Energy storage system single battery consistency detection method considering charging and discharging time sequence characteristics

Abstract

The invention belongs to the technical field of energy storage system battery management and state monitoring, and discloses an energy storage system battery monomer consistency detection method and system considering charging and discharging time sequence characteristics. According to the method, historical operation data of each monomer is acquired and preprocessed, an equivalent circuit model and OCV-SOC mapping are established, and the SOC and ohmic internal resistance time sequence is estimated by adopting filtering recursion; screening key influence factors based on correlation analysis and constructing a consistency detection database; a charging and discharging process is extracted according to a single voltage rising edge / falling edge, and a complete charging and discharging cycle time sequence sample is generated; and clustering the samples, calculating the distance to a clustering center, judging an outlier according to an outlier threshold value, positioning inconsistent monomers, and performing adaptive optimization on the clustering category number and the threshold value to output a final result and an alarm level. The method is suitable for a large-scale energy storage operation scene, the detection accuracy, robustness and traceability can be improved, and the safety risk is reduced.

Description

Technical Field

[0001] This invention belongs to the field of battery management and status monitoring technology for energy storage systems, and involves energy storage systems, battery cell operation data processing, timing characteristic analysis and consistency assessment and anomaly early warning, etc. Specifically, it is a method for detecting the consistency of battery cells in energy storage systems that takes into account charge and discharge timing characteristics. Background Technology

[0002] With the large-scale integration of renewable energy sources such as wind and solar power into the power grid, their inherent randomness and volatility pose significant challenges to the safe and stable operation of the grid. Energy storage systems, acting as buffers against uncertainties on both the source and load sides, can compensate for power shortages caused by fluctuations in wind and solar power generation, enabling grid frequency regulation and maintaining safe and stable grid operation. Energy storage systems typically consist of a large number of individual battery cells connected in series and parallel to form battery clusters, and a Battery Management System (BMS) collects and estimates information such as voltage, current, and temperature. However, due to inevitable variations in individual battery cells during manufacturing and production, and these variations amplified over time due to factors such as temperature, rate conditions, state of charge window, and aging, battery packs are more prone to overcharging and over-discharging. Furthermore, individual cell anomalies can trigger a chain reaction within large-scale energy storage systems. When inconsistent batteries are used in series and parallel, capacity and lifespan losses occur, and increased temperature rise and internal resistance accelerate battery aging, potentially leading to overcharging, over-discharging, thermal runaway, and explosions. Therefore, conducting consistency assessments of energy storage systems and identifying abnormal battery cells is an important means to improve the efficiency and safety of energy storage systems.

[0003] Current research on battery consistency testing mainly focuses on cell-level consistency and module-level consistency, and is mostly based on simulation analysis or conducted under laboratory conditions. The complexities of actual energy storage system operation have not yet been analyzed, and consistency testing methods are typically simplified, resulting in low accuracy and high error. Furthermore, existing battery cell consistency testing methods for energy storage systems mainly include mechanistic model-based consistency testing or data-driven consistency testing, both essentially involving finding outliers—that is, identifying points with significant deviations from data and mechanistic perspectives. The selection of outliers and the degree of deviation are generally measured by distance. Since historical data of energy storage systems includes time-series charging and discharging processes, using only the state parameters at a single moment as a benchmark without considering the charging and discharging process is meaningless for measuring the state of each battery cell.

[0004] In the prior art, Chinese patent CN121186598A discloses a method and related device for verifying the consistency of the operating status of an energy storage system. This method collects real-time deformation data at at least three monitoring points on a single battery cell and compensates for it with local temperature data. It calculates the dispersion of standardized deformation and triggers anomalies based on thresholds to determine the risk of structural failure. However, it focuses on deformation consistency and structural early warning, making it difficult to reflect the temporal differences in electrical performance such as voltage / SOC throughout the charging and discharging process. Another example is CN120630025A, which discloses an active detection method and system for battery consistency monitoring. This method implements phased charging and active balancing based on a random charging control function, records the SOC charging curve, calculates the curve segment offset, and outputs anomaly results based on the number of balancing cycles. However, it relies on charging intervention, and the offset / weight settings are empirical, mainly covering the charging process and lacking sufficient temporal representativeness for discharging and actual operating conditions.

