A dynamic time warping-based energy storage consistency detection method and application

Abstract

The application discloses a kind of based on dynamic time warping's energy storage consistency detection method and application, this method involves energy storage system characteristics, data analysis, machine learning method, solve the problem such as difficult to evaluate the consistency of battery cell in energy storage cabin, it is not easy to position inconsistent battery cell, to effectively avoid the problems such as the capacity of inconsistent battery cell in energy storage cabin cannot be fully utilized, the fire of battery cell caused by overcharge and overdischarge in economy, safety aspect.The main steps of the application are as follows: first, collect and arrange the voltage and temperature data of battery cells in the energy storage cabin;Second, according to the voltage change of battery cells, divide the voltage data into charging, discharging and standing periods, and calculate the distance between sample points according to the data;Third, cluster analysis is performed on the voltage data of each battery cell to determine the inconsistency and locate the abnormal battery monomer.The basic idea of the application is to analyze the time sequence parameters of each battery cell in the energy storage cabin to determine the consistency of energy storage and locate the abnormal monomer.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of state monitoring and battery management of energy storage systems, and relates to parameter analysis and abnormality detection of energy storage cabin cells, in particular to an energy storage consistency detection method based on dynamic time warping and application, which is used for realizing consistency evaluation and abnormal cell positioning of cell operation data in an energy storage cabin. BACKGROUND

[0002] In a battery energy storage system, cells are usually connected in series / parallel to form a battery cluster, and then integrated into an energy storage cabin. Due to manufacturing process differences, uneven running environment (such as inconsistent heat dissipation conditions), and different aging rates, inconsistencies in parameters between cells will occur, mainly in the form of dispersion of voltage, capacity, internal resistance, and other characteristics. Inconsistencies not only lead to a decline in the overall performance of the system, but also cause problems such as overcharging and over-discharging of cells, and even thermal runaway, increasing the safety risk of the system. To ensure the safe and reliable operation of the energy storage system, the battery management system has strict regulations on the voltage level of the cells. When the voltage of a single cell in the energy storage system is high, the entire cluster of cells is not allowed to continue charging, and when the voltage of a single cell is low, the entire cluster is not allowed to discharge. Although this control strategy can effectively prevent overcharging and over-discharging of cells, the overall capacity of the energy storage system will be significantly reduced due to the "short board effect" of the cells, resulting in poor economic efficiency and low energy storage utilization. Therefore, it is of great significance to develop consistency detection and abnormal cell positioning technology for energy storage.

[0003] Currently, energy storage consistency detection and fault detection mainly include two types of methods based on mechanism models and data-driven models. The method based on mechanism model mainly analyzes the physical characteristics and parameter differences of the cells by establishing an equivalent circuit model or an electrochemical model to identify inconsistencies. The detection accuracy of this method is highly dependent on the accuracy of the model. However, the practicality of mechanism models is usually poor, resulting in low accuracy of the consistency detection model. The method based on data-driven models is highly dependent on data quality, i.e. the amount of data, the coverage of data, and the availability of fault sample data must reach a certain level before the detection effect can reach a high precision. However, for an energy storage cabin, most of the data are static period data, and the availability of data is poor. Moreover, it is difficult to obtain fault samples. In addition, the state of the energy storage system is a time-accumulative quantity, i.e. the parameters of the cells at a single time point do not have representative significance. By comparing the parameters of the cells over a period of time, since the historical parameters of the cells are not completely consistent with the actual parameters, there may be a time misalignment, resulting in poor consistency detection effect.

[0004] In the prior art, Chinese patent CN119619863A discloses a battery cluster inconsistency detection method based on pressure difference and residual capacity, but only relies on static voltage comparison, cannot solve the time sequence data misalignment problem, and does not consider the influence of temperature and other environmental factors; CN119291551A proposes to detect battery cluster consistency through a resonant circuit, but is only applicable to cluster-level evaluation, and has insufficient positioning accuracy for single-cell-level inconsistency, and high hardware complexity. Both of them have not effectively solved the core problems of mixed multi-working-condition data and dynamic time offset.

[0005] From the application point of view, the existing consistency detection is mostly concentrated in the laboratory level or offline detection for retired batteries, and there are few online detection methods applicable to engineering scenarios. Such methods usually rely on a large number of pre-labeled samples and do not consider the influence of temperature gradient, heat dissipation structure, and cell spatial layout in complex operating environments on the data of the measuring points, so it is difficult to implement deployment in actual energy storage stations.

