Battery multi-parameter time alignment, battery feature extraction and battery anomaly detection method

Abstract

The invention belongs to the field of batteries, and particularly discloses a battery multi-parameter time alignment method, a battery feature extraction method and a battery anomaly detection method, and the time alignment method comprises the steps: obtaining the current time series data of a battery cluster, and the voltage time series data of each battery cell to which the battery cluster belongs; reconstructing the voltage time sequence data of each battery cell based on the current time sequence data of the battery cluster by adopting an improved DTW algorithm, so that the feature points of the reconstructed voltage time sequence data are aligned with those of the current time sequence data; a distance measurement matrix of the improved DTW algorithm is determined by a physical distance factor and a time distance factor; the physical distance factor is determined according to the current value of the current time sequence data, the voltage value of the voltage time sequence data, the open-circuit voltage of the battery cell and the equivalent internal resistance, and the time distance factor is determined according to the time of the current time sequence data and the voltage time sequence data. According to the invention, the operation and maintenance efficiency of the energy storage system is greatly improved.

Description

Technical Field

[0001] The present application belongs to the field of batteries, and more specifically, relates to a method for battery multi-parameter time alignment, battery feature extraction, and battery anomaly detection. Background Art

[0002] As energy storage systems scale, the number of battery cells in a single station has reached hundreds of thousands, posing significant challenges to their safety and reliability. Currently, energy storage power stations generally employ a non-uniform sampling strategy to accommodate limited data transmission bandwidth. However, this strategy results in inconsistent sampling intervals and timestamp offsets for the voltage and current data of each cell within a battery cluster. Directly extracting features from raw data can easily lead to misjudgments or missed detections.

[0003] In large-scale energy storage scenarios, quickly locating abnormal cells is of great significance for preventing systemic risks. A single abnormal cell may cause local heat accumulation and consistency degradation due to degradation characteristics such as continuous growth of internal resistance or capacity drop during long-term operation. If it is not identified in time, it will lead to a collapse of battery cluster capacity consistency, a significant decrease in operating efficiency, and induce local thermal runaway risks under extreme operating conditions. Existing large-scale energy storage systems have relatively simple detection methods for battery cells, mainly detecting the voltage of each cell and the current of the battery cluster. The anomaly detection method also relies on a single threshold or fixed model, which makes it difficult to cope with data fluctuations and long-term degradation characteristics under dynamic conditions, resulting in unreliable anomaly detection results and greatly limited application scenarios. Summary of the Invention

[0004] In response to the defects of the existing technology, the purpose of this application is to provide a battery multi-parameter time alignment, battery feature extraction and battery anomaly detection method, aiming to solve the problem that the existing battery data collection has timestamp offset, resulting in unreliable battery detection results.

[0005] To achieve the above objectives, in a first aspect, the present application provides a battery multi-parameter time alignment method, comprising: Obtain the current time series data of the battery cluster and the voltage time series data of each battery cell belonging to the battery cluster; An improved dynamic time warping (DTW) algorithm is used to reconstruct the voltage time series data of each battery cell based on the current time series data of the battery cluster, so that the reconstructed voltage time series data is aligned with the characteristic points of the current time series data. The improved DTW algorithm is improved on the basis of the DTW algorithm, and its distance measurement matrix is ​​determined by a physical distance factor and a temporal distance factor. The physical distance factor is determined based on the current numerical information of the current time series data, the voltage numerical information of the voltage time series data, the open circuit voltage of the battery cell, and the equivalent internal resistance of the battery cell, and the temporal distance factor is determined based on the time information of the current time series data and the voltage time series data.

[0006] In one possible implementation, when the absolute value of the time difference between the sampling time of the first feature point in the current time series data and the sampling time of the second feature point in the cell voltage time series data is less than a preset time threshold, the distance measurement matrix between the first feature point and the second feature point is determined by the physical distance factor and the time distance factor; otherwise, the distance measurement matrix between the first feature point and the second feature point is infinite.

[0007] In a possible implementation, the battery cells are connected in series; The physical distance factor is:

[0008] in, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The physical distance factor, I i Represents the characteristic points in the current time series data i Current data, U j Represents the characteristic points in voltage time series data j Voltage data, OCV and R 0 represents the open circuit voltage and equivalent internal resistance of the battery cluster respectively.

[0009] In a possible implementation, the time distance factor is:

[0010] in, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The time distance factor, T Ii Represents the characteristic points in the current time series data i The sampling time, T Uj Represents the characteristic points in voltage time series data j The sampling time.

[0011] In a possible implementation, the distance metric matrix is:

[0012] in, Represents the characteristic points in the current time series data i and the characteristic points in the cell voltage time series data j The distance metric matrix elements of Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The physical distance factor, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The time distance factor, is the weight coefficient of the time distance factor.

