Charging data anomaly detection method and system

Abstract

The invention relates to the technical field of image recognition, in particular to a charging data anomaly detection method and system, and the method comprises the following steps: collecting multi-dimensional operation data in a charging process, and storing the multi-dimensional operation data according to a charging session; performing cleaning and time reference processing on the operation data, and generating a relative time sequence taking the charging start time as a zero point; carrying out abnormal value detection based on the relative time sequence, and removing data records judged to be abnormal; when it is detected that the recording time interval of adjacent data exceeds a preset threshold value, missing data are complemented through an interpolation method, and continuous analysis data are obtained. According to the method, comprehensive calculation and judgment are carried out by utilizing multiple characteristic parameters such as SOC consistency, voltage consistency, temperature consistency, capacity consistency and temperature rise rate, and the operation state of the battery in the charging process is reflected from multiple angles; the multi-index fusion analysis mode can effectively identify potential anomalies which are difficult to discover by a single index.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for detecting abnormal charging data. Background Technology

[0002] The charging data anomaly detection method and system is a technical solution for collecting, processing, and analyzing operational data during the charging process of electric vehicles and battery energy storage devices. It aims to identify abnormal situations during charging by monitoring and processing multi-dimensional data such as voltage, current, temperature, and state of charge. This technology is typically applied in scenarios such as charging safety management, battery health status assessment, and charging quality control. It can provide maintenance personnel with real-time or post-event analysis data to assist in formulating maintenance strategies and optimizing charging strategies, thereby ensuring the safety and stability of the battery during charging and providing data support for subsequent performance evaluation and lifespan prediction.

[0003] In existing technologies, abnormal charging data detection often relies on a single indicator for judgment, lacking comprehensive analysis of multi-dimensional feature parameters. This makes it difficult to fully reflect the battery charging status, resulting in insufficient coverage and low accuracy of abnormal detection. Summary of the Invention

[0004] To overcome the above shortcomings, the present invention provides a charging data anomaly detection method and system, which aims to improve the problems of insufficient anomaly detection coverage and low accuracy caused by relying on a single detection index in the prior art, which makes it difficult to fully reflect the battery charging status.

[0005] In a first aspect, the present invention provides the following technical solution: a method for detecting abnormal charging data, comprising the following steps:

[0006] During the charging process, multi-dimensional operational data is collected and stored according to charging sessions;

[0007] The operational data is cleaned and time-based to generate a relative time series with the charging start time as the zero point;

[0008] Anomaly detection is performed based on the relative time series, and data records determined to be abnormal are removed.

[0009] When the time interval between adjacent data records exceeds a preset threshold, interpolation is used to fill in the missing data and obtain continuous analysis data.

[0010] Based on the continuous analysis data, charging characteristic parameters are calculated, and the presence of abnormal states during the charging process is determined according to the charging characteristic parameters. If an abnormal state exists, an early warning is triggered and relevant information is recorded. The battery health score is calculated based on the charging characteristic parameter set, and the battery safety level is determined according to the battery health score.

[0011] By adopting the above technical solution, a comprehensive analysis of multi-dimensional operational data of the charging process is realized. It can make joint judgments through charging characteristic parameters, thereby effectively overcoming the problem that existing technologies rely on a single indicator to detect the battery charging status, which makes it difficult to fully reflect the battery charging status and results in insufficient coverage and low accuracy of anomaly detection.

[0012] Preferably, the collection of multi-dimensional operational data includes: starting a data collection task at the beginning of the charging session, sequentially acquiring the vehicle identification code, timestamp, vehicle voltage, vehicle current, voltage of each individual cell, maximum temperature of each individual cell, minimum temperature of each individual cell, state of charge (SOC), charging pile output voltage, charging pile output current limit, charging mode, BMS status code, and charging pile status code, and storing the acquired data in the data buffer according to the collection time order.

[0013] Preferably, cleaning the operational data includes the following steps:

[0014] Duplicate records are detected and deleted from the data buffer.

[0015] Check each data field sequentially to ensure that its format conforms to the preset rules;

[0016] Convert voltage, current, temperature, and SOC values ​​to SI units.

[0017] Records with negative voltage, negative current, temperature exceeding the physical reasonable range, or SOC value not within the range of 0% to 100% will be removed.

[0018] Preferably, the time-based standardization process includes the following steps:

[0019] The cleaned records are grouped by charging session, and the timestamp of the first record in each group is taken as zero point.

[0020] Subtract the zero-point timestamp from the timestamps of other records in the group to obtain the relative time value with respect to the charging start time, and use this relative time value as the time field of the record to output to the preprocessed dataset.

