A Data Processing Method for Energy Storage Systems Based on Incremental Storage of Time-Series Data

By combining change detection and compression coding technologies, the problem of low data storage efficiency in time-series energy storage systems has been solved, achieving efficient incremental data storage and abnormal event analysis, and improving the accuracy of data reconstruction.

CN123115773APending Publication Date: 2026-11-17SHENZHEN HUAXING NEW ENERGY TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611574124.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-12
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, time-series data of energy storage systems are stored in a uniform manner during long-term operation, resulting in the simultaneous storage of redundant data and key change data. This leads to low storage efficiency and affects the accuracy of abnormal event tracing and analysis.

Method used

By employing joint change detection, state fragment aggregation fitting, compression coding, and bidirectional correlation reconstruction techniques based on full snapshots, we can distinguish between changed and redundant data, generate compressed codes and boundary anchors, establish bidirectional time pointers, perform data decoding and reconstruction, and monotonic conformal interpolation connections to extract causal association rules for out-of-limit events.

Benefits of technology

It improves the compression and storage efficiency of time-series data in energy storage systems and the accuracy of continuous data reconstruction, realizes incremental storage, rapid retrieval, and correlation analysis of abnormal events of operating data, and reduces data storage redundancy and processing resource consumption.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115773A_ABST
    Figure CN123115773A_ABST
Patent Text Reader

Abstract

This invention discloses a data processing method for energy storage systems based on incremental storage of time-series data, belonging to the field of storage processing technology. It addresses the problem of low storage efficiency by collecting runtime sequence parameters of the energy storage system, performing joint change detection, and calculating change vectors and projection deviation angles. The data is divided into changed data and redundant data. Based on the operating state type, the changed data is aggregated and segmented to form state segments, and trend basis function fitting is performed. Compressed codes and boundary anchors are output. Full sample values ​​are collected to generate a full snapshot and dual-copy storage is performed. A bidirectional time pointer is established between the snapshot and the compressed code. Based on the pointer, compressed code decoding and reconstruction are completed. Monotonic conformal interpolation correction is performed using boundary anchors. The rupture time point is located based on the corrected data sequence. The corresponding compressed code output final data sequence is obtained. Causal association rules for out-of-limit events are extracted and written into a feature library, reducing data redundancy and processing resource consumption.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of storage processing technology, and more specifically, to a data processing method for energy storage systems based on incremental storage of time-series data. Background Technology

[0002] With the large-scale application of energy storage systems, energy storage devices continuously generate time-series data containing information such as battery status, energy conversion status, and changes in the operating environment during operation. Because the operating status of energy storage systems has dynamic characteristics, the magnitude and frequency of data changes vary significantly in different operating stages. During some time periods, the operating parameters remain stable with only minor fluctuations, while during other time periods, due to charging / discharging switching, operating condition adjustments, or status changes, significant data change characteristics will occur.

[0003] The existing technology has the following shortcomings: Currently, during the long-term operation of energy storage systems, continuously collected time-series data are usually stored in a unified manner. This results in the simultaneous storage of redundant data during the stable phase and key change data. There is a lack of incremental data storage and high-precision recovery mechanisms based on the characteristics of changes in operating status, which leads to low storage efficiency, insufficient retention of key state evolution information, and affects the accuracy of abnormal event tracing and analysis. Therefore, a data processing method for energy storage systems based on incremental storage of time-series data is proposed.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a data processing method for energy storage systems based on incremental storage of time-series data. This method addresses the problems mentioned in the background by employing joint change detection, state fragment aggregation fitting, compression coding, and bidirectional correlation reconstruction techniques using full snapshots.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a data processing method for an energy storage system based on incremental storage of time-series data, comprising the following steps: Step S1: Collect the operating sequence parameters of the energy storage system, perform joint change detection on the operating sequence parameters, calculate the change vector and analyze the projection deviation angle, and divide the operating sequence parameters into changed data and redundant data according to the projection deviation angle; Step S2: Obtain the running state type corresponding to the changed data, aggregate and segment the changed data into state segments using the running state type, fit the trend basis function to each state segment, and output the compressed code and boundary anchor point of each state segment. Step S3: Collect all sampled values ​​to generate a full snapshot, perform dual-copy storage on the full snapshot and establish a bidirectional time pointer between it and the compressed code, and after decoding and reconstructing the compressed code according to the bidirectional time pointer, perform monotonic shape-preserving interpolation on the decoding and reconstruction results based on the boundary anchor points to output the corrected data sequence. Step S4: Locate the rupture time point and generate rupture segment markers based on the corrected data sequence, obtain the compressed code corresponding to the rupture segment markers and output the final data sequence, extract the causal association rules of the over-limit event based on the final data sequence and write them into the feature library.

[0007] In a preferred embodiment, in step S1, the battery pack voltage sequence, charge / discharge current sequence, and individual cell temperature sequence are collected, and the three parameter values ​​are combined into a runtime sequence parameter vector at the same sampling time. Perform a difference operation on the runtime sequence parameter vectors corresponding to two adjacent sampling times to obtain the change vector corresponding to the current sampling time; Normalize the changing vectors separately to obtain normalized changing vectors; Read the normalized change vectors obtained from continuous sampling before the current sampling time, and calculate the direction angle between adjacent normalized change vectors; Adjacent normalized change vectors whose directional angle is less than or equal to a preset directional continuity determination value are arranged to form a directional continuity sequence.