[0013] In one possible implementation, the distance metric matrix is ​​defined as:

[0014] in, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The distance metric matrix elements of Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The physical distance factor, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The time distance factor, is the weight coefficient of the time distance factor, M represents the preset time threshold, and inf represents infinity.

[0015] In a second aspect, the present application provides a battery feature extraction method, comprising: Reconstructing the voltage time series data of each battery cell using the method described in the first aspect or any possible implementation of the first aspect; The characteristics of each battery cell are extracted based on the current time series data of the battery cluster and the reconstructed voltage time series data of each battery cell; the characteristics include the charging resistance and the discharging resistance of the battery cell.

[0016] In one possible implementation, extracting characteristics of each cell includes: Refer to the battery cluster current time series data and the reconstructed cell voltage time series data to obtain the battery cluster current at the start of charging, the current at the start of discharging, the instantaneous voltage before and after charging of each cell, and the instantaneous voltage before and after discharging of each cell; Determine the charging resistance of each cell by combining the instantaneous voltage before and after charging of each cell and the current at the start of charging of the battery cluster; The discharge resistance of each cell is determined by combining the instantaneous voltage before and after discharge of each cell and the current at the moment when the battery cluster starts to discharge.

[0017] In a third aspect, the present application provides a battery abnormality detection method, comprising: Extracting characteristics of each cell in the battery cluster using the method described in the second aspect or any possible implementation of the second aspect; Combining the characteristics of each battery cell, the isolation forest algorithm is used to identify abnormal battery cells.

[0018] In a fourth aspect, the present application provides an electronic device comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the programs stored in the memory are executed, the processor is used to execute the first aspect, any possible implementation of the first aspect, the second aspect, any possible implementation of the second aspect, and the method described in the third aspect.

[0019] In a fifth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the first aspect, any possible implementation of the first aspect, the second aspect, any possible implementation of the second aspect, and the method described in the third aspect.

[0020] In a sixth aspect, the present application provides a computer program product, which, when running on a processor, enables the processor to execute the first aspect, any possible implementation of the first aspect, the second aspect, any possible implementation of the second aspect, and the method described in the third aspect.

[0021] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the existing technologies: The present application provides a battery multi-parameter time alignment, battery feature extraction and battery anomaly detection method. Through the improved DTW algorithm, the current sequence of a cluster of batteries is used as a benchmark, and the mutation time points of the current and voltage sequences are captured through the improved physical distance factor, effectively aligning all voltage sequences; the optimal path search range of the algorithm is limited by the time distance factor, thereby accelerating the execution speed of the algorithm. Ultimately, the voltage sequence originally time-shifted can be uniformly aligned to the current sequence, while improving the execution efficiency of the algorithm. The final reconstructed sequence can be used for feature extraction and fault diagnosis, which will improve the fault diagnosis capability of the system and broaden the application scenarios. Therefore, the present application reconstructs the single-core voltage sequence based on the cluster current sequence through the battery multi-parameter time series alignment method, and then uses the reconstructed data to extract the multi-dimensional features of the battery. Finally, the extracted features are used to carry out consistency analysis, which can effectively identify abnormal cells in the energy storage system, greatly improving the operation and maintenance efficiency of the energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1This is a schematic diagram of a cell-level status assessment method using a time alignment algorithm for monitoring data of an energy storage battery cluster, proposed in an embodiment of the present application.

[0023] Figure 2 This is a graph of the original voltage and current data of five cells in a battery cluster of a domestic 200MWh energy storage power station provided in an embodiment of the present application.

[0024] Figure 3 A flowchart of a multi-parameter time alignment method for batteries provided in an embodiment of the present application.

[0025] Figure 4 This is a graph of voltage and current data from a classic DTW reconstruction of five cells in a battery cluster of a domestic 200MWh energy storage power station provided in an embodiment of the present application.

[0026] Figure 5 This is a graph of voltage and current data reconstructed using FastWDTW for five cells in a battery cluster of a domestic 200MWh energy storage power station provided in an embodiment of the present application.

[0027] Figure 6 A first-order Thevenin equivalent circuit model of a lithium battery provided in an embodiment of the present application.

[0028] Figure 7 A schematic diagram of a hybrid pulse power characterization (HPPC) internal resistance test method provided in an embodiment of the present application.

[0029] Figure 8 This is a diagram of the voltage and current waveforms of a battery cell under actual operating conditions of a domestic 200MWh energy storage power station on a certain day provided in an embodiment of the present application.

[0030] Figure 9 This is a flow chart for calculating the internal resistance characteristics of lithium battery charging and discharging based on reconstructing the current and voltage sequence proposed in this application.