[0021] Preferably, the outlier detection includes the following steps:

[0022] The preprocessed dataset is input into the isolation forest algorithm, and multiple isolation trees are constructed by randomly selecting feature fields and their values;

[0023] In each isolation tree, samples are recursively partitioned until the samples are completely isolated. The average path length of each sample is calculated and converted into anomaly score.

[0024] When the abnormal score exceeds the preset threshold, the corresponding record is marked as abnormal and removed from the preprocessed dataset.

[0025] Preferably, the missing data completion includes: after removing abnormal records, detecting the relative time difference between two adjacent records; when the time difference is greater than 60 seconds, constructing a Newton interpolation polynomial based on the time values ​​and feature field values ​​of adjacent valid data points within the time interval, and using the polynomial to calculate and insert the data record of the missing time point, thereby generating a continuous analysis data sequence.

[0026] Preferably, the calculated characteristic parameters include: calculating SOC consistency, voltage consistency, temperature consistency, capacity consistency, and temperature rise rate in a continuous analysis data sequence, and outputting the calculation results as a set of charging characteristic parameters.

[0027] Preferably, the SOC consistency is obtained by calculating the rate consistency coefficient of the SOC change rate of each individual unit;

[0028] Voltage consistency is obtained by calculating the root mean square error or coefficient of variation of the voltage of each individual cell under the same SOC.

[0029] Temperature uniformity is obtained by calculating the root mean square value or coefficient of variation of the difference between the highest and lowest temperatures of monomers under the same SOC.

[0030] Capacity consistency is obtained by calculating the coefficient of variation or range of the capacity of each individual cell.

[0031] Preferably, the calculation of the temperature rise rate includes:

[0032] Within the sliding time window of the charging process, the difference between the highest single-cell temperature at the current time point and the highest single-cell temperature at the previous time point is calculated, and this difference is used as the temperature rise rate.

[0033] When the temperature rise rate is greater than 7°C, the record at the corresponding time point is marked as abnormal and written into the abnormal record table.

[0034] Secondly, the present invention provides the following technical solution: a charging data anomaly detection system, the system comprising:

[0035] The data acquisition module is used to collect multi-dimensional operational data during the charging process and store it according to the charging session.

[0036] The data cleaning module is used to perform duplicate record detection, format verification, unit unification and abnormal value removal on the running data to obtain cleaned running data;

[0037] The time-based standardization module is used to group the cleaned operational data by charging session and set the timestamp of the first record in each group to zero, generating a relative time series.

[0038] The anomaly detection module is used to detect abnormal records based on relative time series and mark missing data when the time interval between adjacent data records exceeds a preset threshold.

[0039] The missing data completion module is used to complete the missing data using an interpolation method to generate continuous analysis data;

[0040] The characteristic parameter calculation module is used to calculate SOC consistency, voltage consistency, temperature consistency, capacity consistency, and temperature rise rate based on continuous analysis data.

[0041] The anomaly detection and recording module is used to determine whether there is an abnormal state during the charging process based on the feature parameters, and to trigger an early warning and record relevant information when an anomaly is found. It also calculates the battery health score based on the feature parameter set and determines the battery safety level based on the battery health score.

[0042] The present invention has the following beneficial effects:

[0043] 1. In this invention, multiple characteristic parameters such as SOC consistency, voltage consistency, temperature consistency, capacity consistency, and temperature rise rate are used for comprehensive calculation and judgment to reflect the battery's operating status during the charging process from multiple perspectives. This multi-index fusion analysis method can effectively identify potential anomalies that are difficult to detect with a single index, and improve the coverage and accuracy of anomaly detection.

[0044] 2. In this invention, by collecting, cleaning, time-based standardization, anomaly detection, and missing data completion of multi-dimensional operational data throughout the entire charging process, the end-to-end processing of data from collection to analysis is achieved. This solution can maintain the consistency of data timing and structure under different charging sessions and different data sources, providing an accurate and continuous input data foundation for subsequent feature parameter calculations.

[0045] 3. In this invention, by establishing an anomaly detection and recording mechanism, real-time warnings are issued for detection results during the charging process, and anomaly records are saved, providing a reliable basis for subsequent fault tracing and operational strategy optimization. Simultaneously, combined with a health status assessment model, the overall safety level of the charging process can be determined, providing intelligent decision support for charging operations and battery management. Attached Figure Description

[0046] Figure 1 This is a flowchart of a charging data anomaly detection method proposed in this invention;

[0047] Figure 2 This is a system architecture diagram of a charging data anomaly detection system proposed in this invention. Detailed Implementation

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1:

[0050] In a first embodiment of the present invention, the present invention provides a method for detecting abnormal charging data, such as... Figure 1 As shown, it includes the following steps:

[0051] During the charging process, multi-dimensional operational data is collected and stored according to charging sessions;

[0052] Furthermore, the collection of multi-dimensional operational data includes: starting a data acquisition task at the beginning of the charging session, sequentially acquiring the vehicle identification code, timestamp, vehicle voltage, vehicle current, voltage of each cell, maximum temperature of each cell, minimum temperature of each cell, state of charge (SOC), charging pile output voltage, charging pile output current limit, charging mode, BMS status code, and charging pile status code, and storing the acquired data in the data buffer according to the acquisition time order.