[0006] In summary, the existing energy storage system consistency detection method still has obvious deficiencies in accuracy, real-time performance, and engineering usability. Therefore, how to construct a consistency detection technology that is applicable to actual energy storage systems, can be implemented online, and has high precision is a technical problem to be solved in the current field. SUMMARY

[0007] (I) Invention purposes

[0008] In view of the above defects and deficiencies of the prior art, the present application takes the energy storage cabin as the research object, aims at the consistency detection index of the cells in the energy storage cabin cluster, and takes the consistency detection reliability as the target, proposes a dynamic time warping-based energy storage consistency detection method and application, constructs a cell historical operation database, divides the charging, discharging and standing states according to the voltage change trend, uses the dynamic time warping algorithm to accurately measure the similarity between voltage data at different times, and introduces the density peak clustering method to realize the unsupervised recognition of the cell consistency state and the positioning of abnormal single cells, effectively avoiding the problems of reduced available capacity caused by inconsistent single cells, overcharging and over-discharging of cells, improving the detection accuracy and robustness, and being applicable to the consistency online evaluation in the actual operation scenario of the energy storage system, providing strong technical support for improving the safety, reliability and operation efficiency of the battery energy storage system.

[0009] (II) Technical solutions

[0010] To achieve the purposes of the application and solve its technical problems, the application adopts the following technical solutions:

[0011] The first inventive objective of the present application is to provide a dynamic time warping-based energy storage consistency detection method for online monitoring and abnormal cell positioning of the consistency state of multiple cells in an energy storage cabin based on historical operation data.

[0012] S100. Multi-source data acquisition and preprocessing:

[0013] Collect and organize the operation data of each cell in the energy storage cabin, including cell voltage data and temperature data of each battery cluster temperature measurement point, and synchronously record the physical location and ambient temperature data of the cells, organize and screen the data according to the preset time granularity, and construct a historical database indexed by time stamp;

[0014] S200. State recognition and data classification:

[0015] According to the change of the voltage of each cell with time in the historical database, the working state of each cell is recognized according to the mode of voltage rise, fall or stability, the cell operation data is divided into charging, discharging and standing state data, and charging, discharging and standing state databases are constructed respectively;

[0016] S300. DTW distance calculation and similarity measurement:

[0017] For each cell in each state database, based on the voltage time series data, the dynamic time warping (DTW) algorithm is used to calculate the DTW distance between each cell and other cells, and the DTW distance matrix between each pair of cells is obtained to measure the similarity of the voltage time series data between the cells and represent the time series consistency feature between the cells;

[0018] S400. Cluster analysis and anomaly recognition:

[0019] Based on the obtained DTW distance matrix between each pair of cells, unsupervised cluster analysis of all cells is performed based on cell voltage data and using density peak clustering algorithm, the local density and relative distance of each cell are calculated, a decision diagram is constructed and cluster centers and outliers are recognized;

[0020] S500. Abnormal cell positioning and output:

[0021] According to the outlier number in the decision diagram, the corresponding abnormal cell recognition result is output, the positioning of the abnormal cell in the energy storage cabin is realized combined with the physical location data of the cells, and the consistency detection result and abnormal cell warning information are output to support subsequent maintenance scheduling and safety control strategy formulation of the energy storage system.

[0022] The second inventive objective of the present application is to provide a computer program product comprising computer instructions for executing the above-mentioned dynamic time warping-based energy storage consistency detection method.

[0023] The third inventive objective of the present application is to provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-mentioned dynamic time warping-based energy storage consistency detection method.

[0024] (III) Technical Effects

[0025] Compared with the prior art, the dynamic time warping-based energy storage consistency detection method and application proposed by the present application has the following beneficial and significant technical effects:

[0026] (1) The present application is based on an energy storage cabin, and according to the historical data of the voltage and temperature of the battery cell in the cabin, a clustering algorithm is used to analyze the inconsistency of the battery cell and locate the abnormal battery monomer. In addition, the present application effectively reduces the distance error caused by time misalignment in distance calculation by using methods such as dynamic time warping (DTW). This method allows the waveform between battery cells to be stretched and matched in the time dimension, significantly improving the accuracy and robustness of the comparison of battery cell behavior in the full time domain, and is particularly suitable for working conditions in actual energy storage systems where there are differences in charging and discharging speed and environmental disturbances.

[0027] (2) The present application combines the density peak clustering algorithm to construct an unsupervised consistency discrimination model, automatically identifies the consistency clustering center and outlier battery cells by constructing a density-relative distance decision diagram, and combines physical location mapping to achieve accurate positioning and hierarchical early warning of abnormal monomers, effectively improving the accuracy of consistency detection, avoiding false labeling of normal battery cells, and thus effectively reducing the maintenance cost of the energy storage cabin, improving the safety of the energy storage cabin operation, and avoiding the occurrence of large-scale accidents in the energy storage cabin. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The flowchart of the dynamic time warping-based energy storage consistency detection method of the present application;

[0029] Figure 2 The schematic diagram of the dynamic time warping-based energy storage consistency detection model architecture of the present application. DETAILED DESCRIPTION

[0030] The application aims to provide a dynamic time warping-based energy storage consistency detection method and application, which is used for monitoring the consistency state of multiple battery cells in an energy storage cabin and positioning abnormal single cells based on operation data. In order to make the purpose, technical scheme and advantages of the application clearer, the technical scheme in the application examples will be described in more detail below in combination with the drawings in the application examples. The described examples are part of the examples of the application, not all examples, and the described examples are exemplary and are intended to explain the application, and cannot be understood as a limitation of the application.