[0031] Figure 10 This is a flowchart for calculating the equivalent capacity characteristics of a lithium battery based on reconstructing the current-voltage sequence proposed in this application.

[0032] Figure 11 This embodiment of the present application provides a line graph showing the abnormal scores of each cell in abnormal battery cluster A in a domestic 200MWh energy storage power station in June of a certain year.

[0033] Figure 12 This is a line graph showing the abnormality scores of each cell in a normal battery cluster B in a domestic 200MWh energy storage power station in June of a certain year, provided in an embodiment of the present application.

[0034] Figure 13The embodiment of the present application provides a feature space scatter plot of abnormal battery cluster A on June 15th of a certain year in June of a domestic 200MWh energy storage power station.

[0035] Figure 14 The embodiment of the present application provides an operating voltage curve of abnormal battery cluster A on July 4th in June of a certain year in a domestic 200MWh energy storage power station.

[0036] Figure 15 This is an architectural diagram of the battery multi-parameter time alignment system provided in an embodiment of the present application.

[0037] Figure 16 This is an architectural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0039] The term "and / or" as used herein describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. The symbol " / " as used herein indicates that the related objects are in an "or" relationship, for example, A / B means either A or B.

[0040] The terms "first" and "second" in this specification and claims are used to distinguish between different objects, rather than to describe a specific order of objects. For example, the terms "first feature point" and "second feature point" are used to distinguish between different feature points, rather than to describe a specific order of feature points.

[0041] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0042] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0043] This application proposes a cell-level status assessment method that uses a time alignment algorithm for energy storage battery cluster monitoring data. The single-cell voltage sequence is reconstructed based on the cluster current sequence through a battery multi-parameter time series alignment method. The reconstructed data is then used to extract multi-dimensional battery features. Finally, the extracted features are used to perform consistency analysis to proactively identify abnormal cells in the energy storage system, greatly improving the operation and maintenance efficiency of the energy storage system.

[0044] With the popularization of large-scale energy storage systems, the safety and reliability of lithium batteries have become core challenges. By carrying out cell-level state estimation and fault diagnosis for energy storage power stations and discovering abnormal cells in advance, the operation and maintenance efficiency of the energy storage system can be greatly improved. This application systematically constructs a complete technical chain from data preprocessing, feature mining, anomaly identification to data offline verification. First, to address the problem of non-stationary time series data alignment, this application introduces the FastWDTW algorithm, which achieves efficient and accurate time series matching while fully preserving the dynamic characteristics of the original signal, thereby significantly improving the accuracy and robustness of the data preprocessing stage. In the feature mining stage, by constructing a multi-dimensional feature system, the key charge and discharge internal resistance and equivalent capacity indicators in the electrochemical behavior are fully explored, the essential characteristics of the signal are effectively captured, and comprehensive and in-depth data support is provided for anomaly diagnosis. Then, to address the accuracy and computational complexity issues in anomaly detection, a feature space density analysis based on isolation forest is proposed, which can effectively reduce the computational complexity and improve the overall stability and reliability of anomaly recognition. Finally, offline verification with actual data shows that the proposed method can accurately identify abnormal phenomena in actual energy storage systems, fully verifying the engineering value of the method.

[0045] Figure 1 This application provides a flow chart of a cell-level state assessment method using a time alignment algorithm for monitoring data of an energy storage battery cluster. The method specifically includes: Step S1. Battery multi-parameter time series alignment method based on the improved FastWDTW algorithm.

[0046] With the rapid development of large-scale energy storage systems, the safety monitoring of lithium battery clusters has become a core challenge for the industry. Currently, the number of battery cells in large-scale energy storage power stations is typically over hundreds of thousands, while the cloud data transmission bandwidth only supports tens of megabytes, which severely restricts the granularity of data collection and transmission quality. To effectively monitor the status of battery cells under limited bandwidth conditions, energy storage power stations currently typically adopt a non-uniform sampling strategy, that is, low-frequency sampling under steady-state conditions and event-triggered supplementary sampling under dynamic conditions. Taking a 200MWh energy storage power station in China as an example, its battery cluster consists of 400 battery cells connected in series. The battery cluster management system performs data collection and transmission according to the following rules: If the data remains steady during the inspection process, the voltage and current data of the entire cluster of cells will be uploaded every 10 minutes; if data changes are detected during the inspection process, additional sampling will be performed and the data will be uploaded.

