponding event sequence.

[0126] Specifically, the abnormal time period corresponding to the abnormal data segment is identified, and all raw data of battery voltage, current and temperature within that time period are extracted. The raw data of each parameter is divided by the mean value of that parameter under normal working conditions, and the processed data is arranged in chronological order to form a standardized data sequence for each parameter.

[0127] Furthermore, by comparing the variation patterns of voltage and current, voltage and temperature, and current and temperature in the standardized data sequence, when the data of one parameter rises or falls, the number of times and the magnitude of the synchronous change of the other parameter are counted. The more times they are synchronized and the closer the magnitudes are, the stronger the correlation between the two. Based on this, the correlation strength information between each pair of parameters is obtained.

[0128] Furthermore, the correlation strength information of each set of parameters is directly compared with the preset correlation threshold. If the correlation strength information of a certain set of parameters exceeds the preset correlation threshold, then the set of parameters is determined to be a strongly correlated parameter pair.

[0129] Furthermore, all parameters contained in all strongly correlated parameter pairs are collected, duplicate parameters are removed, and the remaining unique parameters are summarized to form a set of key coupling parameters.

[0130] Furthermore, starting from the moment the abnormal data fragment occurred, the historical working status records of the battery were reviewed, and the historical data of each parameter in the key coupling parameter set was compared with the normal fluctuation range. The specific time when each parameter first deviated significantly from the normal range and met the statistical significance standard was found, and this time point was determined as the parameter event node.

[0131] Furthermore, the influence of the data sequence corresponding to each parameter event node on the data sequences of other nodes is analyzed. By judging whether the change of the previous parameter can predict the change of the next parameter, this influence is quantified and a matrix is ​​formed. This matrix is ​​the causal strength matrix.

[0132] Furthermore, the value at each position in the causal intensity matrix is ​​compared with a preset causal threshold. If a value exceeds the preset causal threshold, that value is recorded as a determined entropy value.

[0133] Furthermore, based on the parameter event nodes corresponding to the determined entropy value, it is clarified that the preceding parameter event that generates the determined entropy value is the cause of the following parameter event, thereby establishing a causal driving relationship between parameter events.

[0134] Furthermore, based on the causal driving relationship, parameter event nodes with direct causal connections are sequentially connected according to the logic of "cause event - result event" to form a causal event sequence that reflects the abnormal transmission relationship of parameters.

[0135] Finally, the causal event sequence is arranged and integrated according to the chronological order of the occurrence of each parameter event node. Starting from the earliest parameter abnormal event node, subsequent related event nodes are connected in sequence to form a complete abnormal transmission chain.

[0136] It should also be noted that the two different parameter events involved in the transfer entropy value and All of these originate from previously defined parameter event nodes. exist Before the moment The historical data sequence at each time point is a continuous data set of the parameter event extracted from the battery's historical operating state within the corresponding time period, arranged in chronological order. exist Before the moment Similarly, for historical data sequences at a given time point, parameter events... exist The future data value at a given time is the parameter that immediately follows the event. The actual data at the next time point after the given time is obtained by the joint probability distribution, which is the proportion of the number of times three data sequences appear simultaneously to the total number of times. The conditional probability distribution is the proportion of the number of times the target data value or data sequence appears given that a specific data sequence has appeared.

[0137] It should be noted that the meaning of this formula is to quantify the parameter events. To parameter event The strength of causal influence is determined by combining historical data sequences of the two parameter events. Calculate the future data values ​​and their corresponding joint and conditional probability distributions to directly obtain results that reflect... right The transfer entropy value, which generates the degree of causal driving effect, provides a basis for subsequently establishing the causal driving relationship between parameter events.

[0138] It should be noted that when the parameter event... Historical data sequence for parameter events The stronger the predictive power of future data values, that is, the more it combines... Historical data The greater the improvement in the accuracy of future data predictions compared to predictions relying solely on historical data of Q, the larger the calculated transfer entropy value; conversely, when... Historical data Predicting future data offers little help; the smaller the improvement in prediction accuracy, the smaller the transfer entropy value.

[0139] In summary, traditional methods can only describe abnormal segments or simple parameter comparisons, and cannot trace the changes in parameters associated with the abnormality and the transmission logic. This solution, by tracing the events of changes in associated parameters and constructing a transmission chain, clearly shows the evolution path of the abnormality among battery parameters and finds the underlying cause.

[0140] Furthermore, by extracting key coupled parameter sets through parameter decoupling, only parameter changes related to anomalies are traced, avoiding interference from irrelevant parameters, improving the pertinence and efficiency of source tracing analysis, and ensuring that subsequent causal analysis focuses on core correlations.