[0031] Embodiment 1: Energy storage consistency detection method

[0032] The application will be described in detail below in combination with Figure 1 and Figure 2 specific embodiments. As shown in Figure 1 , 2 , the dynamic time warping-based energy storage consistency detection method of the application includes the following steps when implemented:

[0033] S100. Multi-source data acquisition and preprocessing:

[0034] Collect and arrange the operation data of each battery cell in the energy storage cabin, including battery cell voltage data and temperature data of each battery cluster temperature measuring point, and synchronously record the physical position and environmental temperature data of the battery cell, arrange and screen the data according to the preset time granularity, and construct a historical database indexed by time stamp.

[0035] As a preferred embodiment, the multi-source data acquisition and preprocessing in this step includes the following sub-steps when implemented:

[0036] S110. Voltage data acquisition and processing: Collect the voltage data of each battery cluster battery cell in the energy storage cabin, time-align the voltage data according to a uniform time interval (such as 1 second), use high-pass filtering to denoise the voltage data with large fluctuations during state conversion, remove error data introduced by transient disturbance, measuring point jitter or sampling error, and interpolate and complete the abnormal missing values, clean the outliers, and form a continuous and stable voltage time series data sequence;

[0037] S120. Temperature data acquisition and conversion: Synchronously collect the environmental temperature data and the temperature data of each battery cluster temperature measuring point in the energy storage cabin, arrange and screen the temperature data, remove error data, and the temperature measuring point data is closely related to the factors such as the physical position of the battery cell and the environmental temperature, and the influence of the heat field on the behavior of the battery cell is considered, so the difference between the measuring point temperature data and the environmental temperature data is used as a variable to form a temperature difference variable for describing the local heat conduction condition, and the physical position relationship between each temperature measuring point and the corresponding battery cell is recorded as an environmental correction factor for subsequent clustering auxiliary judgment and abnormal positioning;

[0038] S130. History database construction: the screened and preprocessed voltage data and temperature data are constructed into a history database with time stamp, battery cell number, corresponding battery cell voltage data, temperature measurement point data, temperature difference between measurement point and environment, and battery cell physical location as variables.

[0039] Further, in the sub-step S120, the battery cell temperature data processing at different positions in the energy storage cabin includes:

[0040] S121. Position information collection: physical position information of the battery cells in the energy storage cabin is collected, including battery cluster number, layer number and position coordinate in the battery cluster, to ensure accurate thermal variable space mapping;

[0041] S122. Temperature model construction: a battery cell position-temperature relationship model is constructed in combination with the battery cell physical position and temperature measurement point data, which can be modeled by linear fitting, interpolation fitting or thermal diffusion approximation function, for restoring thermal variables that are difficult to directly collect positions;

[0042] S123. Heat dissipation condition correction: based on the position-temperature relationship model, the temperature data is corrected for heat dissipation conditions to eliminate temperature differences caused by different heat dissipation conditions and improve the comparability of the temperature data.

[0043] S200. State recognition and data classification:

[0044] According to the change of the voltage of each battery cell with time in the history database, the working state of each battery cell is recognized according to the voltage rising, falling or stable mode, the battery cell operation data is divided into charging, discharging and standing state data, and charging, discharging and standing state databases are constructed respectively.

[0045] As a preferred, in the step S200, the state recognition and data classification includes the following sub-steps:

[0046] S210. Voltage change rate calculation: based on the voltage time series data of each battery cell in the history database, the voltage change rate between adjacent time points is calculated by using a sliding window method;

[0047] S220. Working state determination: the working state of the battery cell is determined according to the voltage change rate, when the voltage continuously rises and the change rate is greater than the rising threshold, it is determined as the charging state; when the voltage continuously falls and the absolute value of the change rate is greater than the falling threshold, it is determined as the discharging state; when the absolute value of the voltage change rate is less than or equal to the standing threshold, it is determined as the standing state;

[0048] S230. Data classification storage: according to the judgment result, the time series data of each battery in the historical database is respectively stored in the charging, discharging and standing state databases, and the start and end time periods, battery number and associated temperature information of the corresponding state are recorded, a multi-state classification data set is constructed, and a clear structure and explicit semantics input basis is provided for subsequent DTW distance calculation.

