A battery monomer micro-short circuit early detection method for energy storage power station

Abstract

The application relates to a battery cell micro-short circuit early detection method for energy storage power stations, which can realize micro-short circuit detection of battery cells in large-scale energy storage power stations; a clustering algorithm can automatically determine a micro-short circuit detection threshold, which makes the detection algorithm have certain adaptability to errors of a standardized voltage curve change rate, and high-precision detection effects can also be realized in complex environments with small changes in power station working conditions and different battery degradation degrees. Through the breakthrough of three core technologies, namely, a dynamic threshold and trend verification cooperative mechanism, an enhanced HDBSCAN clustering algorithm and double-source data fusion, early, accurate and low-cost detection of battery cell micro-short circuits of energy storage power stations is realized, and a grading early warning strategy is matched to improve operation and maintenance efficiency. Experimental verification shows that the method is significantly superior to the prior art in terms of detection accuracy, real-time performance, anti-interference capability and the like, and has outstanding innovation and engineering application value.

Description

Technical Field

[0001] This invention relates to an early detection method for micro short circuits in individual battery cells for energy storage power stations. Background Technology

[0002] With the transformation of the global energy structure and the continuous growth in demand for clean energy, energy storage technology is receiving increasing attention as a key support for solving the problems of intermittency and volatility of renewable energy and achieving efficient energy utilization and stable grid operation. Energy storage power stations, as an important application form of large-scale energy storage systems, play a vital role in areas such as power peak shaving, frequency regulation, backup power, and distributed energy integration.

[0003] In energy storage power stations, individual battery cells are the core components of the system, and their performance and safety directly affect the overall operational efficiency and reliability of the station. However, during production and use, individual battery cells may experience micro-short circuits due to various factors such as manufacturing defects, material aging, overcharging and over-discharging, and mechanical damage. A micro-short circuit refers to a tiny current leakage path within the battery, causing abnormal reactions of the internal chemical substances, which in turn leads to problems such as performance degradation, overheating, and capacity decay.

[0004] Micro-short circuits are often difficult to detect in their early stages because the current they generate is small, and their impact on battery voltage and capacity is relatively insignificant. However, as the micro-short circuit develops, the current gradually increases, leading to an increase in battery temperature and potentially causing thermal runaway, which in turn can result in serious safety accidents such as fires and explosions.

[0005] Currently, the main shortcomings of the detection methods for micro-short circuits in individual battery cells are as follows:

[0006] Limited detection accuracy: Traditional detection methods, such as monitoring based on battery voltage and current, can often only detect relatively obvious short circuit faults. For early micro short circuit phenomena, due to their weak signals, they are easily interfered with by noise and are difficult to identify accurately.

[0007] Poor real-time performance: Some detection methods require periodic offline testing of batteries, making it impossible to achieve real-time online monitoring of micro-short circuits in individual battery cells. However, energy storage power stations contain a large number of battery cells, making offline testing not only inefficient but also unable to detect early signs of micro-short circuits in a timely manner, thus hindering the implementation of effective preventative measures before a failure occurs.

[0008] High cost: Some advanced detection technologies, such as those using high-precision sensors and complex signal processing algorithms, can improve detection accuracy, but the equipment and maintenance costs are high, making it difficult to widely apply them in large-scale energy storage power stations.

[0009] Therefore, developing an early detection method for micro-short circuits in individual battery cells for energy storage power stations, which can detect micro-short circuit phenomena in individual battery cells in real time and accurately, is of great practical significance for improving the safety and reliability of energy storage power stations and ensuring the stable operation of energy storage systems. Summary of the Invention

[0010] In view of the above situation and to overcome the shortcomings of the existing technology, the purpose of this invention is to provide an early detection method for micro-short circuits in battery cells for energy storage power stations. This method can automatically determine the micro-short circuit detection threshold and detect the cells that have experienced micro-short circuits in real time, effectively overcoming the adverse effects of manually setting unreasonable detection thresholds on the accuracy of the algorithm.

[0011] The technical solution adopted in this invention is:

[0012] A method for early detection of micro-short circuits in individual battery cells for energy storage power stations, characterized by comprising the following steps:

[0013] Step (1): Collect the operating data of the energy storage power station for three consecutive days;

[0014] Step (2): Denoise reduction processing is performed on the collected raw data;

[0015] Step (3): Set a dynamic threshold and combine it with trend verification technology to extract the voltage data in step (2) and obtain the effective charging voltage curve;

[0016] Step (4): Calculate the rate of change of the charging voltage curve obtained in step (3) using a weighted average method, and then standardize it.