[0005] In summary, existing energy storage consistency testing methods have significant shortcomings in terms of accuracy, reliability, and adaptability for large-scale energy storage system applications. Therefore, how to fully utilize the temporal evolution characteristics of battery charging and discharging processes to establish an accurate and reliable battery cell consistency testing method, thereby improving the operational safety and reliability of energy storage systems, is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] (a) Purpose of the invention To address the aforementioned deficiencies and shortcomings of existing technologies, this invention aims to provide a battery cell consistency detection method for energy storage systems that considers charge-discharge timing characteristics. By establishing an equivalent battery model and estimating key state parameters such as State of Charge (SOC) and internal resistance, and combining historical operating data to extract the timing evolution characteristics of the charge-discharge process, a consistency evaluation dataset is constructed. Unsupervised analysis and outlier discrimination are employed to identify, locate, and provide graded early warnings for inconsistent cells. This enables precise location, dynamic early warning, and adaptive optimization of inconsistent battery cells, thereby improving the reliability and accuracy of the detection method and enhancing the operational safety and stability of the energy storage system.

[0007] (II) Technical Solution To achieve the objective of this invention and solve its technical problems, the present invention adopts the following technical solution: A method for detecting the consistency of individual battery cells in an energy storage system, taking into account charge-discharge timing characteristics, wherein the method includes at least the following steps: S100. Historical Operation Data Acquisition: Acquire historical operation data of each battery cell in the energy storage system to form a historical operation dataset for consistency testing; S200. Equivalent model construction and state parameter estimation: Based on historical operating datasets, an equivalent circuit model is established for the energy storage system to characterize the dynamic characteristics of individual battery cells. The model parameters are identified, and the state of charge (SOC) and internal resistance parameters of each battery cell are recursively estimated and updated over time. S300. Correlation analysis of key influencing factors: Based on historical operating datasets, a set of candidate influencing factors is constructed. Correlation analysis is performed on the SOC and internal resistance time series of each battery cell. Candidate influencing factors with an absolute value of correlation coefficient greater than a preset threshold are selected as key influencing factors. S400. Construction of Consistency Detection Database: The fields of the historical operation dataset are standardized and stored in a structured manner. The obtained SOC time series, internal resistance time series and key influencing factors are written into the historical operation dataset to construct the battery cell consistency detection database. S500. Extraction of charging and discharging process and generation of periodic samples: In the consistency detection database, the voltage rise edge and voltage fall edge are identified by the battery cell voltage as the variable. The time sequence data segment corresponding to the voltage rise edge is determined as the charging process data, and the time sequence data segment corresponding to the voltage fall edge is determined as the discharging process data. They are combined to form a complete charging and discharging cycle time sequence sample for each battery cell. S600. Cluster-based consistency detection and outlier determination: Cluster the time series samples of each battery cell's complete charge and discharge cycle, calculate the Euclidean distance between each time series sample and its cluster center, determine the time series samples whose Euclidean distance exceeds the outlier determination threshold as outliers, and output the battery cell corresponding to the outlier as an inconsistent battery cell. S700. Adaptive optimization of clustering parameters: The number of cluster categories and the outlier discrimination threshold are used as variables to be optimized. The variables to be optimized are updated according to the consistency detection effect evaluation index, and the final inconsistent battery cell detection results are output based on the optimized results.

[0008] (III) Technical Effects Compared with the prior art, the energy storage system battery cell consistency detection method of the present invention, which takes into account charge and discharge timing characteristics, has the following beneficial and significant technical effects: (1) This method is geared towards the operation scenario of large-scale energy storage systems. It constructs an equivalent circuit model based on historical operation data and recursively estimates key state parameters such as SOC and internal resistance. Furthermore, it generates complete cycle samples by combining the time-series evolution characteristics of the charging and discharging process. Based on the analysis of the equivalent circuit model of the energy storage system, the SOC of the battery cell, internal resistance and other parameter models, it considers their time-series characteristics, screens and locates inconsistent battery cells, improves the operation safety and reliability of the energy storage system, and enhances the reliability and accuracy of the screening method for inconsistent battery cells in the energy storage system.

[0009] (2) This method screens the key influencing factors of SOC and internal resistance evolution through correlation analysis, and integrates the key factors and state parameter sequences into the consistency detection database, so that the detection process can simultaneously characterize the coupling characteristics of state deviation-cause correlation, reduce the risk of misjudgment caused by a single instantaneous index or isolated outlier judgment, and facilitate the source analysis and graded disposal of inconsistent individuals, thereby improving the stability and engineering applicability of the detection results.