[0047] Under this collection rule, the sampling intervals for cluster current and cell voltage data are not fixed or uniform. At the same time, since each battery cluster contains a large number of cells, the timestamps of the collected data will also be offset during the inspection process. Extract 5 cells from a battery cluster and plot their original voltage and the current data of the battery cluster. The results are as follows: Figure 2 As shown. The charge and discharge end / start feature points of the current data in the figure do not completely match the voltage data, and the charge and discharge end / start feature points of the voltage data of each battery cell do not completely match. It should be noted that the battery cells in the same cluster are in series in the circuit. During the charging or discharging process, the current passing through each battery cell is exactly the same. When the current changes, the voltage of each battery cell should also change at the same time. Figure 2 The graph shows a double vertical axis. The left axis is the current of the battery cluster, and the right axis is the voltage of each battery cell. It can be seen that the voltage of each battery cell changes at different times. This difference in time may be caused by the existing mode of transmitting the voltage data of each battery cell by the energy storage system.

[0048] Specifically, when the battery management system (BMS) in an energy storage system sends cell voltage information to the upper-level monitoring system, it takes some time due to communication rate limitations. For example, if each battery cluster has 400 cells, the BMS collects the voltages of all 400 cells approximately synchronously. However, the BMS does not carry timestamp information when sending voltage information to the upper-level monitoring system. The upper-level monitoring system timestamps the received battery voltage data based on its own clock. As a result, the difference between the timestamp of the voltage data of the first cell in the cluster and the timestamp of the voltage data of the last 400 cells may be the transmission time of the voltage of all 400 cells.

[0049] Therefore, directly using raw voltage and current signals for feature extraction and consistency analysis will inevitably lead to inaccurate results and missed or misjudged abnormalities. Therefore, reconstructing the voltage and current data to align their feature points is very important for cell feature extraction.

[0050] To address the above issues, this application improves the classic DTW algorithm based on the Weighted Dynamic Time Warping (WDTW) algorithm and the Fast Dynamic Time Warping (FastDTW) algorithm, ultimately forming a FastWDTW algorithm that has physical significance and low computational complexity. It reconstructs the voltage data of all cells in the battery cluster based on the battery cluster current data.

[0051] Figure 3 A flow chart of a battery multi-parameter time alignment method provided in an embodiment of the present application; Figure 3 As shown, the following steps are included: Step S101, obtaining the current time series data of the battery cluster and the voltage time series data of each cell belonging to the battery cluster; Step S102: Using an improved dynamic time warping (DTW) algorithm, reconstruct the voltage time series data of each battery cell based on the current time series data of the battery cluster, so that the reconstructed voltage time series data is aligned with the feature points of the current time series data; the improved DTW algorithm is improved on the basis of the DTW algorithm, and its distance measurement matrix is ​​determined by a physical distance factor and a time distance factor; the physical distance factor is determined based on the current numerical information of the current time series data, the voltage numerical information of the voltage time series data, the open circuit voltage of the battery cell, and the equivalent internal resistance of the battery cell, and the time distance factor is determined based on the time information of the current time series data and the voltage time series data.

[0052] Optionally, when the absolute value of the time difference between the sampling time of the first feature point in the current time series data and the sampling time of the second feature point in the cell voltage time series data is less than a preset time threshold, the distance measurement matrix between the first feature point and the second feature point is determined by the physical distance factor and the time distance factor; otherwise, the distance measurement matrix between the first feature point and the second feature point is infinite.

[0053] Specifically, the following dynamic regularized distance metric function that integrates the inherent characteristics of the battery is used to calculate the equivalent distance between the current series and the voltage series, and construct the physical distance factor dist phys ( I i , U j), whose mathematical expression is as follows:

[0054] Additional time label data is introduced, and the time distance factor dist is added to each set of distances t ( T Ii , T Uj ), whose mathematical expression is as follows:

[0055] Although the time axis assigned to data uploaded to the cloud during the signal acquisition phase of a large-scale energy storage system may be offset due to the large number of cells in the system, the offset is limited. For the energy storage case in this embodiment, the baseline sampling period is 600s. Using the time axis data as a constraint, only the distance matrix that satisfies | T Ii - T Uj |<600s distance. The final distance matrix D w The calculation formula is as follows:

[0056] in, I i 、 U j and T Ii 、 T Uj are the numerical information and time information of the current and voltage time series to be reconstructed; OCV and R 0 are the open circuit voltage and equivalent internal resistance of the lithium battery, respectively. To simplify the calculation during data reconstruction, both are approximately considered constants, and the design is performed with reference to the battery manufacturer's data sheet. A k is the weight coefficient of the time distance factor, which can be used to adjust the proportion of the time distance factor and the physical distance factor in the multidimensional distance index, that is, to adjust the emphasis on the time axis information and the physical value axis information in the sequence reconstruction process.

[0057] And due to the introduction of constraints based on time information, D w The computational complexity of O ( n 2 ) is reduced to O ( n), which has obvious speed advantages when processing large-scale battery cell data; at the same time, a time distance factor is introduced during sequence reconstruction, thereby avoiding obvious warping of the reconstructed data on the time axis.