[0053] Specifically, the data acquisition task is initiated at the start of the charging session, triggered by the handshake signal between the vehicle and the charging pile, the charging start status reported by the BMS, or the session start command issued by the charging session management module. This ensures that the data acquisition is only carried out during the effective charging session, which facilitates data organization and traceability according to the session boundaries.

[0054] Data acquisition primarily uses vehicle terminals, battery management systems (BMS), and charging piles as data sources, collecting different types of operating parameters to form a multi-dimensional operating data set. Each data field is structured into data items during acquisition, and these structured records can be saved as key-value pairs, table rows, or serialized objects for subsequent parsing and processing in a unified format within the data buffer. The original source identifier, such as the data source device identifier or port identifier, should be retained in the records for later traceability and responsibility determination.

[0055] Data acquisition adopts a strategy that combines periodic acquisition with event-triggered acquisition: For time series quantities that need to be continuously monitored, such as voltage, current, individual cell voltage, individual cell temperature, and SOC, periodic acquisition is carried out using a preset sampling strategy; for discrete events such as status codes, fault codes, or session switching, event-triggered acquisition is used to collect data in real time and write it into the buffer to ensure that abnormal events are captured in a timely manner and are not missed due to periodic sampling.

[0056] The acquisition results are first written to a local data buffer to ensure that data is not lost during short-term disconnection. The data buffer can be implemented in the form of a sequential queue or a circular buffer. Each record in the buffer is organized in the order of acquisition time and includes necessary verification information such as record sequence number or checksum to support subsequent integrity verification and retransmission mechanisms.

[0057] Records in the data buffer undergo basic validation upon writing, including timestamp validity validation, field integrity validation, and simple range checks to quickly filter out obviously erroneous data. Records that pass validation are saved in the local buffer and queued for reporting or further processing. Records that fail validation, such as those with missing fields or basic format errors, can be marked, recorded, or discarded according to a policy, and logged. To ensure data manageability and retrieval efficiency, the buffer and subsequent persistent storage should use charging sessions as the smallest grouping unit, numbering and indexing data within the same session according to the order of collection time. Each session record simultaneously records the session identifier, vehicle identification code, and session start and end times for rapid location and session-level analysis.

[0058] Data in the buffer is uploaded to the backend server or cloud storage through a secure communication channel when conditions are met, such as when communication is available or when batch reporting trigger conditions are met. During the upload process, transmission confirmation, retransmission, and packet splitting mechanisms can be used to ensure the complete transmission of data. Sensitive fields can be encrypted or de-identified before uploading to meet privacy and compliance requirements.

[0059] To facilitate subsequent unified data processing and cross-source comparison, the semantics and unit conventions of the fields should be clearly defined during the data collection phase. The data collection end can annotate the unit information in the records or store them directly in the preset standard units, thereby reducing data ambiguity and conversion work caused by inconsistent units in the future.

[0060] In response to temporary data interruptions caused by network jitter or device malfunction, the system supports local buffer re-uploading and sequential resending. During resending, the original timestamp and session identifier of the record should remain unchanged to ensure the continuity and traceability of the backend time series.

[0061] The data acquisition module can be implemented by combining software tasks with hardware timing. The software should include data acquisition task scheduling, data acquisition strategy configuration interface, data verification and cache management interface. The hardware can use vehicle-mounted terminals or gateway devices to carry out data acquisition and buffering functions. The implementation method can be selected and adjusted according to the open capabilities of the vehicle platform.

[0062] The operational data is cleaned and time-based to generate a relative time series with the charging start time as the zero point.

[0063] Furthermore, cleaning the operational data includes the following steps:

[0064] Perform duplicate record detection and delete duplicates from the records in the data buffer;

[0065] Check each data field sequentially to ensure that its format conforms to the preset rules;

[0066] Convert voltage, current, temperature, and SOC values ​​to SI units.

[0067] Records with negative voltage, negative current, temperature exceeding the physical reasonable range, or SOC value not within the range of 0% to 100% will be removed.

[0068] Furthermore, the time-based standardization process includes the following steps:

[0069] The cleaned records are grouped by charging session, and the timestamp of the first record in each group is taken as zero point.

[0070] Subtract the zero-point timestamp from the timestamps of other records in the group to obtain the relative time value with respect to the charging start time, and use this relative time value as the time field of the record to output to the preprocessed dataset.