[0141] In summary, the anomaly propagation chain is constructed based on temporal and causal relationships, forming a structured anomaly propagation network. This provides a clear and traceable path for subsequent reverse traversal to locate the origin of the anomaly, avoiding blindness in the tracing process. The propagation chain presents the entire process of an anomaly from initial parameter changes to its spread, allowing maintenance personnel to not only know the anomaly result but also understand how the anomaly is generated and propagated, providing in-depth evidence for accurately judging the root cause of the fault and developing targeted maintenance plans.

[0142] In this embodiment of the invention, the reverse traversal of the anomaly propagation chain to locate the time corresponding to the parameter event at the beginning of the anomaly propagation chain includes:

[0143] The anomaly propagation chain is parsed as a directed acyclic graph structure based on parameter events and causal driving relationships;

[0144] Performing a reverse topological sort on a directed acyclic graph structure yields a reverse event sequence of the directed acyclic graph structure;

[0145] Extract the parameter event node located at the end of the reverse event sequence and obtain its corresponding timestamp, then determine the timestamp as the time of the anomaly origin.

[0146] In this embodiment of the invention, the anomaly propagation chain is parsed as a directed acyclic graph structure based on parameter events and causal driving relationships, including:

[0147] Define each parameter event in the exception propagation chain as a graph node, and assign the graph node its corresponding parameter type and timestamp attribute;

[0148] Based on the causal driving relationship, a directed edge is established between two graph nodes that have a causal relationship, where the direction of the directed edge is from the parameter event node representing the cause to the parameter event node representing the result.

[0149] Topological consistency verification is performed on the directed edges to obtain the directed acyclic graph structure.

[0150] Specifically, each parameter event in the abnormal propagation chain is treated as an independent graph node, and the parameter type corresponding to each graph node is clearly recorded as voltage, current or temperature. The original timestamp corresponding to the parameter event node is directly assigned to the graph node, so that each graph node has two fixed attributes: parameter type and timestamp.

[0151] Specifically, examine all graph nodes one by one, find two graph nodes with a causal driving relationship, and draw a one-way line segment from the parameter event node representing the cause to the parameter event node representing the result. This line segment is the directed edge, and ensure that the direction of the directed edge completely matches the causal logic.

[0152] Specifically, examine all established directed edges to confirm whether there is a cyclic path that starts from a certain graph node, passes through several directed edges, and then returns to itself. If a cycle is found, adjust the connection relationship of directed edges that do not conform to the causal order until all directed edges no longer form cycles, and finally obtain the directed acyclic graph structure.

[0153] Specifically, based on the directed acyclic graph structure, nodes are sorted in the reverse direction of the directed edges to ensure that each graph node is placed after all its original forward successor nodes, without violating the sequential logic of causal driving relationships, and finally forming a reverse event sequence.

[0154] Specifically, examine the reverse event sequence, locate the parameter event node at the last position in the sequence, extract the timestamp attribute attached to the node, and this timestamp is the time corresponding to the parameter event at the beginning of the exception propagation chain, which can be directly determined as the time of the exception origin.

[0155] In summary, traditional methods struggle to pinpoint the initial origin of an anomaly from a temporal perspective. This step, by traversing the propagation chain in reverse, can trace back from the final anomaly result to the initial parameter event, accurately identifying the "source" of the anomaly. This addresses the pain point of "knowing only that the anomaly exists but not when it originated," providing a crucial time anchor for tracing the root cause of an anomaly.

[0156] Before traversal, the propagation chain is parsed into a directed acyclic graph and topologically sorted in reverse. The causal logic is used to backtrack in an orderly manner, avoiding the problems of circular judgment or omission of key nodes that may occur in traditional tracing. This ensures that the positioning process is structured and verifiable, reduces human subjective error, and improves the reliability of the origin time determination.

[0157] Determining the origin time allows us to trace the evolution cycle of an anomaly from initial parameter changes to its manifestation, providing a time reference for predicting anomaly development trends and taking early intervention measures. This helps prevent anomalies from escalating into catastrophic failures such as thermal runaway and enhances the proactiveness of battery system safety control.

[0158] In this embodiment of the invention, the battery state parameters, operating environment data, and propagation logic of the anomaly propagation chain corresponding to the anomaly origin time are integrated to obtain a diagnostic report of the battery detection data, including:

[0159] Extract battery state parameters and operating environment data corresponding to the time of anomaly origin;

[0160] Semantic parsing is performed on the anomaly propagation chain to obtain a description of the propagation path of the anomaly propagation chain;

[0161] A diagnostic report is obtained by structurally integrating battery status parameters, operating environment data, and propagation path descriptions.

[0162] like Figure 2 The diagram shown is a functional block diagram of an abnormal diagnosis and traceability system for battery testing data provided in an embodiment of the present invention.