[0058] Furthermore, after determining the introduction of a new distance metric function that integrates physical and temporal constraints related to the inherent characteristics of the battery, the time series is reconstructed in the same way as the classic DTW algorithm. The basic steps are as follows: First create a distance matrix D ( i , j ), for the sequence A =[ a 1, a 2, … , a n ]and B =[ b 1, b 2, … , b m ], you need to build a n × m The elements of the matrix represent the distances between corresponding points calculated by the new distance metric function.

[0059] According to the above distance matrix D ( i , j ) Further calculate the minimum cumulative distance for each location C ( i , j ), forming the minimum cumulative distance matrix. Among them, C ( i , j ) indicates starting from the starting point i To current location j The shortest path of , its recursive relationship is:

[0060] According to the calculated minimum cumulative distance matrix from C ( n , m ) Backtrack to the minimum cumulative distance matrix C (1, 1), the optimal path can be obtained; according to the optimal path, the alignment and reconstruction of the time series can be completed; express A 、 B The distance between corresponding elements in the sequence is usually the Euclidean distance in the classical algorithm, but in the improved algorithm provided by this application, the distance is a combination of the physical distance factor and the time distance factor, which is the distance mentioned above. Dw .

[0061] The classic DTW algorithm and the improved FastWDTW algorithm are used to Figure 2 The extracted 5 battery cell data are reconstructed, and the results are as follows Figure 4 and Figure 5 As shown in the figure. A comparison shows that the voltage sequence reconstructed using FastWDTW has less warping (i.e., the degree of divergence between multiple voltage sequence curves) than the original sequence, and can accurately align the current mutations at the start and end of charging and discharging, eliminating the impact of sampling period differences. In terms of computational speed, given the same computing resources, reconstructing the voltage sequence of 400 cells using the classic DTW algorithm takes approximately 506 seconds, while using the improved FastWDTW takes only approximately 178 seconds. This fully demonstrates the computational speed advantage of the improved FastWDTW algorithm over the classic DTW algorithm.

[0062] Step S2. A battery feature extraction method based on aligned current and voltage sequences.

[0063] The single-cell voltage sequence is reconstructed based on the cluster current sequence by the battery multi-parameter time series alignment method in step S1. In order to fully evaluate the consistency of the battery, the sequence needs to be further feature extracted for the reconstructed data, including the charging resistance. R c0 , discharge resistor R d0 and equivalent capacity Q x There are three groups of features.

[0064] According to the first-order Thevenin equivalent circuit model of lithium battery, Figure 6 As shown, R 0 is the internal resistance characteristic that needs to be reflected by external voltage. The hybrid pulse power characterization (HPPC) method is usually used to test the internal resistance. The test process is as follows: Figure 7 As shown; Figure 6 middle R 1 is the polarization resistance in the battery equivalent circuit, C 1 is the polarization capacitance in the battery equivalent circuit, U 1 is the polarization voltage in the battery equivalent circuit. The standard test process consists of a 10s pulse current discharge, a 40s rest, and a 10s pulse current charge. The charge and discharge current rates tested can be 0.5C and 1C, respectively. Based on the two current pulses, the charging internal resistance is calculated using Ohm's law. R c0 and discharge internal resistance R d0 .

[0065]

[0066] in, I c and I d are the current magnitudes at the start of charging and discharging respectively; V c0 、 V c1 are the extracted characteristic voltages before and after charging, respectively; V d0 、 V d1 are the extracted characteristic voltages before and after discharge, respectively.

[0067] For energy storage battery systems, their operating conditions have typical cyclical characteristics, which are mainly characterized by constant current charging during the off-peak electricity price period of the grid and constant power discharge during the peak electricity price period. The actual operating current and cell voltage waveforms of the energy storage site (such as Figure 8 ) and the Hybrid Pulse Power Characterization (HPPC) test method (as shown in Figure 7 ) have similar step change characteristics.

[0068] Based on the reconstructed current and voltage time series data, the process of calculating the charge and discharge internal resistance of the battery cell by dynamic feature extraction method is as follows: Figure 9 As shown in the figure, the specific process is as follows: First, the current mutation detection algorithm is used to obtain the characteristic voltage and characteristic current. Discrete difference operation is performed on the current sequence to calculate the current change rate, and the current index that meets the requirements is screened by the set threshold. Based on the reconstructed voltage sequence aligned with the current sequence, the characteristic voltage is obtained according to the current label. Then, the charging internal resistance is calculated based on the time series data before and after the characteristic point. R c0 and discharge internal resistance R d0 Finally, a three-dimensional data storage matrix is ​​constructed, which traverses all battery cells daily and stores the charging and discharging internal resistances. If the energy storage site does not have suitable calculation conditions on a certain day due to maintenance or other reasons, the data for that day will be recorded as empty.