[0008] In a preferred embodiment, in step S2, the unit direction of the most recent normalized change vector with a non-zero magnitude is taken as the predicted direction of the current system trajectory by tracing back in the continuous sequence in the direction of the current sampling time. The angle between the current normalized change vector and the prediction direction is used as the projection deviation angle. When the projection deviation angle is greater than the preset allowable deviation judgment value, the sampled values ​​of the three parameters at the current moment are all marked as changed data; When the projection deviation angle is not greater than the preset allowable deviation judgment value or there is no predicted direction, the absolute value of the change of the three parameters is compared with the change judgment value determined by their respective historical statistics. If it exceeds the limit, it is marked as changed data, and if it does not exceed the limit, it is marked as redundant data.

[0009] In a preferred embodiment, in step S2, three change quantities corresponding to the change data are read, each change quantity is converted into a change direction and a normalized change amplitude, and combined into a state feature vector in a fixed order. Clustering is performed on all state feature vectors, and the cluster category label is used as the running state type at each time. Read the running status type corresponding to each change data in the order of sampling time. If the type is the same at adjacent time points, arrange them continuously. Use the point where the type changes as the dividing point to cut off the continuous arrangement and form multiple status segments. For each state segment, a linear basis and a parabolic basis are fitted respectively. The selected basis function order, the basis function coefficients corresponding to each parameter, the start time stamp and the end time stamp of the state segment are encoded to form a compressed code. The first and last two actual sampling points of the state segment are taken as the first boundary anchor point and the last boundary anchor point, respectively.

[0010] In a preferred embodiment, in step S3, the full sample value is the complete sample value obtained by collecting all operating sequence parameters of the energy storage system at a preset full snapshot acquisition time. Read all sampled values ​​at preset time intervals and write them together with the collection timestamp into a full snapshot; Perform dual-copy storage on the full snapshot, writing the complete sampled value of the full snapshot to the first storage copy and the second storage copy respectively; In the compressed encoding of each state segment, the time position of the most recent full snapshot before its start time is recorded, and a positive time pointer is formed by the compressed encoding pointing to the full snapshot.

[0011] In a preferred embodiment, in step S3, the time position of the first state segment compressed code after each full snapshot is recorded, and a reverse time pointer is formed by pointing from the full snapshot to the compressed code; A bidirectional time pointer is established using forward and reverse pointers; When a recall request for the target time interval is received, the nearest full snapshot before the start is located by the forward pointer of the start time, and the first full snapshot after the end is located by the reverse pointer of the end time. The compressed code between the two snapshots is used as the decoding range. Using the three parameter values ​​from the most recent full snapshot before the decoding range as the initial state for decoding, trend basis function decoding is performed on the state segment according to the order and coefficient of the basis function in each compressed code, so as to obtain the three parameter reconstruction values ​​at each sampling time and form the decoding reconstruction result.

[0012] In a preferred embodiment, in step S3, the first and last boundary anchor points of each state segment are read, the monotonically increasing or monotonically decreasing direction is determined according to the relationship between the actual parameter values ​​of adjacent boundary anchor points, the two end anchor points are used as endpoints, the interpolation result is limited to the closed interval of the two anchor point values, and the interpolation connection is performed in the monotonically increasing direction. The full snapshot before the target interval, the decoding and reconstruction results, the interpolation and connection results, and the full snapshot after the target interval are arranged in chronological order to form a correction data sequence.

[0013] In a preferred embodiment, in step S4, the actual values ​​of the three parameters at each sampling time in the correction data sequence are subtracted from the reconstructed values ​​of the trend basis function of the corresponding state segment to obtain the battery pack voltage reconstruction residual, the charge and discharge current reconstruction residual, and the individual cell temperature reconstruction residual. The absolute value of the reconstructed residuals of each parameter is compared with the residual judgment value determined by the statistics of their respective historical normal state segments, and the residual over-limit sampling points are marked. The sequence of residual out-of-limit sampling points with continuous time was used as a candidate rupture section; The joint reconstruction residual vector Euclidean norm at each time step within the candidate rupture segment is examined to obtain the rupture segment label; Read the bidirectional time pointer based on the start and end times in the broken section markings, locate the compressed code covering the broken section, re-execute the trend basis function decoding with the preceding full snapshot corresponding to the code as the initial state, and re-execute the monotonic conformal interpolation connection with the corresponding boundary anchor points.

[0014] In a preferred embodiment, in step S4, the data in the corresponding time range of the correction data sequence are replaced by the broken segment data obtained by re-decoding and interpolation connection, the data outside the segment is retained and rearranged according to time, and the final data sequence is output. The actual values ​​of the three parameters at each sampling time in the final data sequence are compared with their respective rated operating ranges, and the sampling times that exceed the limits are marked. The consecutive sampling times that exceed the limits are aggregated into an out-of-limit event. Establish the preceding event state sequence and the following event state sequence based on the temporal relationship of adjacent out-of-limit events, and calculate the transition entropy from the preceding event to the following event; The event time interval is calculated based on the end time of the preceding event and the start time of the subsequent event. The statistical distribution of the transfer entropy of the same event type combination in history is read to determine the transfer entropy judgment value. When the transfer entropy is not lower than the judgment value, causal association rules are established and written into the feature library.