[0049] S300. DTW distance calculation and similarity measurement:

[0050] For each battery in each state database, based on the voltage time series data, the DTW distance between each battery and other batteries is calculated, and the DTW distance matrix between each battery pair is obtained, which is used to measure the similarity of the voltage time series data between the batteries and represent the time series consistency feature between the batteries.

[0051] As preferred, in step S300, the DTW distance calculation and similarity measurement includes the following sub-steps:

[0052] S310. Data normalization processing:

[0053] In order to eliminate the influence of the difference in amplitude scale of different battery voltage data on the distance calculation result, the voltage time series data of each battery in the state database is normalized. The normalization adopts the minimum-maximum scaling method, and the algorithm formula is Where V i (t) represents the original voltage value of the i-th battery at time t, V i norm V n (t) is the normalized voltage value, V m and V i are the minimum and maximum values of its voltage sequence, respectively;

[0054] S320. DTW distance calculation:

[0055] For any two batteries' normalized voltage time series data P = {p1, p2, …, p n} and Q = {q1, q2, …, q m}, where n and m are the sequence lengths, allow them to be nonlinearly stretched and deformed on the time axis, use the DTW algorithm to find the optimal alignment path of the waveforms to minimize the cumulative distance between them, and obtain the DTW distance between them based on the best alignment path π is a legal alignment path that satisfies the boundary condition, d(p i , q j ) = (p i -q j ) 2 is the element p i and q jEuclidean distance between the i-th and j-th points, i∈[1, n], j∈[1, m], the waveform optimal alignment path is obtained by solving the state transition equation D(i, j) = min{D(i-1, j-1), D(i-1, j), D(i, j-1)} + d(p i j ) in dynamic programming, the initial boundary condition in the dynamic programming process is set as dp(0, 0) = 0, dp(i, 0) = dp(0, j) = ∞;

[0056] S330. DTW distance matrix construction:

[0057] The similarity DTW distance between all the voltage time series data between the cells in the same state database (charging, discharging or standing) is calculated pair by pair using the DTW calculation method, and the results are recorded in the symmetric distance matrix M, which is used to comprehensively depict the dynamic response similarity degree between the cells.

[0058] S400. Cluster analysis and anomaly identification:

[0059] Based on the obtained DTW distance matrix between each pair of cells, the density peak clustering algorithm is used to perform unsupervised cluster analysis on all the cells based on the cell voltage data, the local density and relative distance of each cell are calculated, the decision graph is constructed, and the cluster center and outliers are identified.

[0060] As preferred, in step S400, the density peak clustering algorithm is used for clustering, and the cluster analysis and anomaly identification includes the following sub-steps:

[0061] S410. Local density calculation: based on the DTW distance matrix, for each cell point, the cell parameter is set as the center, the distance cutoff threshold is set, the number of adjacent cell points within the cutoff threshold is counted to obtain the local density, and the calculation formula is wherein d ij is the DTW distance between the i-th and j-th cells, d c is a preset cutoff distance threshold;

[0062] S420. Relative distance calculation: for each cell point i, the DTW distance between the cell point i and the nearest point with higher density than the cell point i is calculated as the relative distance δ i , and for the cell point with the highest density, the relative distance δ i is defined as the DTW distance between the cell point and the farthest point, and the related calculation formulas are respectively:

[0063]

[0064] S430. Decision graph construction and outlier identification: the local density ρ​i The horizontal axis represents the relative distance δ. i Using the vertical axis, a cluster distribution of all battery cells is plotted on a two-dimensional plane to construct a decision map. Battery cells with high density and large relative distance in the upper left region of the decision map are regarded as cluster centers, and battery cells with low density and large relative distance in the lower right region of the decision map are regarded as outliers. Faulty battery cells are located based on the outlier numbers.

[0065] Furthermore, in sub-step S430, the identification of cluster centers and outliers includes the following sub-steps:

[0066] S431. Comprehensive score calculation: Calculate the local density ρ i and relative distance δ i The product γ i =ρ i ×δ i This serves as a comprehensive score for the degree of abnormality in the battery cell;

[0067] S432. Cluster center determination: according to γ i Sort the cells from largest to smallest value, and select γ. i The cells with the highest values ​​are selected as cluster centers;

[0068] S433. Outlier Detection: Set a relative distance threshold δ c and local density threshold ρ c When cell i satisfies δ i >δ c And ρ i <ρ c When this happens, the cell is identified as an outlier, i.e., an abnormal cell.

[0069] S500 Abnormal Cell Location and Output:

[0070] Based on the outlier numbers in the decision map, the system outputs the corresponding abnormal cell identification results and combines the cell's physical location data to locate the abnormal cell within the energy storage compartment. It also outputs consistency detection results and abnormal cell early warning information to support the subsequent maintenance scheduling and safety control strategy formulation of the energy storage system.