[0017] Step (5): Based on the HDBSCAN algorithm, the battery voltage change rate after standardization in step (4) is clustered by constructing a distance map and agglomeration tree. Combined with parameter tuning and semi-supervised learning to fuse labeled data, abnormal battery cells are accurately identified.

[0018] Step (6): If a battery cell is marked as abnormal twice in three consecutive days of data, it is identified as a micro-short-circuited cell.

[0019] Preferably, the running data in step (1) includes the voltage and current of a single battery cell.

[0020] Preferably, the noise reduction process in step (3) achieves noise filtering through the synergistic effect of dynamic threshold adjustment and multi-dimensional trend verification, including: screening effective charging data to construct a voltage sequence, setting a start threshold and a cutoff threshold; at the same time, ensuring the rationality of the trend through dynamic incremental constraints, establishing an abnormal interruption response mechanism to terminate abnormal segments in a timely manner and reset the monitoring status, thereby improving the accuracy of charging segment identification and anti-interference capability.

[0021] Preferably, in step (4), the difference in curve shape is quantified by calculating the overall rate of change of the battery charging voltage curve, and the calculated overall rate of change sequence is subjected to Z-Score standardization processing to improve the accuracy of the clustering detection algorithm. The specific steps are as follows:

[0022] Preferably, in step (5), the HDBSCAN algorithm is used to perform cluster analysis on the battery charging voltage change rate. Anomaly pattern recognition is achieved by constructing a distance map, agglomeration tree, and stability assessment: First, a clustering tree structure is constructed based on density stratification. Stable clusters are screened by cluster persistence measurement. The clustering granularity is controlled by adjustable parameters. Finally, the largest cluster is determined to be a normal battery group, and small clusters and noise points are marked as potential micro-short-circuit cells, i.e., abnormal battery cells. The semi-supervised training mechanism that integrates a small amount of labeled data can further optimize the clustering boundary and effectively improve the accuracy and noise resistance of micro-short-circuit detection. After clustering by the HDBSCAN algorithm, when the same cell triggers the abnormal label twice within three consecutive operating cycles, it is considered that the battery cell has triggered a micro-short-circuit warning. By integrating dual-source data from actual power plants and simulation systems, combined with dynamic threshold verification and enhanced HDBSCAN clustering algorithm, an intelligent detection system of "feature extraction-anomaly identification-dual verification-tiered warning" is constructed to achieve early and accurate identification and quantitative assessment of micro-short-circuit faults, which is conducive to improving the safety and economy of energy storage power plant operation and maintenance.

[0023] Compared with existing technologies, this invention enables micro-short circuit detection in individual battery cells of large-scale energy storage power stations. The clustering algorithm automatically determines the micro-short circuit detection threshold, making the detection algorithm adaptable to errors in the rate of change of the standardized voltage curve. It achieves high-precision detection even in complex environments with small variations in power station operating conditions and varying degrees of battery degradation. Through breakthroughs in three core technologies—a dynamic threshold and trend verification collaborative mechanism, an enhanced HDBSCAN clustering algorithm, and dual-source data fusion—this invention achieves early, accurate, and low-cost detection of micro-short circuits in individual battery cells of energy storage power stations, and is complemented by a tiered early warning strategy to improve operation and maintenance efficiency. Experimental verification shows that this method significantly outperforms existing technologies in terms of detection accuracy, real-time performance, and anti-interference capability, demonstrating outstanding innovation and engineering application value. Attached Figure Description

[0024] Figure 1 Flowchart of the method of this invention.

[0025] Figure 2 A flowchart illustrating the steps of the method for extracting the charging voltage curve in this invention.

[0026] Figure 3 This invention provides an application example of battery voltage and current variation data over three consecutive days.

[0027] Figure 4The present invention applies the three-charge voltage curves of the first eight cells.

[0028] Figure 5 The present invention applies the standardized rate of change of two sets of charging voltage curves.

[0029] Figure 6 This invention applies the algorithm to detect battery micro-short circuits. Detailed Implementation

[0030] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples.

[0031] like Figure 1 As shown, the present invention provides an early detection method for micro-short circuits in battery cells for energy storage power stations, comprising the following steps:

[0032] Step (1): Collect the operating data of the energy storage power station and its simulation system for three consecutive days; the data of the energy storage power station is mostly exported from its simulation system. Collecting data from the energy storage power station can be regarded as collecting data from the simulation system.