[0069] This method combines operating condition analysis with HPPC test principles. Starting from the operating characteristics of the energy storage system, it calculates the internal resistance by screening suitable operating conditions without introducing additional test conditions. It has high practical value. Among them, only when the current change rate reaches the threshold does it indicate that the appropriate operating conditions exist on that day.

[0070] Energy storage batteries typically consist of multiple cells connected in series to form a battery cluster to achieve higher voltage levels. For a series-connected battery cluster, the current flowing through each cell is always the same, so the capacity changes consistently during charge and discharge. However, due to variations among the cells in a battery cluster, the capacity of some cells will inevitably decay more rapidly during use, affecting the capacity of the entire cluster. Therefore, it is necessary to estimate the capacity of each battery cell to identify abnormal cells whose capacity significantly deviates from the majority.

[0071] According to the operating characteristics of energy storage batteries, this application proposes a battery equivalent capacity estimation algorithm based on specific operating conditions and SOC-OCV curves to measure the capacity level of each battery cell in the same cluster; wherein the above-mentioned specific operating conditions refer to the operating conditions where the current is 0 and SOC>80% or SOC<30%.

[0072] To calculate the number of cells n Equivalent capacity characteristics of the day Q x, n , then select the characteristic points and calculate the estimated value of the equivalent capacity of the battery cell Q x, n The process is as follows Figure 10 As shown. First, extract the characteristic voltage based on the state of charge of the battery cluster. The battery management system collects voltage, current and SOC time series data, filters the static operating condition segment with zero current, and extracts the high charge interval data of SOC>80% and the low charge interval data of SOC<30% in the static segment. This interval is the non-plateau curve of the SOC-OCV curve. Select the voltage stable segment at the end of the high / low SOC interval respectively, and then calculate the characteristic voltage corresponding to the high SOC U h and low SOC corresponding characteristic voltage U l .

[0073] Then correct the equivalent capacity based on the SOC-OCV relationship curve Q x, n . Using the Gaussian SOC-OCV relationship model:

[0074] in c 1. c 2. c 3 are constant coefficients, U It represents the battery terminal voltage collected during the battery's rest phase. At this time, the battery terminal voltage can be approximately considered equal to the open-circuit voltage.

[0075] The characteristic voltage U h 、 U lSubstitute into the formula to obtain the corrected high SOC value SOC h and low SOC value SOC l The daily equivalent capacity is then calculated based on the principle of charge conservation:

[0076] Among them, SOC meas-h , SOC meas-l is the measured high / low SOC value, Q x, n-1 is the estimated equivalent capacity of the previous day, Q new The equivalent capacity calculation value for the day. The capacity estimate is updated using the exponential smoothing method:

[0077] in, a 1 is the smoothing coefficient, ranging from [0, 1]. a 1 will make the latest data have a greater impact on the prediction results, while smaller a 1 will make the impact of historical data greater; Q x, n It is the estimated value of equivalent capacity for the day. At the same time, a data validity verification mechanism is established, that is, when the data index of the high / low SOC interval is empty, the Q x, n = Q x, n-1 .

[0078] The algorithm combines dynamic operating condition feature extraction with static SOC-OCV model correction, realizes data segmentation processing through a sliding time window, and combines a recursive update strategy to ensure the stability of capacity estimation.

[0079] Step S3. A multi-feature consistency analysis method based on the extracted multi-dimensional features.

[0080] The charging resistance is extracted based on the aligned voltage and current sequences by the battery feature extraction method in step S2. R c0 , discharge resistor R d0 and equivalent capacity Q x There are three groups of characteristics. R c0 and discharge resistor R d0 The correlation is very strong, and the discharge resistance R d0 It has better robustness, so the discharge resistor is selected R d0 and equivalent capacityQ x Analyze the consistency of battery cells together.

[0081] Compared to outlier detection algorithms based on feature point distance, the isolation forest algorithm does not use distance or density metrics to detect anomalies, which greatly reduces computational costs and only requires a small sub-sampling size to achieve efficient detection performance. The isolation forest algorithm abandons the idea of ​​training a high-precision anomaly detection network, constructs multiple simple sub-detectors, namely isolated trees, and fuses them to achieve better detection performance with lower complexity. The low complexity and distributed deployment of the isolation forest enable it to perform excellent performance when performing massive data calculations, and is particularly suitable for processing problems such as consistency of battery cells at energy storage sites. Combined with the low computational cost of the isolation forest algorithm under large amounts of data, this application selects this algorithm to perform multi-feature consistency analysis on batteries.