[0071] Preferably, step S500, the abnormal cell location and output, includes the following sub-steps:

[0072] S510. Abnormal Cell Identification: Based on the constructed density-relative distance decision map, cell points located in low-density areas and with large relative distances are identified as outliers. Combined with the preset outlier identification threshold, cell numbers that meet the conditions are selected as suspected abnormal individuals. This process does not rely on manually setting the number of clusters, and achieves unsupervised automatic determination of consistency degradation trends.

[0073] S520. Physical location positioning: match the abnormal cell number with the cell physical location data to determine the specific location of the abnormal cell in the energy storage cabin, output its physical arrangement coordinates in the energy storage cabin (such as battery cluster number, row and column position or rack number), support the graphical interface to visualize and label the fault point position in three-dimensional or two-dimensional space, and provide intuitive positioning basis for maintenance personnel;

[0074] S530. Abnormal degree scoring: for each identified abnormal cell, based on its relative distance δ i , local density p i and the minimum DTW distance from the cluster center, an abnormal scoring index is constructed, and the higher the score value, the more serious the abnormality;

[0075] S540. Risk level classification and early warning: according to the abnormal degree score, the abnormal cells are divided into high risk, medium risk and low risk three levels, and the consistency detection result and grading warning information are output.

[0076] S600. Consistency evaluation and capacity optimization, including:

[0077] S610. Consistency score: internal consistency evaluation is performed on the cell cluster members represented by each cluster center. A consistency scoring model can be constructed based on the group DTW mean distance, outlier proportion or density gradient change, etc. to give each cluster a numerical score, quantifying its internal performance balance;

[0078] S620. Cell grouping reconstruction: according to the scoring results, the original battery cluster configuration is optimized, and the cells with high consistency score and similar time sequence behavior characteristics are recombined into new battery clusters. This reconstruction scheme can be used for manufacturing screening in the design stage, or for allocation and control in the system operation and maintenance stage;

[0079] S630. Charge and discharge strategy optimization: according to the characteristics of the new grouping, different charge and discharge strategies are configured for different cell clusters. High consistency clusters can use standard strategies to improve efficiency, and low consistency clusters can use flexible strategies to suppress degradation and ensure safe operation. The strategy can realize adaptive control in combination with SOC dynamic window adjustment, constant current-constant voltage joint control, etc.

[0080] S640. Dynamic adjustment and update: periodically run the detection process of S300-S500 to update the consistency score and clustering division results, dynamically adjust the system operation parameters and maintenance strategies according to the monitoring trend, form a data-driven battery group life cycle management mechanism, and improve the whole life cycle performance of the energy storage system.

[0081] Through the above steps, the application not only realizes high-precision monitoring and abnormal identification of the consistency state of the energy storage battery cell, can effectively cover the dynamic response characteristic changes under multiple state working conditions, and improves the accuracy and sensitivity of the cell abnormal detection; at the same time, with the help of the consistency evaluation mechanism and the visual positioning method based on the clustering structure, the digital management ability of the energy storage battery is further promoted. The method can realize state perception, strategy adjustment and capacity optimization of the energy storage system in different running stages, significantly improve the operation reliability, safety and energy efficiency level of the system, and provide strong technical support and scientific decision basis for intelligent scheduling and long-period operation and maintenance of large-scale energy storage systems.

[0082] Embodiment 2: Application example

[0083] On the basis of the above-mentioned embodiment 1, as a more specific application example, this embodiment 2 shows the specific application of the above-mentioned method of the application in a certain energy storage power station. Specifically, in a certain 100MWh lithium battery energy storage power station, the energy storage cabin is composed of 20 battery clusters, each battery cluster contains 200 series-connected lithium iron phosphate battery cells (nominal voltage 3.2V, capacity 280Ah). The operating environment temperature of the power station ranges from-10℃ to 40℃, and a BMS system is equipped to collect battery cell voltage and temperature data with a granularity of 1 second. This embodiment applies the above-mentioned method of the application to conduct consistency detection on the 5th battery cluster (containing 200 battery cells), and the specific implementation process is as follows:

[0084] S100. Multi-source data acquisition and preprocessing

[0085] Voltage data acquisition: Collect 24 hours of battery cell voltage data, the original data contains 5 charging cycles (0.3C constant current charging to 3.65V), 3 discharging cycles (0.5C constant current discharging to 2.5V) and a standing period. The voltage fluctuation data (amplitude jump>0.5V) at the charging-to-discharging instant is high-pass filtered (cutoff frequency 15Hz), and after filtering out the noise, the missing value filling rate is≤0.1%.