[0015] The technical effects and advantages of this invention are as follows: This invention collects runtime timing parameters of an energy storage system, performs joint change detection on these parameters, calculates the change vector and projection deviation angle, divides the runtime timing parameters into variable data and redundant data, aggregates and segments the variable data using operating state types to form state segments, performs trend basis function fitting, outputs compressed codes and boundary anchors, collects full sample values ​​to generate a full snapshot and performs dual-copy storage, establishes a bidirectional time pointer between the full snapshot and the compressed code, completes compressed code decoding and reconstruction based on the bidirectional time pointer, and uses boundary anchors to perform monotonic conformal interpolation connection, locates the rupture time point based on the corrected data sequence, re-acquires compressed codes for the rupture segment, outputs the final data sequence, extracts causal association rules for over-limit events based on the final data sequence and writes them into a feature library, thereby improving the compression storage efficiency and continuous data reconstruction accuracy of time-series data in energy storage systems, realizing incremental storage, rapid retrieval, and abnormal event correlation analysis of operating data, and reducing data storage redundancy and data processing resource consumption. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the implementation of a data processing method for an energy storage system based on incremental storage of time-series data, according to the present invention.

[0017] Figure 2 This is a schematic diagram illustrating the steps of a data processing method for an energy storage system based on incremental storage of time-series data according to the present invention. Detailed Implementation

[0018] The technical solutions of 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.

[0019] This invention collects runtime timing parameters of an energy storage system, performs joint change detection on these parameters, calculates the change vector and projection deviation angle, divides the runtime timing parameters into variable data and redundant data, aggregates and segments the variable data using operating state types to form state segments, performs trend basis function fitting, outputs compressed codes and boundary anchors, collects full sample values ​​to generate a full snapshot and performs dual-copy storage, establishes a bidirectional time pointer between the full snapshot and the compressed code, completes compressed code decoding and reconstruction based on the bidirectional time pointer, and uses boundary anchors to perform monotonically conformal interpolation connection, locates the rupture time point based on the corrected data sequence, re-acquires compressed codes for the rupture segment, outputs the final data sequence, extracts causal association rules for over-limit events based on the final data sequence and writes them into a feature library, thereby improving the compression storage efficiency and continuous data reconstruction accuracy of time series data in energy storage systems.

[0020] Example 1, as Figures 1 to 2 As shown, a data processing method for an energy storage system based on incremental storage of time-series data includes the following steps: Step S1: Collect the operating sequence parameters of the energy storage system, perform joint change detection on the operating sequence parameters, calculate the change vector and analyze the projection deviation angle, and divide the operating sequence parameters into changed data and redundant data according to the projection deviation angle; Step S2: Obtain the running state type corresponding to the changed data, aggregate and segment the changed data into state segments using the running state type, fit the trend basis function to each state segment, and output the compressed code and boundary anchor point of each state segment. Step S3: Collect all sampled values ​​to generate a full snapshot, perform dual-copy storage on the full snapshot and establish a bidirectional time pointer between it and the compressed code, and after decoding and reconstructing the compressed code according to the bidirectional time pointer, perform monotonic shape-preserving interpolation on the decoding and reconstruction results based on the boundary anchor points to output the corrected data sequence. Step S4: Locate the rupture time point and generate rupture segment markers based on the corrected data sequence, obtain the compressed code corresponding to the rupture segment markers and output the final data sequence, extract the causal association rules of the over-limit event based on the final data sequence and write them into the feature library.

[0021] The specific implementation is as follows: In step S1, the operating sequence parameters of the energy storage system are collected, joint change detection is performed on the operating sequence parameters, the change vector is calculated and the projection deviation angle is analyzed, and the operating sequence parameters are divided into changed data and redundant data according to the projection deviation angle.

[0022] Specifically, operational timing parameters are collected during the operation of the energy storage system. These parameters include battery pack voltage sequences, charge / discharge current sequences, and individual cell temperature sequences. These three parameters are sampled synchronously at a uniform sampling period, ensuring that each sampling moment corresponds to one battery pack voltage value, one charge / discharge current value, and one individual cell temperature value. The battery pack voltage sequence reflects the change in the battery pack's terminal voltage over operating time; the charge / discharge current sequence reflects the current changes during charging or discharging; and the individual cell temperature sequence reflects the temperature changes of the individual cell during operation. The three operational timing parameters are arranged according to their sampling moments to form a continuous operational timing parameter sequence, and the three parameter values ​​at the same sampling moment constitute an operational timing parameter vector.

[0023] A difference operation is performed on the runtime sequence parameter vectors corresponding to two adjacent sampling times. The runtime sequence parameter vector at the current sampling time is subtracted from the runtime sequence parameter vector at the previous sampling time to obtain the change vector corresponding to the current sampling time. The change vector includes the change in battery pack voltage, the change in charging and discharging current, and the change in individual cell temperature. A positive value of the change indicates that the corresponding parameter has increased relative to the previous sampling time, and a negative value indicates that the corresponding parameter has decreased relative to the previous sampling time. The absolute value of the change indicates the magnitude of the change of the corresponding parameter within one sampling period.

[0024] To eliminate the dimensional differences between battery pack voltage, charge / discharge current, and individual cell temperature, normalization is performed on the three change vectors. Historical variation statistics of the corresponding parameters within the previous full-capacity snapshot time interval are read. Based on these historical variation statistics, the respective change scales for battery pack voltage, charge / discharge current, and individual cell temperature are determined. The changes in battery pack voltage, charge / discharge current, and individual cell temperature at the current sampling time are divided by their corresponding change scales to obtain the normalized change vectors. When the corresponding change scale is zero, the normalized change component of the current parameter is set to zero to avoid the normalization calculation failing due to a zero change scale.