[0033] Step (2): Denoise reduction processing is performed on the collected raw data;

[0034] Step (3): Set a dynamic threshold and combine it with trend verification technology to extract the voltage data in step (2) and obtain the effective charging voltage curve;

[0035] Step (4): Calculate the rate of change of the charging voltage curve obtained in step (3) using a weighted average method, and then standardize it.

[0036] Step (5): Based on the HDBSCAN algorithm, the battery voltage change rate after standardization in step (4) is clustered by constructing a distance map and agglomeration tree. Combined with parameter tuning and semi-supervised learning to fuse labeled data, abnormal battery cells are accurately identified.

[0037] Step (6): If a battery cell is marked as abnormal twice in three consecutive days of data, it is identified as a micro-short-circuited cell.

[0038] The data for step (1) includes the voltage and current of a single battery cell.

[0039] The noise reduction process in step (3) achieves noise filtering through the synergistic effect of dynamic threshold adjustment and multi-dimensional trend verification, including: screening effective charging data to construct a voltage sequence, setting a start threshold and a cutoff threshold; at the same time, ensuring the rationality of the trend through dynamic incremental constraints, establishing an abnormal interruption response mechanism to terminate abnormal segments in a timely manner and reset the monitoring status, thereby improving the accuracy of charging segment identification and anti-interference capability.

[0040] The specific steps are as follows:

[0041] Step (3.1): Extract charging period data: Filter the voltage data points corresponding to the charging process where current I < 0 from the original data, and construct a charging voltage sequence set;

[0042] Step (3.2): Determine the dynamic threshold: Based on the distribution of historical charging data, calculate the mean μ and standard deviation σ of the historical voltage data, determine the charging start threshold start and the stop threshold stop, and form a dynamic monitoring window [start, stop]; the calculation formulas for start and stop are as follows:

[0043] start=μ-3σ

[0044] stop=μ+3σ

[0045] Step (3.3): Based on the monitoring window obtained in step (3.2), perform dual verification on the voltage sequence obtained in step (3.1), including dynamic increment constraint and abnormal interruption handling; the dynamic increment constraint requires the later value to be greater than the previous value, and a small tolerance δ is set to deal with measurement errors; when the sequence interruption is detected, the current sequence is immediately terminated and the monitoring status is reset.

[0046] Step (3.4): Repeat steps (3.2) and (3.3) to obtain multiple voltage sequences, and then integrate and splice them to obtain the noise-reduced battery charging voltage curve.

[0047] 4. The method for early detection of micro-short circuits in battery cells for energy storage power stations according to claim 1, characterized in that, in step (4), the curve shape difference is quantified by calculating the overall change rate of the battery charging voltage curve, and the calculated overall change rate sequence is subjected to Z-Score standardization processing to improve the accuracy of the clustering detection algorithm. The specific steps are as follows:

[0048] Step (4.1): Calculate the rate of change of the charging voltage curve for each battery cell between day 1 and day 2, and between day 2 and day 3; specifically, calculate the overall rate of change between the curves by averaging the rate of change at each point on the curve, using the following formula:

[0049]

[0050] Where: n is the total number of data points on the battery charging voltage curve, u 1ij and u 2ij This refers to the voltage value at a specific time point in the first and second charging voltage curves.

[0051] Step (4.2): The voltage curve change rate obtained in step (4.1) is standardized using Z-Score to further separate abnormal data from normal data within small intervals, which is beneficial for subsequent clustering detection algorithms. The formula is as follows:

[0052]

[0053] Where: r i Let μ be the rate of change of the voltage curve of the i-th cell during a single charge, and let μ and σ be r i The mean and standard deviation of the distribution.