[0086] Temperature data processing: The environmental temperature data range is 8℃-32℃, the internal measuring point temperature of the battery cluster is as high as 45℃ (center position) and as low as 38℃ (edge position). Correct the temperature according to the position of the battery cell: the heat dissipation weight k of the center battery cell is 0.9, and the weight k of the edge battery cell is 1.1. The corrected temperature difference variable range is 3-7℃.

[0087] History database construction: Storage fields include timestamp, battery cell number (CL5-001 to CL5-200), voltage (2.5V-3.65V), corrected temperature difference, position coordinates (X / Y / Z).

[0088] S200. State identification and data classification

[0089] State determination result: charging state data proportion 32% (voltage rise slope ≥ 0.012V / s); discharge state data proportion 28% (voltage drop slope ≤ -0.015V / s); standing state data proportion 40% (voltage fluctuation ≤ 0.03V).

[0090] Classification data set: charging database contains 6400 time series records, discharge database 5600, and standing database 8000.

[0091] S300. DTW distance calculation and similarity measurement

[0092] Normalization processing: charging state voltage is normalized to [0, 1], and discharge state is normalized to [0.2, 0.8] to avoid cross-state interference.

[0093] DTW distance calculation: Sakoe-Chiba Band constraint (path bending width is 10% of time series length) is used to calculate the DTW distance between cells. The DTW distance between the charging curves of cells CL5-048 and CL5-132 is 1.24 (after normalization), and the distance between the discharge curves is 0.87.

[0094] DTW distance matrix construction: a 200x200 symmetric matrix is generated, with the maximum distance being 2.15 (CL5-003 and CL5-197) and the minimum distance being 0.12 (CL5-045 and CL5-046).

[0095] S400. Clustering analysis and anomaly identification

[0096] Parameter setting: cutoff distance dc = 0.65, local density calculation uses Gaussian weighting.

[0097] Decision graph analysis: cluster centers are cells CL5-045 (p = 18.7, d = 0.21) and CL5-089 (p = 16.2, d = 0.25); outliers are cells CL5-112 (p = 2.1, d = 1.98p = 2.1, d = 1.98) and CL5-156 (p = 3.4, d = 1.75p = 3.4, d = 1.75).

[0098] Second verification: the DTW-Var of cell CL5-112 charging and discharging curve is 0.56 (normal cell average is 0.18), confirming it as an anomaly.

[0099] S500. Abnormal cell positioning and output

[0100] Anomaly score and alarm: cell CL5-112 scores 1.25 (corresponding to high risk, red alarm), and CL5-156 scores 0.92 (corresponding to medium risk, yellow warning).

[0101] Three-dimensional visual positioning: CL5-112 is located in the 3rd layer and 12th column of the battery cluster, and the voltage difference between adjacent cells CL5-111 and CL5-113 is >0.2V (normal difference <0.05V).

[0102] Maintenance response: offline detection is performed on CL5-112, and the internal resistance is found to be increased to 12mΩ (normal value 8mΩ), and the capacity is attenuated to 82% (SOH), and after replacement, the system capacity utilization is improved by 6.3%.

[0103] S600. Consistency evaluation and capacity optimization

[0104] Reorganization strategy implementation: the cells with the top 50% consistency scores (CL5-001 to CL5-100) are reorganized into a new battery cluster, and the capacity difference is reduced from 5.7% to 1.2%.

[0105] Charge-discharge strategy adjustment: the charging rate of high-risk clusters is reduced to 0.3C, and the overall cycle efficiency of the system is improved to 94.5% (originally 92.1%).

[0106] Implementation effect verification: abnormality detection accuracy, accurately identifying 3 abnormal cells (CL5-112, CL5-156, CL5-178), false positive rate <1% (2 false positive points are excluded after review). Economic improvement, the effective capacity of the system is increased from 89.4MWh to 93.1MWh. Safety guarantee, after replacing the abnormal cells, the maximum temperature of the battery cluster is reduced by 4℃, and no thermal runaway event occurs.

[0107] Through the above examples, the purpose of the present application is completely and effectively achieved. Those skilled in the art can understand that the present application includes but is not limited to the contents described in the drawings and the above specific embodiments. Although the present application has been described with respect to the presently preferred and most advantageous embodiments, it should be understood that the present application is not limited to the disclosed embodiments, and any modification that does not deviate from the functional and structural principles of the present application will be included in the scope of the claims.