[0071] Specifically, to ensure the data quality and time alignment accuracy of subsequent analysis, after collecting multi-dimensional operational data, it is first cleaned and time-based to generate a relative time series with the charging start time as the zero point.

[0072] The data cleaning process operates on records in the data buffer. Cleaning begins with duplicate record detection, which compares the timestamps and data field values ​​of adjacent records to identify duplicates acquired at the same time or due to communication retransmissions. Duplicates are then removed from the dataset to prevent statistical bias in subsequent analysis. After deduplication, each data field of each record undergoes format validation to ensure it conforms to preset parsing rules. For example, timestamp fields must be in a valid time format, numeric fields must meet numeric type requirements, and status code fields must be within a predefined status range. If a field format is found to be non-compliant, the record is marked as invalid and removed.

[0073] During the cleaning process, the units of voltage, current, temperature, and SOC values ​​from different sources need to be standardized. To avoid unit differences caused by different acquisition devices or protocols, the system converts all numerical fields to SI units to ensure the comparability and consistency of calculation results in subsequent analysis stages. The system performs physical rationality checks on the operating data. For example, voltage values ​​must not be negative, current values ​​must not be negative, temperature values ​​should be within the physically feasible range, and SOC values ​​should be between 0% and 100%. Any records that do not meet the above physical constraints are considered abnormal data and are removed during the cleaning stage to prevent them from interfering with subsequent feature analysis and anomaly judgment.

[0074] After data cleaning, the cleaned records undergo time-based standardization. First, the records are grouped by charging session to ensure that the data within each session is interconnected and temporally continuous. Within each charging session group, the timestamp of the first record in that session is set to zero, serving as the starting reference time for that session. This zero-point selection ensures that data from different sessions are aligned relative to their respective charging start times, supporting cross-session comparison and analysis. Then, the zero-point timestamp is subtracted from the timestamps of the other records in the group to calculate the relative time value relative to the charging start time. This relative time value is added as a new time field to the data record. In this way, records originally on an absolute timeline are mapped to a relative timeline with the charging start time as the reference, thus more intuitively reflecting the temporal evolution characteristics of the charging process in subsequent analysis.

[0075] The data records that have completed time benchmarking are output to the preprocessed dataset. This preprocessed dataset retains the original absolute timestamp information for traceability, and also includes relative time fields for subsequent time series analysis and model calculation, thus achieving a balance between data traceability and computability.

[0076] Outlier detection is performed based on relative time series data, and data records determined to be abnormal are removed.

[0077] Furthermore, outlier detection includes the following steps:

[0078] The preprocessed dataset is input into the isolation forest algorithm, and multiple isolation trees are constructed by randomly selecting feature fields and their values;

[0079] In each isolation tree, samples are recursively partitioned until the samples are completely isolated. The average path length of each sample is calculated and converted into anomaly score.

[0080] When the abnormal score exceeds the preset threshold, the corresponding record is marked as abnormal and removed from the preprocessed dataset.

[0081] Specifically, after completing the time benchmarking process, outlier detection is performed on the charging process data based on the generated relative time series to remove records that deviate significantly from normal charging behavior in terms of data distribution characteristics, thereby ensuring that the data on which subsequent analysis depends has high authenticity and reliability.

[0082] In the outlier detection process, the preprocessed dataset, after time benchmarking, is first input into the Isolation Forest algorithm for modeling and analysis. The Isolation Forest algorithm, by constructing multiple independent decision tree structures, can efficiently identify outliers located in sparse regions or far from normal clusters in the feature space. When constructing the isolation forest, the system randomly selects some feature fields and their value ranges in each isolation tree as the basis for partitioning. These feature fields may include multi-dimensional operating parameters such as voltage, current, temperature, and SOC during the charging process. By selecting different feature combinations on different isolation trees, the model's ability to capture various types of anomaly patterns can be improved.

[0083] Within each isolation tree, the algorithm recursively partitions samples until a sample is completely isolated as a single node along the partitioning path. At this point, the path length of that sample in that tree is recorded. Since outliers are usually located at the edges of the data distribution, they require fewer partitions to be isolated, and therefore their path lengths are often significantly shorter than those of normal samples. The system calculates the average path length of each sample across all isolation trees and converts this average path length into an anomaly score. The anomaly score reflects the sparsity of the sample in the overall data distribution; a higher score indicates a greater likelihood of it being an outlier.

[0084] When the anomaly score of a record exceeds a preset threshold, the system will mark the record as an anomaly and remove it from the preprocessed dataset to prevent it from interfering with subsequent steps such as missing data completion, feature parameter calculation, and anomaly status determination.

[0085] Through the above method, this embodiment can effectively eliminate abnormal data points in the charging process while ensuring computational efficiency, providing a reliable input data foundation for subsequent analysis based on continuous and accurate data.