[0025] The Euclidean norm is calculated for the normalized change vector, and the result is determined as the magnitude of the change vector. The magnitude of the change vector is obtained by taking the square root of the sum of the squares of the three normalized change components. A change vector magnitude of zero indicates that the battery pack voltage, charge / discharge current, and individual cell temperature have not changed in the current sampling period. An increase in the change vector magnitude indicates an increase in the combined change of the three operating sequence parameters.

[0026] Read the normalized change vectors obtained from continuous sampling before the current sampling time, calculate the directional angle between adjacent normalized change vectors, and read the statistical distribution of the directional angles between adjacent normalized change vectors in the historical normal operation data of the energy storage system. Determine the directional continuity judgment value based on the statistical distribution. Arrange adjacent change vectors whose directional angle between adjacent normalized change vectors is less than or equal to the directional continuity judgment value to form a directional continuity relationship. When the directional angle between two adjacent normalized change vectors is greater than the directional continuity judgment value, take the previous normalized change vector corresponding to that angle as the end of the previous directional continuity sequence, and re-establish the directional continuity sequence from the next normalized change vector. Thus, determine the boundary of the directional continuity sequence based on whether the directional angle exceeds the directional continuity judgment value.

[0027] In the continuous sequence of directions corresponding to the current sampling time, the normalized change vector with a non-zero magnitude that is closest to the current sampling time is read backwards from the current sampling time. The unit direction of this normalized change vector is determined as the predicted direction of the current system trajectory. If there is no normalized change vector with a non-zero magnitude before the current time, the predicted direction for the current sampling time is not generated.

[0028] When a predicted direction for the current system trajectory exists, the angle between the current normalized change vector and the predicted direction is calculated, and the resulting angle is determined as the projection deviation angle. The projection deviation angle represents the directional difference between the joint change direction formed by the three runtime sequence parameters at the current sampling time and the predicted direction of the current system trajectory. A projection deviation angle of zero indicates that the two directions are consistent, while an increase in the projection deviation angle indicates an increase in the directional deflection of the current joint change direction relative to the predicted direction.

[0029] The statistical distribution of projection deviation angles in historical normal operation data is read, and an allowable deflection judgment value is determined based on the statistical distribution. When the projection deviation angle at the current sampling time is greater than the allowable deflection judgment value, the battery pack voltage, charging / discharging current, and individual cell temperature at the current sampling time are all marked as changed data.

[0030] When the projection deviation angle at the current sampling time is less than or equal to the allowable deflection judgment value, or when there is no prediction direction at the current sampling time, calculate the absolute value of the battery pack voltage change, the absolute value of the charge / discharge current change, and the absolute value of the individual cell temperature change at the current sampling time, and read the statistical distribution of the absolute values ​​of the corresponding parameter changes in the previous full-capacity snapshot collection time interval. Determine the battery pack voltage change judgment value, the charge / discharge current change judgment value, and the individual cell temperature change judgment value based on the corresponding statistical distribution.

[0031] The absolute values ​​of the changes in the three parameters are compared with the corresponding parameter change judgment values. When the absolute value of the change in battery pack voltage is greater than the corresponding battery pack voltage change judgment value, the current battery pack voltage sample value is marked as changed data; otherwise, it is marked as redundant data. When the absolute value of the change in charging and discharging current is greater than the corresponding charging and discharging current change judgment value, the current charging and discharging current sample value is marked as changed data; otherwise, it is marked as redundant data. When the absolute value of the change in individual cell temperature is greater than the corresponding individual cell temperature change judgment value, the current individual cell temperature sample value is marked as changed data; otherwise, it is marked as redundant data.

[0032] In step S2, the changes in battery pack voltage, charge / discharge current, and individual cell temperature at each sampling time of the change data are read, and these three changes are converted into change direction and normalized change amplitude, respectively. The change direction is determined by the sign of the change: a positive change is recorded as an increasing direction, a negative change as a decreasing direction, and a zero change as a direction with no change. The normalized change amplitude is obtained by dividing the absolute value of the corresponding change in step S1 by the corresponding change scale.

[0033] Following a fixed arrangement of battery pack voltage, charging and discharging current, and individual cell temperature, the direction of change and normalized amplitude of these three parameters are combined to form a state feature vector. Clustering is performed on the state feature vectors corresponding to all the change data. Based on the category labels in the clustering results, the corresponding change data is assigned an operating state type. Change data with the same category label correspond to the same operating state type, and different category labels correspond to different operating state types.

[0034] The operating state type is a discrete classification label obtained by clustering state feature vectors, used to represent the joint change pattern formed by battery pack voltage, charging / discharging current, and individual cell temperature at the same sampling time. The change direction in the state feature vector is used to distinguish the increase or decrease of the three operating sequence parameters, and the normalized change amplitude is used to distinguish the degree of change of the three operating sequence parameters relative to their respective historical change scales, so that the operating state type is jointly determined by the joint change direction and normalized change amplitude of the three operating sequence parameters.

[0035] The operating state type corresponding to each changing data point is read in chronological order of sampling time. When adjacent changing data points have the same operating state type, the corresponding changing data points are arranged consecutively in chronological order of sampling time. When the operating state type of adjacent changing data points changes, the position corresponding to the change in operating state type is determined as the segmentation position. The consecutive arrangement of changing data points is broken at the segmentation position, thereby forming multiple state segments. Each state segment contains changing data points with the same operating state type within a continuous sampling time range.