[0054] In step (5), the HDBSCAN algorithm is used to perform cluster analysis on the battery charging voltage change rate. Abnormal pattern recognition is achieved by constructing a distance map, a clustering tree, and stability assessment: First, a clustering tree structure is constructed based on density stratification. Stable clusters are screened by cluster persistence measurement. The clustering granularity is controlled by adjustable parameters. Finally, the largest cluster is determined to be a normal battery group, and small clusters and noise points are marked as potential micro-short-circuit cells, i.e., abnormal battery cells. The semi-supervised training mechanism that integrates a small amount of labeled data can further optimize the clustering boundary and effectively improve the accuracy and noise resistance of micro-short-circuit detection. The specific steps are as follows:

[0055] Step (5.1): Based on the standardized rate of change obtained in step (4), construct the cross-distance graph and obtain its minimum spanning tree; the HDBSCAN algorithm estimates the density between points by calculating the cross-distance of each point, where the cross-distance is defined as:

[0056] reachability(p,q)=max(core_dist(p),core_dist(q),distance(p,q))

[0057] In the above formula, core_dist(p) and core_dist(q) are the min_samples nearest neighbor distances between points p and q, and distance(p,q) is the distance between points p and q. The min_samples nearest neighbor distance refers to the minimum radius that expands around the point from small to large until it contains no less than min_samples neighboring points.

[0058] The minimum spanning tree obtained by constructing a reachability graph is an acyclic connected graph that connects all points and has the minimum total edge weight.

[0059] Step (5.2): Construct a hierarchical agglomerative tree based on the minimum spanning tree in step (5.1); where each node of the agglomerative tree represents a cluster, the root node of the tree represents the entire dataset, and the leaf nodes represent a single data point; each node of the agglomerative tree contains density information of a cluster, including the size and density of the cluster.

[0060] Step (5.3): Evaluate the clustering tree obtained in step (5.2) and select the level with optimal stability. Its stability is measured by calculating the persistence λ(C) of each cluster, as shown in the following formula:

[0061]

[0062] in, The maximum density of the cluster, This represents the minimum density of the cluster.

[0063] Step (5.4): Prune the agglomeration tree obtained in step (5.2) to remove unstable clusters and extract the final clusters. During the pruning process, clusters with high persistence are retained, while small clusters with low persistence and instability are discarded. Extract the final clusters from the pruned agglomeration tree and mark noise points to achieve clustering of data samples.

[0064] In step (6), after clustering using the HDBSCAN algorithm, if the same cell triggers an anomaly marker twice within three consecutive operating cycles, it is considered that the battery cell has triggered a micro-short circuit warning. Based on this, a tiered warning mechanism is further introduced. As shown in Table 1, the fault level is determined by quantifying the deviation of the voltage change rate, thus achieving differentiated alarm triggering and providing maintenance personnel with precise guidance on fault handling priorities.

[0065] Table 1 Warning Levels and Judgment Logic

[0066]

[0067] In practical applications, this invention has achieved the same or similar technical effects. Examples of applications are as follows:

[0068] An experiment was conducted based on three consecutive days of operating data from a battery cabinet at an energy storage power station. The cabinet structure consists of 12 parallel battery clusters per cabinet, with each cluster containing 364 series-connected battery cells, resulting in 4368 battery cells per cabinet. The energy storage power station performs one constant-power charge-discharge cycle daily. The changes in some electrical parameters from the experimental data are shown below. Figure 3 As shown.

[0069] Combining the formula in step 3.2, the average voltage is calculated to be 2518.18V and the standard deviation is 623.74V. The dynamic monitoring window can be set to [646.96V, 4389.40V].

[0070] By artificially injecting faults into the experimental data, the data distribution of a real micro-short circuit in a battery cell is simulated, and the algorithm's detection performance of micro-short circuit phenomena is verified. The voltage curves of cells 1, 101, 201, and 301 in each battery cluster are artificially lowered to simulate a micro-short circuit. The following example only uses the first eight cells in the first cluster (rack 1) as an example, employing... Figure 2 The voltage extraction method shown extracts the three charging voltage curves of rack1, as follows: Figure 4 As shown.

[0071] Based on the formula in step (4.1), the data from step (3) are further calculated, and the two-day variation rate of normal monomers is 0.38%, while the two-day variation rates of experimental monomers are -0.42% and -0.41%, respectively. It can be seen that the standardized variation rate of abnormal monomers is significantly lower than that of normal monomers.

[0072] Based on the formula in step (4.2), further calculate the standardized rate of change of the two sets of charging voltage curves for each battery cell in rack1 during the three cycles, as shown in the figure. Figure 5 As shown, the standardized rate of change of the two sets of charging voltage curves shows that, due to fluctuations in battery charging conditions and errors in voltage curve extraction, the standardized rate of change of normal cells deviates from 0, while the standardized rate of change of abnormal cells is significantly lower than that of other normal cells.