[0086] When the time interval between adjacent data records exceeds a preset threshold, interpolation is used to fill in the missing data and obtain continuous analysis data.

[0087] Furthermore, missing data completion includes: after removing abnormal records, detecting the relative time difference between two adjacent records; when the time difference is greater than 60 seconds, constructing a Newton interpolation polynomial based on the time values ​​and feature field values ​​of adjacent valid data points within this time interval, and using this polynomial to calculate and insert the data record of the missing time point, thereby generating a continuous analysis data sequence.

[0088] Specifically, after outlier detection and removal, to avoid discontinuous data analysis due to large time intervals in the data records, the time intervals between adjacent data records are further checked to determine if there is any missing data that needs to be filled in.

[0089] Specifically, the system sequentially reads the relative time series after removing abnormal records and calculates the time difference between two adjacent valid records. When the time difference exceeds a preset time threshold, it is considered that there is data missing within that time period, requiring completion processing to maintain the temporal continuity of the charging process data. During the completion process, the system uses the previous and next valid records within the time interval as boundary points to obtain their relative time values ​​and corresponding feature field values. Feature fields include multi-dimensional operational data reflecting the charging status, such as vehicle voltage, vehicle current, individual cell voltage, individual cell temperature, and SOC. Subsequently, a Newton interpolation polynomial is constructed based on the above time and feature field values. This polynomial estimates the feature values ​​of the missing time points by utilizing the changing trends of adjacent known data points, thereby closely approximating the actual charging process's changing patterns in both time and numerical dimensions.

[0090] The system will generate data records corresponding to the missing time points according to the calculation results of the interpolation polynomial, and insert these records into the original time series, so that the processed analysis data forms a continuous sequence on the time axis.

[0091] Based on continuous analysis data, charging characteristic parameters are calculated, and the presence of abnormal states during the charging process is determined according to the charging characteristic parameters. If an abnormal state exists, an early warning is triggered and relevant information is recorded. The battery health score is calculated based on the charging characteristic parameter set, and the battery safety level is determined based on the battery health score.

[0092] Furthermore, the calculation of characteristic parameters includes: calculating SOC consistency, voltage consistency, temperature consistency, capacity consistency, and temperature rise rate in a continuous analysis data sequence, and outputting the calculation results as a set of charging characteristic parameters.

[0093] Furthermore, SOC consistency is obtained by calculating the rate consistency coefficient of the SOC change rate of each individual unit;

[0094] Voltage consistency is obtained by calculating the root mean square error or coefficient of variation of the voltage of each individual cell under the same SOC.

[0095] Temperature uniformity is obtained by calculating the root mean square value or coefficient of variation of the difference between the highest and lowest temperatures of monomers under the same SOC.

[0096] Capacity consistency is obtained by calculating the coefficient of variation or range of the capacity of each individual cell.

[0097] Furthermore, the calculation of the temperature rise rate includes:

[0098] Within the sliding time window of the charging process, the difference between the highest single-cell temperature at the current time point and the highest single-cell temperature at the previous time point is calculated, and this difference is used as the temperature rise rate.

[0099] When the temperature rise rate is greater than 7°C, the record at the corresponding time point is marked as abnormal and written into the abnormal record table.

[0100] Specifically, based on continuous analysis of data sequences, characteristic parameters are calculated on the data to determine the state of the charging process;

[0101] In the specific implementation process, the data such as the state of charge (SOC), voltage, temperature and capacity of individual cells involved in the continuous analysis data sequence are first extracted, and characteristic parameters such as SOC consistency, voltage consistency, temperature consistency, capacity consistency and temperature rise rate are calculated according to the extraction results to form a charging characteristic parameter set;

[0102] For SOC consistency, the SOC rate is first calculated using the following formula:

[0103]

[0104] Then calculate the consistency coefficient, using the following formula:

[0105]

[0106] The formula for calculating the SOC consistency score is:

[0107] SOC Consistency Score = 100 - (Consistency Coefficient);

[0108] For voltage consistency, first calculate the root mean square value of the difference between the maximum and minimum voltage values ​​of each cell under the same SOC, using the following formula:

[0109]

[0110] Next, calculate the standardized value for voltage consistency using the following formula:

[0111]

[0112] in, U is the average value of the voltage curve. max-j and U min-j These represent the maximum and minimum single-cell voltages measured at the same SOC, respectively, where k is the SOC change from the start of charging to the end of charging; σ v This represents the root mean square error between the highest and lowest voltage curve data during the charging phase; δv The coefficient of variation of the charging voltage of each individual cell, δ, is a parameter for evaluating voltage consistency. v The larger the value, the worse the voltage consistency of the battery pack.