[0082] The specific detection process of the isolation forest algorithm is as follows. First, create an isolated tree (iTree), given the data sample space X , randomly select some samples to form a training space Y , arbitrarily select features q and its partition value p , recursively divide the training space so that the point on the left branch of the current node is less than p , and will be greater than or equal to p The points are placed on the right branch of the current node until any of the following conditions are met: (1) the subsample space is indivisible; (2) the specified cutting depth is reached. The above process is repeated to construct t An isolated tree forms an isolated forest. X Test space in Z They are input into the trained isolation tree for data evaluation, and outliers among them are identified based on the anomaly scores of the sample points.

[0083] Sample points x Anomaly score The definition is as follows:

[0084] in, h ( x ) is the sample point in iForest x The average depth of c ( n ) is from n The average path length of the isolation tree constructed by points is calculated as follows:

[0085] in, H ( k) = ln( k ) + 0.577215665.

[0086] In order to verify the method's ability to predict the alarm risk of each battery cell in a large-scale energy storage site, the alarm data for July and the operating data for June of a domestic 200MWh energy storage power station were used as an example. The method was used to identify the battery cells in the battery cluster with significant internal resistance or capacity deviations, and then verified in combination with the site's alarm data.

[0087] Based on the July alarm data, we identified a group of clearly abnormal cells and their respective battery clusters, A, and a group of normal cells and their respective battery clusters, B. We then used the June operating data of each to predict the July alarm behavior, verifying the accuracy of our proposed method. The hyperparameters for the isolation forest algorithm were set as follows: number of random trees, 100, number of subtree samples, 256, maximum tree depth, 7, and anomaly prediction threshold, 0.65.

[0088] Calculate the isolation forest anomaly scores of abnormal battery cluster A and battery cluster B in June, and draw the anomaly score change curve of each battery cell by date. The anomaly score edge curves of battery cluster A and battery cluster B are as follows: Figure 11 and Figure 12 shown.

[0089] As can be seen from the figure, the abnormal scores of two cells in battery cluster A gradually increase, while there is no cell in battery cluster B whose abnormal score increases significantly over time, so it is believed that there are abnormal cells in battery cluster A. Then the discharge resistance of battery cluster A on June 15th is plotted. R d0 and equivalent capacity Q x The feature space scatter plot of the feature is as follows Figure 13 As shown, it is found that the equivalent capacity of abnormal cells Cell2 and Cell3 is significantly lower, while the discharge internal resistance of the cells does not deviate significantly from the abnormality. Therefore, it is determined that the capacity and initial SOC of the cells are abnormal, and the low voltage alarm may be caused during the static stage of the subsequent charge and discharge cycle.

[0090] Combining the operation and alarm data in July, it was found that battery cluster A had no charging or discharging conditions on July 4th, but several battery cell low voltage alarms appeared on that day. The voltage curve of battery cluster A on July 4th was drawn as follows: Figure 14As shown in the figure. Since no low cell voltage alarms occurred during the operating time from June 1st to June 15th, while the voltages of the abnormal cells Cell 2 and Cell 3 during the static phase on July 4th were lower than the set alarm threshold of 2.80V, continuously triggering low voltage alarms, this demonstrates that the method of this application can identify abnormal cells in the energy storage system before alarm signals are generated, and can locate the type of anomaly based on further analysis of the feature space scatter plot after the anomaly is identified, greatly improving the operation and maintenance efficiency of the energy storage system.

[0091] Figure 15 This is a diagram of the battery multi-parameter time alignment system architecture provided by an embodiment of the present application, such as Figure 15 Shown, including: A sequence data acquisition module 1510 is used to acquire the current time series data of the battery cluster and the voltage time series data of each battery cell included in the battery cluster; The time alignment module 1520 is configured to reconstruct the voltage time series data of each battery cell based on the current time series data of the battery cluster using an improved DTW algorithm, so that the reconstructed voltage time series data is aligned with the feature points of the current time series data. The improved DTW algorithm is an improvement on the DTW algorithm, and its distance metric matrix is ​​determined by a physical distance factor and a temporal distance factor. The physical distance factor is determined based on the current numerical information of the current time series data, the voltage numerical information of the voltage time series data, the open circuit voltage of the battery cell, and the equivalent internal resistance of the battery cell. The temporal distance factor is determined based on the temporal information of the current time series data and the voltage time series data.

[0092] In the improved DTW algorithm adopted by the above-mentioned time alignment module 1520: when the absolute value of the time difference between the sampling time of the first feature point in the current time series data and the sampling time of the second feature point in the battery cell voltage time series data is less than the preset time threshold, the distance measurement matrix between the first feature point and the second feature point is determined by the physical distance factor and the time distance factor; otherwise, the distance measurement matrix between the first feature point and the second feature point is infinite.