[0036] For each state segment, a trend basis function fitting is performed. The sampling time of each sampling moment within the state segment is used as the independent variable, and the actual sampled values ​​of the battery pack voltage, charging and discharging current, and individual cell temperature at the corresponding sampling moment are used as the dependent variables. The fitting is performed using a first-order linear basis and a second-order parabolic basis, respectively, to obtain the corresponding first-order linear fitting results and second-order parabolic fitting results.

[0037] Calculate the squared difference between the actual sampled value of the battery pack voltage and the first linear fitting value, the squared difference between the actual sampled value of the charging and discharging current and the first linear fitting value, and the squared difference between the actual sampled value of the individual cell temperature and the first linear fitting value at each sampling time within the state segment. Then, sum all the squared differences to obtain the sum of squared residuals of the first linear basis fitting. The sum of squared residuals of the second parabolic basis fitting is obtained by the same calculation method.

[0038] Subtracting the sum of squared residuals from the sum of squared residuals of the quadratic parabolic basis from the sum of squared residuals of the first-order linear basis yields the reduction in residuals after upgrading the basis function order. Dividing this reduction by the sum of squared residuals of the first-order linear basis gives the marginal gain ratio. This marginal gain ratio represents the relative reduction in the sum of squared residuals after upgrading from a first-order linear basis to a quadratic parabolic basis. An increase in the marginal gain ratio indicates a larger proportion of the reduction in the sum of squared residuals of the first-order linear basis relative to the quadratic parabolic basis.

[0039] Read the statistical distribution of marginal gain ratios corresponding to historical state segments, and determine the marginal gain judgment value based on the statistical distribution. When the marginal gain ratio of the current state segment is greater than or equal to the marginal gain judgment value, select a quadratic parabolic basis as the trend basis function of the current state segment; when the marginal gain ratio of the current state segment is less than the marginal gain judgment value, select a linear basis as the trend basis function of the current state segment.

[0040] Based on the selected trend basis function, the order of the basis function corresponding to the current state segment, the basis function coefficients corresponding to each runtime timing parameter, the start timestamp of the state segment, and the end timestamp of the state segment are encoded in a fixed field order to form a compressed code for the state segment. This allows the trend values ​​of battery pack voltage, charging and discharging current, and individual cell temperature within the state segment to be recovered during decoding.

[0041] The sampling time, actual sampled value of battery pack voltage, actual sampled value of charging / discharging current, and actual sampled value of individual cell temperature corresponding to the first and last actual sampling points of the state segment are read. The sampling time of the first actual sampling point and the actual sampled values ​​of the three parameters are used to form the first boundary anchor point, and the sampling time of the last actual sampling point and the actual sampled values ​​of the three parameters are used to form the last boundary anchor point. The boundary anchor points are used to record the actual sampled values ​​at both ends of the state segment and serve as endpoint constraints when connecting the decoding and reconstruction results in step S3.

[0042] In step S3, the full sample value is the complete sample value obtained by collecting all operating sequence parameters of the energy storage system at the preset full snapshot acquisition time. All operating sequence parameter sample values ​​corresponding to the current full snapshot acquisition time are read according to the preset full snapshot acquisition time interval. The complete sample values ​​of battery pack voltage, charging / discharging current, and individual cell temperature at the current full snapshot acquisition time, along with the full snapshot acquisition timestamp, are written into the full snapshot.

[0043] The generated full snapshots are stored in dual-copy format, with the complete sampled values ​​of each full snapshot written to the first and second storage copies respectively. The first storage copy stores each full snapshot in the order of its acquisition timestamp. The second storage copy stores incremental data using the parameter differences between adjacent full snapshots. Specifically, the battery pack voltage value in the current full snapshot is subtracted from the battery pack voltage value in the previous full snapshot to obtain the battery pack voltage increment; the charge / discharge current value in the current full snapshot is subtracted from the charge / discharge current value in the previous full snapshot to obtain the charge / discharge current increment; and the individual cell temperature value in the current full snapshot is subtracted from the individual cell temperature value in the previous full snapshot to obtain the individual cell temperature increment. These three parameter increments are then associated with the current full snapshot acquisition timestamp to form an incremental snapshot.

[0044] The first full snapshot in the second storage copy is stored completely, while subsequent full snapshots only store the three parameter increments relative to the previous full snapshot and their corresponding timestamps. Based on the complete sampled values ​​of the first full snapshot and the parameter increments in each subsequent incremental snapshot, the sampled values ​​of battery pack voltage, charge / discharge current, and individual cell temperature in subsequent full snapshots are sequentially accumulated and restored in chronological order. This enables the second storage copy to restore all full snapshots based on the first complete full snapshot and subsequent incremental data.

[0045] Read the time correspondence between each full snapshot and the state segment compressed code. Record the time position of the most recent full snapshot before the start time of the state segment in each state segment compressed code. A forward time pointer is formed by pointing from the state segment compressed code to the full snapshot. At the same time, record the time position of the first state segment compressed code after the full snapshot in each full snapshot. A reverse time pointer is formed by pointing from the full snapshot to the corresponding state segment compressed code. A bidirectional time pointer is established between the full snapshot and the state segment compressed code by the forward time pointer and the reverse time pointer.

[0046] When a recall request for the target data time interval is received, the forward time pointer of the corresponding state segment compressed code is read according to the start time of the target data time interval to locate the nearest full snapshot before the start time of the target data time interval; the reverse time pointer of the corresponding full snapshot is read according to the end time of the target data time interval to locate the first full snapshot after the end time of the target data time interval, and the state segment compressed code between the two full snapshots is determined as the decoding range corresponding to the target data time interval.