[0113] Cell voltage difference = highest cell voltage - lowest cell voltage;

[0114] The formula for calculating the voltage consistency score is as follows:

[0115] Indicator voltage consistency = 100 - δ v ;

[0116] For temperature consistency, first calculate the average of the sum of the highest temperatures of all monomers under the same SOC value, using the following formula:

[0117]

[0118] Where T represents the sum of the highest power battery temperatures measured at the same SOC value, and k represents the number of measurements;

[0119] Next, calculate the average maximum temperature value during the entire charging process from start to finish. The formula is:

[0120]

[0121] Where T m-j This represents the average highest temperature measured in the database from the start to the end of charging, T. j This represents the average temperature of all the highest power batteries measured at the same SOC value, where n represents the change in SOC from the start of charging to the end of charging.

[0122] The maximum and minimum measured temperatures of the power battery at each SOC value are compared with the average of the maximum measured temperatures from the start to the end of charging. Data with large deviations from the average are removed. In data processing, normal deviations are generally quantified using the standard deviation. The remaining data are used as the maximum and minimum values ​​of the highest temperature. Then, the temperature consistency coefficient is calculated based on the difference between the highest and lowest temperatures of individual cells at the same SOC. The formula is as follows:

[0123]

[0124] Where T max-j and T min-j σ represents the maximum and minimum values ​​of the highest temperature measured by the power battery at a certain S0C. T σ represents the root mean square error between the highest and lowest temperature curves during the charging phase, and is used as a parameter to evaluate temperature consistency.T The larger the size, the worse the temperature uniformity of the battery pack;

[0125] The temperature difference refers to the difference between the highest and lowest temperatures of the battery at the same moment during the charging process; let the temperature difference at time t be T. t The highest battery temperature at this moment is T. h The lowest battery temperature is T. i Therefore, the temperature difference formula is:

[0126] T t =T h -T i ;

[0127] The formula for calculating the temperature consistency score is:

[0128] Temperature uniformity = 100 - σ T ;

[0129] To ensure capacity consistency, first calculate the capacity of this charge using the following formula:

[0130] Q = R cc +Q sys ;

[0131] Where Q represents the battery capacity during this charge; R cc Q represents the battery's internal resistance, which can be obtained from the battery charging voltage and battery charging current; sys The static capacity under BSC conditions is represented by the following formula:

[0132]

[0133] Where I i This is the current value sampled per minute, where T is the total charging time. Converting time to hours, we divide by 60; Q sys BSC represents the basic static capacity of a battery, which refers to the amount of charge or energy that a battery can store under specific conditions; the total battery capacity of a fully charged battery usually refers to the total energy or total charge that a battery can provide in a fully charged state; in the database of the embodiment, the difference between the two can be ignored and is assumed to be equal.

[0134] The average capacity after multiple charging is calculated using the following formula:

[0135]

[0136] in This represents the average battery capacity after n charges;

[0137] Its standard deviation formula is:

[0138]

[0139] The formula for the coefficient of variation is:

[0140]

[0141] Where σ Q δ represents the standard deviation of the battery capacity after n charges; Q The standard deviation of the volume is represented by the coefficient of variation.

[0142] The formula for capacity range coefficient is:

[0143]

[0144] in The range coefficient of variation represents the capacity of each individual cell. This indicates the maximum value of the measured battery capacity. δ represents the minimum battery capacity among the measured data; Q and These are all parameters used to evaluate capacity consistency; δ Q and The larger the size, the worse the battery pack capacity consistency;

[0145] The formula for calculating the capacity consistency score is:

[0146]

[0147] The rate of temperature rise is the change in the battery's highest temperature per minute during charging; let the rate of increase be V. T T a Let T be the temperature value at time a. a-1 Let be the temperature value one minute before time 'a'. Therefore, the formula for calculating the rate of temperature difference is:

[0148] V T =(T a -T a-1 ) / 1;

[0149] During the charging process, if the difference between the current highest power battery temperature and the previous highest power battery temperature, as transmitted by the BMS to the charging pile, exceeds 7°C, an early warning will be issued.

[0150] The score calculation formula for abnormal termination is as follows:

[0151] Abnormal termination = 100 - (Number of abnormal terminations);

[0152] The final battery health score is calculated using the following formula:

[0153] Final battery health score = (battery capacity + temperature consistency + voltage consistency + SOC consistency + maximum temperature + abnormal termination) / 6;

[0154] Battery safety levels are divided into five categories: absolutely safe, safe, generally safe, dangerous, and extremely dangerous, which correspond to the final battery health scores of 90-100, 80-90, 70-79, 60-69, and 0-59, respectively.