[0093] It should be understood that the above-mentioned system is used to execute the method in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the system are similar to those described in the above-mentioned method. The working process of the system can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0094] Based on the method in the above embodiment, the embodiment of the present application provides an electronic device, such as Figure 16As shown, the electronic device may include: a processor 1610, a communication interface 1620, a memory 1630, and a communication bus 1640, wherein the processor 1610, the communication interface 1620, and the memory 1630 communicate with each other via the communication bus 1640. The processor 1610 may call the logic instructions in the memory 1630 to execute the method in the above embodiment.

[0095] In addition, the logic instructions in the aforementioned memory 1630 can be implemented in the form of a software functional unit and, when sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0096] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0097] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0098] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0099] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and storage medium can be located in an ASIC.

[0100] The above embodiments can be implemented in whole or in part using software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions. When loaded and executed on a computer, the computer program instructions fully or partially produce the processes or functions described in the embodiments of this application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted via the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state drive (SSD)).

[0101] It will be understood that the various numerical numbers involved in the embodiments of the present application are merely distinctions for the convenience of description and are not intended to limit the scope of the embodiments of the present application.

[0102] It is easy for those skilled in the art to understand that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A battery multi-parameter time alignment method, characterized in that: include: Acquiring current time series data of a battery cluster and voltage time series data of each battery cell belonging to the battery cluster; Using an improved dynamic time warping (DTW) algorithm, reconstructing the voltage time series data of each battery cell based on the current time series data of the battery cluster, so that the reconstructed voltage time series data is aligned with feature points of the current time series data; The improved DTW algorithm is improved on the basis of the DTW algorithm, and its distance measurement matrix is ​​determined by a physical distance factor and a time distance factor; the physical distance factor is determined according to the current numerical information of the current time series data, the voltage numerical information of the voltage time series data, the open circuit voltage of the battery cell, and the equivalent internal resistance of the battery cell, and the time distance factor is determined according to the time information of the current time series data and the voltage time series data.

2. The method according to claim 1, characterized in that When the absolute value of the time difference between the sampling time of the first feature point in the current time series data and the sampling time of the second feature point in the cell voltage time series data is less than a preset time threshold, the distance measurement matrix between the first feature point and the second feature point is determined by the physical distance factor and the time distance factor; otherwise, the distance measurement matrix between the first feature point and the second feature point is infinite.

3. The method according to claim 1, characterized in that The battery cells are connected in series; The physical distance factor is: in, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The physical distance factor, I i Represents the characteristic points in the current time series data i Current data, U j Represents the characteristic points in voltage time series data j Voltage data, OCV and R 0 represents the open circuit voltage and equivalent internal resistance of the battery cluster respectively.

4. The method according to claim 1, wherein The time distance factor is: in, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The time distance factor, T Ii Represents the characteristic points in the current time series data i The sampling time, T Uj Represents the characteristic points in voltage time series data j The sampling time.

5. The method according to claim 1, characterized in that The distance metric matrix is: in, Represents the characteristic points in the current time series data i and the characteristic points in the cell voltage time series data j The distance metric matrix elements of Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The physical distance factor, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The time distance factor, is the weight coefficient of the time distance factor.

6. The method according to claim 2, characterized in that The distance metric matrix is ​​defined as: in, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The distance metric matrix elements of Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The physical distance factor, Represents the characteristic points in the current time series data i and characteristic points in voltage time series data j The time distance factor, is the weight coefficient of the time distance factor, M represents the preset time threshold, and inf represents infinity.

7. A battery feature extraction method, characterized in that: include: Reconstructing the voltage time series data of each battery cell using the method described in any one of claims 1 to 6; The characteristics of each cell are extracted based on the current time series data of the battery cluster and the reconstructed voltage time series data of each cell.

8. The method according to claim 7, characterized in that Cell characteristics include: charging resistance and discharging resistance of the cell; extracting characteristics of each cell, including: Refer to the battery cluster current time series data and the reconstructed cell voltage time series data to obtain the battery cluster current at the start of charging, the current at the start of discharging, the instantaneous voltage before and after charging of each cell, and the instantaneous voltage before and after discharging of each cell; Determine the charging resistance of each cell by combining the instantaneous voltage of each cell before and after charging and the current at the start of battery cluster charging; The discharge resistance of each cell is determined by combining the instantaneous voltage before and after discharge of each cell and the current at the moment when the battery cluster starts to discharge.

9. A battery abnormality detection method, characterized in that: include: Extracting the characteristics of each cell in the battery cluster using the method described in claim 7 or 8; Combining the characteristics of each battery cell, the isolation forest algorithm is used to identify abnormal battery cells.

10. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory, and when the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1-6, 7-8 or 9.

Images