[0047] Read the most recent full snapshot before the start time of the target data time interval. Use the battery pack voltage value, charge / discharge current value and individual cell temperature value in the full snapshot as the initial state for decoding. Based on the basis function order, basis function coefficients, state segment start time and state segment end time recorded by the compressed encoding of each state segment within the decoding range, perform trend basis function decoding on the corresponding state segment, calculate the reconstructed battery pack voltage value, charge / discharge current value and individual cell temperature value corresponding to each sampling time, and form the decoding and reconstruction result.

[0048] The system reads the first and last boundary anchor points corresponding to each state segment, arranges the decoding and reconstruction results between adjacent boundary anchor points in the order of sampling time, and determines the interpolation direction based on the actual parameter values ​​of the two boundary anchor points. When the actual parameter value of the last boundary anchor point is greater than that of the first boundary anchor point, interpolation is performed on the data between the two boundary anchor points according to a monotonically increasing constraint; when the actual parameter value of the last boundary anchor point is less than that of the first boundary anchor point, interpolation is performed on the data between the two boundary anchor points according to a monotonically decreasing constraint; when the actual parameter values ​​of the two boundary anchor points are equal, the interpolation result of the corresponding parameters between the two boundary anchor points is limited to the same actual value.

[0049] When performing monotonically conformal interpolation, the first and last boundary anchor points are used as interpolation endpoints. The interpolation result is limited to the closed interval formed by the actual parameter values ​​of the two boundary anchor points, and the interpolation result is kept to change monotonically in the direction of parameter change from the first to the last boundary anchor point. This prevents the interpolation result from producing local extrema that exceed the range of parameter values ​​of the two boundary anchor points.

[0050] The corrected data sequence is formed by arranging the full snapshots before the start of the target data time interval, the reconstruction results obtained by decoding the trend basis function, the connection results obtained by monotonic conformal interpolation of the boundary anchors, and the full snapshots after the end of the target data time interval in chronological order of sampling time. The full snapshots in the corrected data sequence retain complete sampled values, and the data within the state segments consists of the corresponding compressed encoding and decoding results and the boundary anchor interpolation results.

[0051] In step S4, the rupture time point is located and rupture segment markers are generated based on the corrected data sequence. The compressed code corresponding to the rupture segment marker is obtained and the final data sequence is output. The causal association rules of the over-limit event are extracted based on the final data sequence and written into the feature library.

[0052] Specifically, the actual values ​​of battery pack voltage, charge / discharge current, and individual cell temperature at each sampling moment in the calibration data sequence are read, and the corresponding reconstructed values ​​of battery pack voltage trend basis function, charge / discharge current trend basis function, and individual cell temperature trend basis function are calculated based on the trend basis function, basis function coefficients, and sampling time of the state segment to which the current sampling moment belongs.

[0053] The battery pack voltage reconstruction residual is obtained by subtracting the corresponding trend basis function reconstruction value from the actual value of the battery pack voltage in the corrected data sequence; the charging and discharging current reconstruction residual is obtained by subtracting the corresponding trend basis function reconstruction value from the actual value of the charging and discharging current; and the individual cell temperature reconstruction residual is obtained by subtracting the corresponding trend basis function reconstruction value from the actual value of the individual cell temperature.

[0054] The statistical distributions of the battery pack voltage reconstruction residual, charge / discharge current reconstruction residual, and individual cell temperature reconstruction residual in historical normal state segments are read respectively. Based on the corresponding statistical distributions, the judgment values ​​for the battery pack voltage residual, charge / discharge current residual, and individual cell temperature residual are determined. The absolute values ​​of the reconstruction residuals of the three parameters at the current sampling time are compared with the corresponding residual judgment values. When any one of the absolute values ​​of the battery pack voltage reconstruction residual, charge / discharge current reconstruction residual, and individual cell temperature reconstruction residual is greater than the corresponding residual judgment value, the current sampling time is marked as a residual exceeding the limit sampling point.

[0055] Residual excess sampling points are read sequentially according to sampling time. A sequence of residual excess sampling points that is temporally continuous and has no non-residual excess sampling points between adjacent sampling points is formed into candidate rupture segments. Three reconstructed residuals corresponding to each sampling time within the candidate rupture segment are read. The battery pack voltage reconstruction residual, charge / discharge current reconstruction residual, and individual cell temperature reconstruction residual are arranged in a fixed order to form a joint reconstruction residual vector, and the Euclidean norm of the joint reconstruction residual vector is calculated.

[0056] The statistical distribution of the Euclidean norm of the joint reconstructed residual vector in historical normal state segments is read, and a rupture determination value is determined based on the statistical distribution. The Euclidean norm of the joint reconstructed residual vector at each sampling time within the candidate rupture segment is checked in sampling time sequence. When the Euclidean norm of the joint reconstructed residual vector at consecutive sampling points is greater than or equal to the rupture determination value, the sampling time in the sequence of consecutive sampling points that first satisfies the condition is determined as the rupture time point. The rupture time point represents the time position at which the joint residual between the joint actual value of the three runtime timing parameters and the corresponding reconstructed value of the trend basis function in the corrected data sequence first satisfies the rupture determination condition.

[0057] Based on the rupture time point, read the continuous residual over-limit sampling points in the candidate rupture segment backward, and read the continuous residual over-limit sampling points in the candidate rupture segment backward. Determine the earliest sampling time in the sequence of continuous residual over-limit sampling points as the start time of the rupture segment, and determine the latest sampling time as the end time of the rupture segment. Then, combine the rupture segment start time, rupture time point, and rupture segment end time into a rupture segment marker according to a fixed field order.