[0155] Example 2:

[0156] During electric vehicle charging, due to the diversity of charging environment, equipment performance, and battery status, the collected charging data often contains outliers, missing values, and inconsistent time bases, leading to reduced accuracy and reliability of subsequent data analysis. To address these issues, this invention provides a charging data anomaly detection system, the structure of which is as follows: Figure 2 As shown. The specific implementation process of this system is as follows:

[0157] A charging data anomaly detection system, the system comprising:

[0158] The data acquisition module is used to collect multi-dimensional operational data during the charging process and store it according to the charging session.

[0159] The data cleaning module is used to perform duplicate record detection, format verification, unit unification, and abnormal value removal on the running data to obtain cleaned running data.

[0160] The time-based standardization module is used to group the cleaned operational data by charging session and set the timestamp of the first record in each group to zero, generating a relative time series.

[0161] The anomaly detection module is used to detect abnormal records based on relative time series and mark missing data when the time interval between adjacent data records exceeds a preset threshold.

[0162] The missing data completion module is used to complete missing data using interpolation methods to generate continuous analysis data;

[0163] The characteristic parameter calculation module is used to calculate SOC consistency, voltage consistency, temperature consistency, capacity consistency, and temperature rise rate based on continuous analysis data.

[0164] The anomaly detection and recording module is used to determine whether there is an abnormal state during the charging process based on feature parameters, and to trigger an alarm and record relevant information when an anomaly is found. It also calculates the battery health score based on the feature parameter set and determines the battery safety level based on the battery health score.

[0165] Specifically, the data acquisition module is used to initiate data acquisition tasks during the charging process of electric vehicles and organize and store the data according to the charging session; the acquired data is stored in the data buffer in chronological order to ensure the consistency of the timing of subsequent processing;

[0166] The data cleaning module is used to process the validity and consistency of the collected operational data. It first detects and deletes duplicate records; then it performs format validation on each field to ensure that it conforms to the preset parsing and processing rules; then it converts parameters such as voltage, current, temperature, and SOC into standard values ​​in the International System of Units (SI) to ensure the consistency of subsequent analysis and calculations; finally, it removes data records with negative voltage, negative current, temperature exceeding the physical reasonable range, or SOC values ​​outside the reasonable range, thereby obtaining a cleaned and valid dataset.

[0167] The time-based standardization module is used to perform time alignment and standardization on the cleaned operational data. Specifically, the module groups the data by charging session and sets the timestamp of the first record in each group to zero. Then, it performs a difference operation on the timestamps of the remaining records in the group to obtain a relative time value relative to the charging start time. This relative time value is then output as the record time field to the preprocessing dataset so that it can be calculated and analyzed with a unified time base in subsequent processing stages.

[0168] The anomaly detection module is used to detect anomalies in the dataset based on relative time series. It can call algorithms such as Isolation Forest to randomly select and recursively partition the samples in the preprocessed dataset using feature fields, generating multiple isolation trees, calculating the average path length of the samples, and converting it into anomaly scores. When a sample's anomaly score exceeds a set threshold, the module marks it as an anomaly record and adds an anomaly label to the record. Furthermore, this module is also responsible for detecting the time interval between adjacent data records; when the time interval exceeds a preset threshold, it marks it as a missing data point for use in subsequent data completion processes.

[0169] After detecting missing data points, the missing data completion module calls the interpolation calculation method to complete the records at the missing time. The module constructs an interpolation polynomial based on the time values ​​of the valid data points at both ends of the missing interval and the feature field values, and generates interpolation records at the missing time accordingly, thus forming a time-continuous and feature field-complete analysis data sequence.

[0170] The feature parameter calculation module is used to calculate the feature parameters of the charging process based on continuous analysis data. The module will call the corresponding calculation formula, extract and analyze the key feature values ​​of each individual cell during the charging process, and organize the calculation results into a set of charging feature parameters for the anomaly judgment link to call.

[0171] The anomaly detection and recording module performs a comprehensive analysis of the charging process based on the feature parameter set to determine whether there is an abnormal state. When the analysis results indicate that there is an anomaly, the module will trigger an early warning mechanism and record the relevant information when the anomaly occurs in the anomaly log for subsequent analysis, fault tracing and strategy optimization. The module calculates the battery health score based on the feature parameter set and determines the battery safety level based on the battery health score.

[0172] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting abnormal charging data, characterized in that, Includes the following steps: During the charging process, multi-dimensional operational data is collected and stored according to charging sessions; The operational data is cleaned and time-based to generate a relative time series with the charging start time as the zero point; Anomaly detection is performed based on the relative time series, and data records determined to be abnormal are removed. When the time interval between adjacent data records exceeds a preset threshold, interpolation is used to fill in the missing data and obtain continuous analysis data. Based on the continuous analysis data, charging characteristic parameters are calculated, and the presence of abnormal states during the charging process is determined according to the charging characteristic parameters. If an abnormal state exists, an early warning is triggered and relevant information is recorded. The battery health score is calculated based on the charging characteristic parameter set, and the battery safety level is determined according to the battery health score.