Claims

1. A dynamic time warping-based energy consistency detection method, characterized in that, Comprise: S100. Multi-source data collection and preprocessing: Collect and organize the operation data of each battery cell in the energy storage tank, including battery cell voltage data and temperature data of each battery cluster temperature measuring point, record the physical location and environmental temperature data of the battery cell synchronously, arrange and screen the data according to the preset time granularity, and construct a historical database; S200. State recognition and data classification: According to the change of voltage of each battery cell with time in the historical database, the working state of each battery cell is recognized according to the mode of voltage rising, falling or stabilizing, and the battery cell data is divided into charging, discharging and standing states, and charging, discharging and standing state databases are constructed respectively; S300. DTW distance calculation and similarity measurement: For each battery cell in each state database, based on the voltage time series data, the DTW algorithm is used to calculate the distance between each battery cell and other battery cells, and the DTW distance matrix between each battery cell pair is obtained to measure the similarity of the voltage time series data between the battery cells; S400. Cluster analysis and anomaly identification: Based on the obtained DTW distance matrix, unsupervised cluster analysis is performed on all battery cells based on the voltage data of the battery cells and using the density peak clustering algorithm, the local density and relative distance of each battery cell are calculated, a decision graph is constructed, and the cluster center and outliers are identified; S500. Abnormal battery cell positioning and output: According to the outlier number in the decision graph, the corresponding abnormal battery cell identification result is output, the positioning of the abnormal battery cell in the energy storage tank is realized combined with the physical location data of the battery cell, and the consistency detection result and abnormal battery cell warning information are output.

2. The dynamic time warping-based energy consistency detection method of claim 1, wherein, In step S100, multi-source data collection and preprocessing includes the following sub-steps: S110. Voltage data collection and processing: Collect the voltage data of each battery cell in the energy storage tank, time align the voltage data according to a uniform time interval, use high-pass filtering to denoise the voltage data with large fluctuations during state transition, remove error data, and interpolate missing values, clean outliers, and form continuous and stable voltage time series data sequences; S120. Temperature data collection and conversion: Collect environmental temperature data and temperature data of each battery cluster temperature measuring point in the energy storage tank synchronously, arrange and screen the temperature data, remove error data, and use the difference between the temperature data of the measuring point and the environmental temperature data as a variable; S130. Historical database construction: The voltage data and temperature data after screening and preprocessing are used to construct a historical database, with time stamp, battery cell number, corresponding battery cell voltage data, temperature measuring point data, temperature difference between measuring point and environment, and battery cell physical location as variables.

3. The dynamic time warping-based energy consistency detection method of claim 2, wherein, In sub-step S120, the temperature data processing for battery cells at different positions in the energy storage tank includes: S121. Position information collection: Collect the physical location information of the battery cells in the energy storage tank, including the battery cluster number, layer number and position coordinates in the battery cluster where the battery cell is located; S122. Temperature model construction: Construct a battery cell position-temperature relationship model based on the physical location and temperature measuring point data of the battery cell; S123. Heat dissipation condition correction: Based on the position-temperature relationship model, correct the temperature data based on the heat dissipation condition, eliminate the temperature difference caused by different heat dissipation conditions, and improve the comparability of the temperature data. 4.The dynamic time warping-based energy consistency detection method of claim 1, wherein, In step S200, state recognition and data classification includes the following sub-steps: S210. Voltage rate of change calculation: based on the voltage time series data of each battery cell in the historical database, the voltage rate of change between adjacent time points is calculated using a sliding window method; S220. Working state determination: determine the working state of the battery cell according to the voltage rate of change, when the voltage continues to rise and the rate of change is greater than the rising threshold, it is determined to be in the charging state; when the voltage continues to fall and the absolute value of the rate of change is greater than the falling threshold, it is determined to be in the discharging state; when the absolute value of the voltage rate of change is less than or equal to the resting threshold, it is determined to be in the resting state; S230. Data classification storage: according to the determination result, the time series data of each battery cell in the historical database is stored in the charging, discharging and resting state databases respectively, and a multi-state classification data set is constructed.

5. The dynamic time warping-based energy consistency detection method of claim 1, wherein, In step S300, DTW distance calculation and similarity measurement includes the following sub-steps: S310. Data normalization processing: The cell voltage time series data in each state database is normalized by using the minimum-maximum scaling method, and the algorithm formula is Wherein V i (t) represents the original voltage value of the i th cell at time t, V i norm (t) is the normalized voltage value, The minimum and maximum values of the voltage sequence are respectively S320. DTW distance calculation: P = {p1, p2, …, pn} and Q = {q1, q2, …, qm} are the normalized voltage time series data of any two cells, where n, m are the sequence lengths, allowing them to be nonlinearly stretched and deformed on the time axis, the DTW algorithm is used to find the optimal alignment path of the waveforms to minimize the cumulative distance between the two, and the DTW distance between the two is obtained based on the optimal alignment path n} and Q = {q1, q2, …, q m} are the normalized voltage time series data of any two cells, where n, m are the sequence lengths, allowing them to be nonlinearly stretched and deformed on the time axis, the DTW algorithm is used to find the optimal alignment path of the waveforms to minimize the cumulative distance between the two, and the DTW distance between the two is obtained based on the optimal alignment path π is a legal alignment path that satisfies the boundary conditions, d(p i , q j ) = (p i - q j ) 2 is the Euclidean distance between elements p i and q j , i ∈ [1, n], j ∈ [1, m], the optimal alignment path of the waveforms is obtained by solving the state transition equation D(i, j) = min{D(i-1, j-1), D(i-1, j), D(i, j-1)} + d(p i , q j ), the initial boundary conditions are set as dp(0,0) = 0, dp(i,0) = dp(0,j) = ∞; S330. DTW distance matrix construction: For the voltage time series data between all battery cells in the same state database, the similarity DTW distance is calculated pair by pair using the DTW calculation method, and the result is recorded in the symmetric distance matrix M.