[0010] (3) This method uses the distance between cluster centers as the deviation metric and sets an outlier detection threshold to perform outlier detection on samples of complete charge and discharge cycles. Furthermore, it adaptively optimizes the number of cluster categories and the outlier threshold so that the algorithm parameters can be dynamically adjusted according to the data distribution and operating conditions. This enhances the robustness and generalization ability of consistency detection across battery clusters and operating conditions, thereby improving the accuracy and real-time availability of inconsistent individual cell screening. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic flowchart of a battery cell consistency detection method for an energy storage system that takes into account charge and discharge timing characteristics, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the equivalent circuit model structure used in this invention. The area inside the dashed box is the equivalent circuit part of a single battery cell, and the area outside the dashed box is the external load circuit. Figure 3 This is a schematic diagram showing the Euclidean distance distribution and outlier threshold from different battery cell periodic samples to the cluster center in Embodiment 2 of the present invention; Figure 4 This is a schematic diagram comparing the internal resistance timing curves of inconsistent monomer #17 with those of normal monomers in the same cluster. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. The described embodiments are some, but not all, embodiments of this invention, and are exemplary and intended to explain the invention, not to limit it. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0014] Example 1: Method for Testing the Consistency of Individual Battery Cells in Energy Storage Systems like Figure 1 As shown in the embodiments of the present invention, the battery cell consistency detection method for energy storage systems considering charge-discharge timing characteristics is used to evaluate the consistency of battery cells in a battery cluster and locate inconsistent cells based on historical operating data of the energy storage system. Its implementation mainly includes the following steps: S100. Historical Operational Data Acquisition: Historical operating data of each battery cell in the energy storage system is acquired to form a historical operating dataset for consistency detection. The historical operating data includes at least the cell number, sampling timestamp, terminal voltage, branch current or cell current, temperature measurement point number and temperature, charge / discharge condition identifier, charge / discharge rate information, SOC-related original records, and operating environment-related data for each battery cell in the battery cluster. Each data point forms a traceable time series based on the battery cell dimension. The acquired historical operating data is preprocessed, including at least the following: resampling and timestamp alignment of data from different sampling sources at a unified sampling period; filling in missing segments using interpolation or forward hold and marking the filled-in intervals; removing or correcting outliers based on the physical and statistical boundaries of voltage, current, and temperature; and performing denoising and smoothing processing on the time series data to obtain a continuously usable historical operating dataset.

[0015] S200. Equivalent Model Construction and State Parameter Estimation: Based on historical operational datasets, equivalent circuit models are established for energy storage systems to characterize the dynamic characteristics of individual battery cells. Model parameters are identified, and the state of charge (SOC) and internal resistance parameters of each battery cell are recursively estimated and updated over time. The equivalent model for a single battery cell uses circuit elements such as voltage sources, resistors, and capacitors to simulate the dynamic characteristics of lithium-ion batteries, enabling the estimation and prediction of battery temperature, capacity, and SOC. An example is the Thevenin equivalent circuit model, and based on this, an open-circuit voltage-SOC relationship curve is established. The specific model is as follows: Figure 2 As shown, the equivalent circuit of a single battery cell includes an open-circuit voltage source U. OC Capacitance C characterizes battery capacity and slow dynamic processes. b The ohmic internal resistance R0, which characterizes the ohmic voltage drop, is used to characterize the RC polarization branch, where the polarization resistance is R. p The polarization capacitance is C p And corresponding to the branch current I respectively Rp (t) and I Cp (t); The battery's output terminal is connected to a load resistor R. L The battery terminal voltage is U L The loop current is I(t).

[0016] In this embodiment of the invention, the equivalent circuit model is constructed as a discrete state-space representation with the state of charge (SOC) and polarization voltage of a battery cell as state variables, the battery cell current as input variable, and the battery cell terminal voltage as output variable. The discrete state-space representation is discretized within the sampling period to obtain state equations and observation equations. The state equations are used to describe the dynamic process of SOC changing with the integral of the input current and the polarization voltage decaying with the RC time constant and changing under current excitation. The observation equations are used to describe the terminal voltage and are composed of open-circuit voltage terms, ohmic voltage drop terms, and polarization voltage superposition terms.

[0017] Preferably, the equivalent circuit model can be implemented using any one of the following: Rint model, Thevenin model, first-order RC model, second-order RC model, or PNGV model. The Rint model includes at least an open-circuit voltage source and an ohmic internal resistance branch. The Thevenin model adds at least one RC polarization branch to the Rint model. The second-order RC model includes at least two parallel RC polarization branches to characterize the fast and slow time constant polarization processes, respectively. The PNGV model further includes equivalent elements to describe voltage hysteresis or dynamic bias based on the Thevenin structure.

[0018] Furthermore, based on the completed construction of the equivalent circuit model, Kalman filtering, sliding membrane observer, and other methods are used to fit parameters and predict the state of charge (SOC) of the energy storage system. These methods, given an input signal, obtain an estimate of the output quantity using the state quantities at the current and previous times, as well as the model parameters. The error in the output quantity is used as feedback to iteratively calculate the estimated state quantity. That is, the state estimate is updated based on the observed values ​​and the system model. This invention uses the SOC and internal resistance as estimated values ​​and employs the above methods to fit the SOC and internal resistance parameters of the energy storage system.