2. The charging data anomaly detection method according to claim 1, characterized in that, The collection of multi-dimensional operational data includes: starting a data collection task at the beginning of the charging session, sequentially acquiring the vehicle identification code, timestamp, vehicle voltage, vehicle current, voltage of each cell, maximum temperature of each cell, minimum temperature of each cell, state of charge (SOC), charging pile output voltage, charging pile output current limit, charging mode, BMS status code, and charging pile status code, and storing the acquired data in the data buffer according to the collection time order.

3. The charging data anomaly detection method according to claim 1, characterized in that, Cleaning the operational data includes the following steps: Duplicate records are detected and deleted from the data buffer. Check each data field sequentially to ensure that its format conforms to the preset rules; Convert voltage, current, temperature, and SOC values ​​to SI units. Records with negative voltage, negative current, temperature exceeding the physical reasonable range, or SOC value not within the range of 0% to 100% will be removed.

4. The charging data anomaly detection method according to claim 1, characterized in that, The time-based standardization process includes the following steps: The cleaned records are grouped by charging session, and the timestamp of the first record in each group is taken as zero point. Subtract the zero-point timestamp from the timestamps of other records in the group to obtain the relative time value with respect to the charging start time, and use this relative time value as the time field of the record to output to the preprocessed dataset.

5. The charging data anomaly detection method according to claim 1, characterized in that, The outlier detection includes the following steps: The preprocessed dataset is input into the isolation forest algorithm, and multiple isolation trees are constructed by randomly selecting feature fields and their values; In each isolation tree, samples are recursively partitioned until the samples are completely isolated. The average path length of each sample is calculated and converted into anomaly score. When the abnormal score exceeds the preset threshold, the corresponding record is marked as abnormal and removed from the preprocessed dataset.

6. The charging data anomaly detection method according to claim 1, characterized in that, The missing data completion includes: after removing abnormal records, detecting the relative time difference between two adjacent records; when the time difference is greater than 60 seconds, constructing a Newton interpolation polynomial based on the time values ​​and feature field values ​​of adjacent valid data points within the time interval, and using the polynomial to calculate and insert the data record of the missing time point, thereby generating a continuous analysis data sequence.

7. The charging data anomaly detection method according to claim 1, characterized in that, The calculated characteristic parameters include: calculating SOC consistency, voltage consistency, temperature consistency, capacity consistency, and temperature rise rate in a continuous analysis data sequence, and outputting the calculation results as a set of charging characteristic parameters.

8. The charging data anomaly detection method according to claim 7, characterized in that, The SOC consistency is obtained by calculating the rate consistency coefficient of the SOC change rate of each individual unit; Voltage consistency is obtained by calculating the root mean square error or coefficient of variation of the voltage of each individual cell under the same SOC. Temperature uniformity is obtained by calculating the root mean square value or coefficient of variation of the difference between the highest and lowest temperatures of monomers under the same SOC. Capacity consistency is obtained by calculating the coefficient of variation or range of the capacity of each individual cell.

9. The charging data anomaly detection method according to claim 7, characterized in that, The calculation of the temperature rise rate includes: Within the sliding time window of the charging process, the difference between the highest single-cell temperature at the current time point and the highest single-cell temperature at the previous time point is calculated, and this difference is used as the temperature rise rate. When the temperature rise rate is greater than 7°C, the record at the corresponding time point is marked as abnormal and written into the abnormal record table.

10. A charging data anomaly detection system, characterized in that, A charging data anomaly detection method according to any one of claims 1-9, the system comprising: The data acquisition module is used to collect multi-dimensional operational data during the charging process and store it according to the charging session. The data cleaning module is used to perform duplicate record detection, format verification, unit unification and abnormal value removal on the running data to obtain cleaned running data; The time-based standardization module is used to group the cleaned operational data by charging session and set the timestamp of the first record in each group to zero, generating a relative time series. The anomaly detection module is used to detect abnormal records based on relative time series and mark missing data when the time interval between adjacent data records exceeds a preset threshold. The missing data completion module is used to complete the missing data using an interpolation method to generate continuous analysis data; The characteristic parameter calculation module is used to calculate SOC consistency, voltage consistency, temperature consistency, capacity consistency, and temperature rise rate based on continuous analysis data. The anomaly detection and recording module is used to determine whether there is an abnormal state during the charging process based on the feature parameters, and to trigger an early warning and record relevant information when an anomaly is found. It also calculates the battery health score based on the feature parameter set and determines the battery safety level based on the battery health score.

Images