[0058] Based on the start and end times of the broken segment in the broken segment marker, read the corresponding bidirectional time pointers, locate the state segment compressed code covering the broken segment, use the preceding full snapshot corresponding to the located compressed code as the initial state for re-decoding, and re-execute the trend basis function decoding based on the basis function order, basis function coefficients and the start and end times of the state segment in the located compressed code to obtain the re-decoding result corresponding to the broken segment.

[0059] Read the boundary anchor points corresponding to the ruptured section, determine the interpolation directions corresponding to the battery pack voltage, charge / discharge current, and individual cell temperature based on the actual parameter values ​​of the boundary anchor points at both ends of the ruptured section, and perform interpolation connection on the re-decoding results according to the monotonic conformal interpolation method. Replace the data in the time range corresponding to the ruptured section marker in the correction data sequence with the ruptured section data obtained from the re-decoding and interpolation connection, retain the correction data sequence data outside the ruptured section, and rearrange them according to the sampling time order to output the final data sequence.

[0060] The battery pack voltage, charge / discharge current, and individual cell temperature are read from the final data sequence. The rated operating range of the corresponding parameters of the energy storage system is read, and the actual values ​​of the three parameters at each sampling time are compared with the corresponding rated operating range. When any parameter of the battery pack voltage, charge / discharge current, or individual cell temperature exceeds the corresponding rated operating range, the current sampling time is marked as an out-of-limit sampling time. Continuous out-of-limit sampling times are aggregated into out-of-limit events according to the sampling time sequence, and the event type, event start time, and event end time corresponding to each out-of-limit event are recorded.

[0061] The out-of-limit events are sorted according to their start time to form a cross-parameter out-of-limit event sequence. Based on the temporal relationship between adjacent out-of-limit events, a preceding out-of-limit event state sequence and a following out-of-limit event state sequence are established. The probability distributions of the preceding event state, the following event state, and the joint occurrence of the preceding and following event states are then statistically analyzed.

[0062] The transition entropy is calculated based on the probability distribution. The conditional entropy of the subsequent over-limit event state without the preceding over-limit event state is subtracted from the conditional entropy of the subsequent over-limit event state after the preceding over-limit event state is introduced. This difference yields the transition entropy from the preceding over-limit event to the subsequent over-limit event. The transition entropy represents the degree to which the uncertainty of the subsequent over-limit event state is reduced after the preceding over-limit event state is known. An increase in transition entropy indicates an increase in the increment of temporal information provided by the preceding over-limit event state to the subsequent over-limit event state.

[0063] The event time interval is calculated based on the end time of the preceding out-of-limit event and the start time of the subsequent out-of-limit event. The statistical distribution of transfer entropy corresponding to the same combination of preceding and subsequent event types in the historical out-of-limit event sequence is read, and a transfer entropy judgment value is determined based on this statistical distribution. When the transfer entropy from the current preceding out-of-limit event to the subsequent out-of-limit event is greater than or equal to the transfer entropy judgment value, a corresponding causal association rule is established, and the preceding out-of-limit event type, the subsequent out-of-limit event type, the event time interval, and the transfer entropy are jointly written into the feature database.

[0064] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0065] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0066] In this document, the singular forms “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that terms such as “comprising / including” or “having” specify the presence of the stated features, integrals, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, integrals, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0067] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0068] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A data processing method for an energy storage system based on incremental storage of time-series data, characterized in that: Includes the following steps: Step S1: Collect the operating sequence parameters of the energy storage system, perform joint change detection on the operating sequence parameters, calculate the change vector and analyze the projection deviation angle, and divide the operating sequence parameters into changed data and redundant data according to the projection deviation angle; Step S2: Obtain the running state type corresponding to the changed data, aggregate and segment the changed data into state segments using the running state type, fit the trend basis function to each state segment, and output the compressed code and boundary anchor point of each state segment. Step S3: Collect all sampled values ​​to generate a full snapshot, perform dual-copy storage on the full snapshot and establish a bidirectional time pointer between it and the compressed code, and after decoding and reconstructing the compressed code according to the bidirectional time pointer, perform monotonic shape-preserving interpolation on the decoding and reconstruction results based on the boundary anchor points to output the corrected data sequence. Step S4: Locate the rupture time point and generate rupture segment markers based on the corrected data sequence, obtain the compressed code corresponding to the rupture segment markers and output the final data sequence, extract the causal association rules of the over-limit event based on the final data sequence and write them into the feature library.

2. The data processing method for an energy storage system based on incremental storage of time-series data according to claim 1, characterized in that: In step S1, the battery pack voltage sequence, charge / discharge current sequence, and individual cell temperature sequence are collected, and the three parameter values ​​are combined into a runtime sequence parameter vector at the same sampling time. Perform a difference operation on the runtime sequence parameter vectors corresponding to two adjacent sampling times to obtain the change vector corresponding to the current sampling time; Normalize the changing vectors separately to obtain normalized changing vectors; Read the normalized change vectors obtained from continuous sampling before the current sampling time, and calculate the direction angle between adjacent normalized change vectors; Adjacent normalized change vectors whose directional angle is less than or equal to a preset directional continuity determination value are arranged to form a directional continuity sequence.