[0073] The HDBSCAN algorithm was used to detect abnormal voltage curve changes, and the abnormal cell was output as a micro-short-circuit cell. The experimental output results are as follows: Figure 6 As shown, micro-short circuit anomalies were detected in cell numbers 1, 101, 201, and 301.

Claims

1. A method for early detection of micro-short circuits in individual battery cells for energy storage power stations, characterized in that, Includes the following steps: Step (1): Collect the operating data of the energy storage power station for three consecutive days; Step (2): Denoise reduction processing is performed on the collected raw data; Step (3): Set a dynamic threshold and combine it with trend verification technology to extract the voltage data in step (2) and obtain the effective charging voltage curve; Step (4): Calculate the rate of change of the charging voltage curve obtained in step (3) using a weighted average method, and then standardize it. Step (5): Based on the HDBSCAN algorithm, the battery voltage change rate after standardization in step (4) is clustered by constructing a distance map and agglomeration tree. Combined with parameter tuning and semi-supervised learning to fuse labeled data, abnormal battery cells are accurately identified. Step (6): If a battery cell is marked as abnormal twice in three consecutive days of data, it is identified as a micro-short-circuited cell; The operational data for step (1) includes the voltage and current of a single battery cell; The noise reduction process in step (3) achieves noise filtering through the synergistic effect of dynamic threshold adjustment and multi-dimensional trend verification, including: screening effective charging data to construct a voltage sequence, setting a start threshold and a cutoff threshold; at the same time, ensuring the rationality of the trend through dynamic incremental constraints, establishing an abnormal interruption response mechanism to terminate abnormal segments in a timely manner and reset the monitoring status, thereby improving the accuracy of charging segment identification and anti-interference capability. The specific steps are as follows: Step (3.1): Extract charging period data: Filter the voltage data points corresponding to the charging process where the current I < 0 from the original data, and construct a charging voltage sequence set; Step (3.2): Determine the dynamic threshold: Based on the distribution of historical charging data, calculate the mean μ and standard deviation σ of the historical voltage data, determine the charging start threshold start and the stop threshold stop, and form a dynamic monitoring window [start, stop]; the calculation formulas for start and stop are as follows: start = μ - 3σ stop = μ + 3σ Step (3.3): Based on the monitoring window obtained in step (3.2), perform dual verification on the voltage sequence obtained in step (3.1), including dynamic increment constraint and abnormal interruption handling; the dynamic increment constraint requires the later value to be greater than the previous value, and a small tolerance δ is set to deal with measurement errors; when the sequence interruption is detected, the current sequence is immediately terminated and the monitoring status is reset. Step (3.4): Repeat steps (3.2) and (3.3) to obtain multiple voltage sequences, and then integrate and splice them to obtain the noise-reduced battery charging voltage curve; In step (4), the difference in curve shape is quantified by calculating the overall rate of change of the battery charging voltage curve, and the calculated overall rate of change sequence is subjected to Z-Score standardization to improve the accuracy of the clustering detection algorithm. The specific steps are as follows: Step (4.1): Calculate the rate of change of the charging voltage curve for each battery cell between day 1 and day 2, and between day 2 and day 3; specifically, calculate the overall rate of change between the curves by averaging the rate of change at each point on the curve, using the following formula: in: The total number of data points for the battery charging voltage curve. and This refers to the voltage value at a specific time point in the first and second charging voltage curves. Step (4.2): The voltage curve change rate obtained in step (4.1) is standardized using Z-Score to further separate abnormal data from normal data within small intervals, which is beneficial for subsequent clustering algorithms. The formula is as follows: in: For the first charge Rate of change of voltage curve for each individual cell and They are respectively The mean and standard deviation of the distribution.

2. The method for early detection of micro-short circuits in battery cells for energy storage power stations according to claim 1, characterized in that, In step (5), the HDBSCAN algorithm is used to perform cluster analysis on the battery charging voltage change rate. Abnormal pattern recognition is achieved by constructing a distance map, agglomeration tree and stability assessment: First, a clustering tree structure is constructed based on density stratification. Stable clusters are screened by cluster persistence measurement. The clustering granularity is controlled by adjustable parameters. Finally, the largest cluster is determined to be a normal battery group, and small clusters and noise points are marked as potential micro-short-circuit cells, i.e. abnormal battery cells. A semi-supervised training mechanism that incorporates a small amount of labeled data can further optimize cluster boundaries and effectively improve the accuracy and noise resistance of micro-short circuit detection.

Images