[0163] This invention discloses a battery testing data anomaly diagnosis and traceability system 100, which can be installed in an electronic device. Depending on the functions implemented, the battery testing data anomaly diagnosis and traceability system 100 may include a preprocessing module 101, a data comparison module 102, an anomaly judgment module 103, an anomaly propagation chain generation module 104, a traceability analysis module 105, and a diagnostic report generation module 106. The modules of this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0164] In this embodiment, the functions of each module / unit are as follows:

[0165] The preprocessing module 101 is used to normalize the real-time detection data of the battery to obtain the preprocessed dataset of the real-time detection data.

[0166] The data comparison module 102 is used to compare the real-time data distribution of the preprocessed dataset with the pre-stored historical data benchmark distribution to obtain the adaptive detection threshold of the preprocessed dataset.

[0167] The anomaly detection module 103 is used to determine the abnormal state of the preprocessed dataset using an adaptive detection threshold, and to obtain abnormal data fragments of the preprocessed dataset.

[0168] The anomaly propagation chain generation module 104 is used to trace the change events of related parameters from the battery's historical operating state based on the anomaly data fragments, and to construct the anomaly propagation chain according to the temporal relationship of the change events.

[0169] The source analysis module 105 is used to reverse the abnormal transmission chain, locate the time corresponding to the parameter event at the beginning of the abnormal transmission chain, and obtain the time of the abnormal origin.

[0170] The diagnostic report generation module 106 is used to integrate the battery state parameters, working environment data and the propagation logic of the abnormality transmission chain corresponding to the time of the abnormality origin to obtain a diagnostic report of the battery detection data.

[0171] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0172] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0173] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0174] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0175] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for anomaly diagnosis and source tracing of battery test data, characterized in that, The method includes: S1. Normalize the real-time detection data of the battery to obtain the preprocessed dataset of the real-time detection data; S2. Based on the pre-stored historical data baseline distribution and historical baseline threshold, the real-time data distribution of the preprocessed dataset is compared to obtain the adaptive detection threshold of the preprocessed dataset. S3. Use the adaptive detection threshold to determine the abnormal state of the preprocessed dataset to obtain abnormal data fragments of the preprocessed dataset. S4. Based on the abnormal data fragments, trace the change events of related parameters from the battery's historical operating states, and construct an abnormal propagation chain according to the temporal relationship of the change events. S5. Perform reverse traversal of the abnormal propagation chain to locate the time corresponding to the parameter event at the beginning of the abnormal propagation chain, and obtain the time of the abnormal origin. S6. Integrate the battery state parameters, operating environment data, and the propagation logic of the anomaly propagation chain corresponding to the anomaly origin time to obtain a diagnostic report of the battery detection data; The step of comparing the real-time data distribution of the preprocessed dataset based on the pre-stored historical data benchmark distribution and historical benchmark threshold includes: Real-time features of battery voltage, current, and temperature data are extracted from the preprocessed dataset. From the pre-stored historical data benchmark distribution, obtain the benchmark features of the corresponding voltage, current and temperature data; Calculate the volatility stability index of the data parameters in the historical data benchmark distribution, and assign weight factors to the data parameters according to the volatility stability index to obtain the weight vector of the data parameters; The weight vector is used to weight and synthesize the real-time feature and the baseline feature respectively, to obtain a weighted real-time feature vector and a weighted baseline feature vector; Distance quantization is performed between the weighted real-time feature vector and the weighted baseline feature vector to obtain the distribution difference between the two. The historical baseline threshold and the distribution difference are linearly combined, and the combined result is used as the adaptive detection threshold. The formula for calculating the volatility stability index is as follows: ; In the formula, Indicates the first The volatility stability index of each data parameter, the Corresponding to voltage, current and temperature respectively ; indicates the first [number] calculated based on the historical data baseline distribution. The standard deviation of each data parameter; It is a pre-defined, extremely small positive number.

2. The method for anomaly diagnosis and source tracing of battery test data as described in claim 1, characterized in that, The step of using the adaptive detection threshold to determine abnormal states in the preprocessed dataset includes: Based on the adaptive detection threshold, continuous data points in the preprocessed dataset are labeled with their states, and data points exceeding the adaptive detection threshold are labeled as candidate outliers. The candidate anomalies are continuously verified, and the set of candidate anomalies that appear continuously for a duration exceeding a preset window is determined as a valid anomaly state. Based on the start and end points of the valid abnormal states on the time axis, the corresponding data sequences are extracted from the preprocessed dataset to obtain the abnormal data segments.

3. The method for anomaly diagnosis and source tracing of battery test data as described in claim 1, characterized in that, The step of tracing changes in related parameters from the battery's historical operating states based on the anomalous data fragments, and constructing an anomaly propagation chain according to the temporal relationship of the changes, includes: The abnormal data segment is decoupled from its parameters to obtain the set of key coupling parameters for the abnormal data segment. Starting from the time when the abnormal data segment occurred, the time when each parameter in the key coupling parameter set first showed a statistically significant deviation is traced back in the battery's historical operating state to obtain a set of parameter event nodes; A causal association analysis was performed on all the parameter event nodes to obtain the causal driving relationship between the parameter events; By linking the parameter event nodes with the causal driving relationship, a causal event sequence of the abnormal data fragment is obtained; The causal event sequence is integrated in chronological order to obtain the anomaly propagation chain.