3. The data processing method for an energy storage system based on incremental storage of time-series data according to claim 2, characterized in that: In step S2, backtracking is performed in the continuous sequence in the direction of the current sampling time, and the unit direction of the most recent normalized change vector with a non-zero magnitude is taken as the predicted direction of the current system trajectory. The angle between the current normalized change vector and the prediction direction is used as the projection deviation angle. When the projection deviation angle is greater than the preset allowable deviation judgment value, the sampled values ​​of the three parameters at the current moment are all marked as changed data; When the projection deviation angle is not greater than the preset allowable deviation judgment value or there is no predicted direction, the absolute value of the change of the three parameters is compared with the change judgment value determined by their respective historical statistics. If it exceeds the limit, it is marked as changed data, and if it does not exceed the limit, it is marked as redundant data.

4. The data processing method for an energy storage system based on incremental storage of time-series data according to claim 3, characterized in that: In step S2, three change quantities corresponding to the change data are read, each change quantity is converted into change direction and normalized change amplitude, and combined into a state feature vector in a fixed order. Clustering is performed on all state feature vectors, and the cluster category label is used as the running state type at each time. Read the running status type corresponding to each change data in the order of sampling time. If the type is the same at adjacent time points, arrange them continuously. Use the point where the type changes as the dividing point to cut off the continuous arrangement and form multiple status segments. For each state segment, a linear basis and a parabolic basis are fitted respectively. The selected basis function order, the basis function coefficients corresponding to each parameter, the start time stamp and the end time stamp of the state segment are encoded to form a compressed code. The first and last two actual sampling points of the state segment are taken as the first boundary anchor point and the last boundary anchor point, respectively.

5. The data processing method for an energy storage system based on incremental storage of time-series data according to claim 1, characterized in that: In step S3, the full sample value is the complete sample value obtained by collecting all operating sequence parameters of the energy storage system at the preset full snapshot acquisition time; Read all sampled values ​​at preset time intervals and write them together with the collection timestamp into a full snapshot; Perform dual-copy storage on the full snapshot, writing the complete sampled value of the full snapshot to the first storage copy and the second storage copy respectively; In the compressed encoding of each state segment, the time position of the most recent full snapshot before its start time is recorded, and a positive time pointer is formed by the compressed encoding pointing to the full snapshot.

6. The data processing method for an energy storage system based on incremental storage of time-series data according to claim 5, characterized in that: In step S3, the time position of the first state segment compressed and encoded after each full snapshot is recorded, and a reverse time pointer is formed by pointing from the full snapshot to the compressed encoding. A bidirectional time pointer is established using forward and reverse pointers; When a recall request for the target time interval is received, the nearest full snapshot before the start is located by the forward pointer of the start time, and the first full snapshot after the end is located by the reverse pointer of the end time. The compressed code between the two snapshots is used as the decoding range. Using the three parameter values ​​from the most recent full snapshot before the decoding range as the initial state for decoding, trend basis function decoding is performed on the state segment according to the order and coefficient of the basis function in each compressed code, so as to obtain the three parameter reconstruction values ​​at each sampling time and form the decoding reconstruction result.

7. The data processing method for an energy storage system based on incremental storage of time-series data according to claim 6, characterized in that: In step S3, the first and last boundary anchor points of each state segment are read, and the monotonically increasing or monotonically decreasing direction is determined according to the relationship between the actual parameter values ​​of adjacent boundary anchor points. The two end anchor points are used as endpoints, the interpolation result is limited to the closed interval of the two anchor point values, and the interpolation connection is performed in the monotonically increasing direction. The full snapshot before the target interval, the decoding and reconstruction results, the interpolation and connection results, and the full snapshot after the target interval are arranged in chronological order to form a correction data sequence.

8. The energy storage system data processing method based on incremental storage of time-series data according to claim 7, characterized in that: In step S4, the actual values ​​of the three parameters at each sampling time in the correction data sequence are subtracted from the reconstructed values ​​of the trend basis function of the corresponding state segment to obtain the battery pack voltage reconstruction residual, the charge and discharge current reconstruction residual, and the individual cell temperature reconstruction residual. The absolute value of the reconstructed residuals of each parameter is compared with the residual judgment value determined by the statistics of their respective historical normal state segments, and the residual over-limit sampling points are marked. The sequence of residual out-of-limit sampling points with continuous time was used as a candidate rupture section; The joint reconstruction residual vector Euclidean norm at each time step within the candidate rupture segment is examined to obtain the rupture segment label; Read the bidirectional time pointer based on the start and end times in the broken section markings, locate the compressed code covering the broken section, re-execute the trend basis function decoding with the preceding full snapshot corresponding to the code as the initial state, and re-execute the monotonic conformal interpolation connection with the corresponding boundary anchor points.

9. A data processing method for an energy storage system based on incremental storage of time-series data according to claim 8, characterized in that: In step S4, the data in the corresponding time range of the correction data sequence are replaced by the data of the broken segment obtained by re-decoding and interpolation connection, the data outside the segment are retained and rearranged according to time, and the final data sequence is output. The actual values ​​of the three parameters at each sampling time in the final data sequence are compared with their respective rated operating ranges, and the sampling times that exceed the limits are marked. The consecutive sampling times that exceed the limits are aggregated into an out-of-limit event. Establish the preceding event state sequence and the following event state sequence based on the temporal relationship of adjacent out-of-limit events, and calculate the transition entropy from the preceding event to the following event; The event time interval is calculated based on the end time of the preceding event and the start time of the subsequent event. The statistical distribution of the transfer entropy of the same event type combination in history is read to determine the transfer entropy judgment value. When the transfer entropy is not lower than the judgment value, causal association rules are established and written into the feature library.