6. The dynamic time warping-based energy consistency detection method of claim 1, wherein, In step S400, clustering analysis and anomaly identification includes the following sub-steps: S410. Local density calculation: based on the DTW distance matrix, for each battery point, set the distance threshold, and count the number of adjacent battery points within the threshold range to obtain the local density, the calculation formula is wherein d ij is the DTW distance between the i-th and j-th batteries, d c is the preset distance threshold. S420. Relative distance calculation: for each cell point i, calculate its DTW distance to the closest one among all the points with higher density than it, as its relative distance δ i For the cell point with the highest density, its relative distance δ i is defined as its DTW distance to the farthest one among all the other points, and the relevant calculation formulae are respectively: S430. Decision graph construction and outlier identification: with local density p of each cell point i Relative distance d is the horizontal axis i Relative distance d is the horizontal axis i All cell clustering distribution is plotted on two-dimensional plane to construct decision graph, cell points with high density and large relative distance in the upper left area of decision graph are considered as clustering centers, cell points with low density and large relative distance in the lower right area of decision graph are considered as outliers, and faulty cell is located according to outlier number.

7. The dynamic time warping-based energy consistency detection method of claim 6, wherein, In sub-step S430, the identification of cluster centers and outliers includes: S431. Comprehensive score calculation: calculate the local density p i and the product y of the relative distance d i i = p i x d i as the comprehensive score of the degree of abnormality of the battery cell;​ S432. Cluster center determination: sort the battery cells by γ i value in descending order, and select the battery cells with the largest γ i values as the cluster centers; S433. Outlier determination: set relative distance threshold δ c and local density threshold ρ c , when the cell i satisfies δ i > δ c and ρ i < ρ c , determine that the cell is an outlier, i.e., an abnormal cell. 8.The dynamic time warping-based energy consistency detection method of claim 1, wherein, In step S500, abnormal battery cell positioning and output includes the following sub-steps: S510. Abnormal battery cell identification: based on the constructed density-relative distance decision diagram, the battery cell points located in the low density area and with large relative distance are determined as outliers, and the battery cell numbers that meet the conditions are screened out as suspected abnormal monomers combined with the preset outlier identification threshold; S520. Physical location positioning: associate and match the abnormal battery cell number with the physical location data of the battery cell to determine the specific location of the abnormal battery cell in the energy storage cabin, and output the physical arrangement coordinates of the abnormal battery cell in the energy storage cabin; S530. Abnormality degree score: for each identified abnormal cell, based on its relative distance δ in the decision graph i , local density p i and the minimum DTW distance from the cluster center to construct an abnormal score indicator, the higher the score value, the more serious the abnormality degree; S540. Risk level classification and early warning: according to the abnormal degree score, the abnormal battery cell is divided into three levels of high risk, medium risk and low risk, and the consistency detection result and grading warning information are output.

9. The dynamic time warping-based energy consistency detection method of claim 1, wherein, The method further includes step S600 of performing consistency evaluation and capacity optimization, wherein: S610. Consistency score: based on the clustering result, the consistency score of the battery cell group represented by each cluster center is calculated; S620. Battery cell grouping reconstruction: according to the consistency score, the battery cells in the energy storage cabin are grouped, and the battery cells with high consistency are grouped into battery clusters; S630. Charging and discharging strategy optimization: according to the consistency characteristics of the battery cell group, the charging and discharging control strategy is optimized to improve the effective capacity utilization rate of the energy storage system; S640. Dynamic adjustment and update: regularly update the battery cell consistency evaluation result, and dynamically adjust the operating parameters and maintenance plan of the energy storage system.

10. A computer program product, characterised in that, Computer instructions for executing the energy storage consistency detection method based on dynamic time warping of any one of claims 1-9. Computer instructions for executing the energy storage consistency detection method based on dynamic time warping of any one of claims 1-9.

Images