4. The method for anomaly diagnosis and source tracing of battery test data as described in claim 3, characterized in that, The parameter decoupling of the abnormal data fragment includes: Extract standardized data sequences of battery voltage, current, and temperature parameters during the abnormal period from the abnormal data fragments; Correlation analysis is performed on the parameters in the standardized data sequence to obtain information on the correlation strength between the parameters in the standardized data sequence; Based on the association strength information, strongly associated parameter pairs exceeding a preset association threshold are identified; The key coupling parameter set is obtained by combining the parameters involved in the strongly correlated parameters through a union.

5. The method for anomaly diagnosis and source tracing of battery test data as described in claim 3, characterized in that, The causal correlation analysis of all the parameter event nodes includes: Calculate the transition entropy between the data sequences corresponding to the parameter event nodes, and construct the causal strength matrix; Each element in the causal intensity matrix is ​​compared with a preset causal threshold, and the transition entropy value that exceeds the causal threshold is recorded as the determined entropy value; Establish the causal driving relationship between the parameter events based on the determined entropy value; The formula for calculating the transfer entropy is as follows: ; in, Indicates the event from the parameter. arrive The transfer entropy value, and These represent two different parameter events. Indicates parameter event exist Before the moment Historical data sequences at various points in time. Indicates parameter event exist Before the moment Historical data sequences at various points in time. Indicates parameter event exist Future data values ​​at any given time This represents the joint probability distribution or conditional probability distribution of the corresponding event sequence.

6. The method for anomaly diagnosis and source tracing of battery test data as described in claim 1, characterized in that, The step of reverse traversing the anomaly propagation chain to locate the time corresponding to the parameter event at the beginning of the anomaly propagation chain includes: The anomaly propagation chain is parsed as a directed acyclic graph structure based on parameter events and causal driving relationships; Perform a reverse topological sort on the directed acyclic graph structure to obtain the reverse event sequence of the directed acyclic graph structure; Extract the parameter event node located at the end of the reverse event sequence and obtain its corresponding timestamp, and determine the timestamp as the time of the anomaly origin.

7. The method for anomaly diagnosis and source tracing of battery test data as described in claim 6, characterized in that, The step of resolving the anomaly propagation chain into a directed acyclic graph structure based on parametric events and causal driving relationships includes: Define each parameter event in the anomaly propagation chain as a graph node, and assign the graph node its corresponding parameter type and timestamp attribute; Based on the causal driving relationship, a directed edge is established between two graph nodes that have a causal relationship, wherein the direction of the directed edge is from the parameter event node representing the cause to the parameter event node representing the result. The directed edges are subjected to topological consistency verification to obtain the directed acyclic graph structure.

8. The method for anomaly diagnosis and source tracing of battery test data as described in claim 1, characterized in that, The process of integrating the battery state parameters, operating environment data, and the propagation logic of the anomaly propagation chain corresponding to the anomaly origin time to obtain a diagnostic report of the battery detection data includes: Extract the battery state parameters and operating environment data corresponding to the time of the anomaly origin; Semantic parsing is performed on the anomaly propagation chain to obtain a description of the propagation path of the anomaly propagation chain; The diagnostic report is obtained by structurally integrating the battery status parameters, operating environment data, and propagation path description.

9. A system for diagnosing and tracing anomalies in battery testing data, characterized in that, The system, applied to the method for anomaly diagnosis and tracing of battery test data as described in any one of claims 1-8, further includes: The preprocessing module is used to normalize the real-time detection data of the battery to obtain the preprocessed dataset of the real-time detection data. The data comparison module is used to compare the real-time data distribution of the preprocessed dataset with the pre-stored historical data benchmark distribution to obtain the adaptive detection threshold of the preprocessed dataset. An anomaly detection module is used to determine the anomaly state of the preprocessed dataset using the adaptive detection threshold, and to obtain the abnormal data fragments of the preprocessed dataset. An anomaly propagation chain generation module is used to trace the change events of related parameters from the battery's historical operating state based on the anomaly data fragments, and to construct an anomaly propagation chain according to the temporal relationship of the occurrence of the change events. The source tracing analysis module is used to traverse the anomaly propagation chain in reverse, locate the time corresponding to the parameter event at the beginning of the anomaly propagation chain, and obtain the anomaly origin time. The diagnostic report generation module is used to integrate the battery state parameters, operating environment data, and the propagation logic of the abnormality transmission chain corresponding to the time of the abnormality origin to obtain a diagnostic report of the battery detection data.

Images