[0019] Specifically, in this embodiment of the invention, the estimation and identification of equivalent circuit model parameters, as well as the recursive estimation and timing update of SOC and internal resistance parameters, include the following sub-steps: S201. Data Segment Selection and Initialization: Select voltage and current time series data containing charging and discharging segments from the historical running dataset, and perform noise reduction and normalization on the data to reduce the impact of sampling noise and dimensional differences on the stability of subsequent identification and estimation. S202. Equivalent Circuit Model Definition and Discretization: Define an equivalent circuit model, which includes at least an ohmic internal resistance branch and an RC network polarization branch used to characterize the polarization effect; Discretize the model based on the sampling period, and establish discrete state equations and observation equations, wherein the state variables include at least the state of charge (SOC) and polarization voltage, the input is the cell current, and the observation is the cell terminal voltage. S203. Open Circuit Voltage Mapping and Initial Value Setting: Establish the mapping relationship between SOC and open circuit voltage OCV (Open Circuit Voltage). This mapping relationship is obtained through a preset mapping table, offline calibration curve, or fitting model. Set the initial values ​​of SOC, polarization voltage, and model parameters, and set the initial matrices for process noise covariance and observation noise covariance to meet the initialization conditions of recursive estimation. S204. Model Parameter Identification: Taking the minimization of the predicted residual of the battery cell terminal voltage as the objective function, the parameters of the equivalent circuit model are identified and updated. The model parameters include at least the ohmic internal resistance and the RC network parameters. Pre-set physical constraints such as positive value, range constraints or convergence constraints are applied to each identified model parameter to keep the ohmic internal resistance and RC network parameters non-negative and within the preset reasonable engineering range, thereby avoiding parameter divergence and ensuring interpretability. S205. SOC recursive estimation and time series update: SOC is predicted based on discrete state equations, and SOC is corrected and updated based on the observation equations using the measured values ​​of battery cell terminal voltages to form a SOC time series. The recursive estimation uses a filtering framework to correct the prediction error online, and updates the corresponding error covariance matrix at each sampling time to obtain continuously usable SOC time series estimation results. S206. Recursive estimation and time series update of internal resistance parameter: Based on the recursive estimation of SOC, the ohmic internal resistance is recursively corrected based on the observation residuals and the sensitivity of model parameters, and physical constraints and smoothing constraints are applied to the internal resistance estimate, and the internal resistance time series is output. S207. Result validity verification: Perform boundary verification and consistency verification on the SOC time series and internal resistance time series, remove abnormal estimation points that do not meet the voltage, current and temperature boundary conditions, and write the verified time series data into the consistency detection database for subsequent steps.

[0020] S300. Key influencing factors in analyzing the equivalent circuit model: Based on historical operation datasets, a set of candidate influencing factors is constructed. Correlation analysis methods such as Pearson correlation analysis are used to conduct correlation analysis on the time series of parameters such as the state of charge (SOC) and internal resistance of individual battery cells in the energy storage system. Candidate influencing factors with an absolute value of correlation coefficient greater than a preset threshold (such as 0.3) are selected as key influencing factors, such as the number of charge and discharge cycles, service life, and meteorological factors.

[0021] As a preferred option, the candidate influencing factors set includes one or more of the following: ambient temperature and humidity, cooling medium temperature or air-cooled operation status, charge / discharge rate, SOC operating window, individual cell temperature rise characteristic, cumulative charge / discharge cycles, years of operation and resting time; the correlation analysis uses Pearson correlation coefficient to calculate the correlation between each candidate influencing factor and the SOC time series and internal resistance time series, and the candidate influencing factors with an absolute value of correlation coefficient greater than a preset threshold are identified as key influencing factors.

[0022] S400. Construction of Consistency Detection Database: The historical operational dataset is structured and stored in a structured manner. The resulting SOC time series, internal resistance time series, and key influencing factors are then written into the historical operational dataset to construct a battery cell consistency detection database. The field structured and structured storage of the historical operational dataset includes: establishing an association index between battery cluster-level and cell-level data according to cell number, forming a wide table structure with cell number-timestamp as the primary key. The values ​​of SOC time series, internal resistance time series, and key influencing factors are written into the wide table structure. At the same time, a data quality identification field is recorded to identify the timestamp alignment results, missing test data completion status, and anomaly removal status, so as to support the traceability of consistency detection.

[0023] S500. Extraction of charging and discharging processes and generation of periodic samples: In the consistency detection database, the battery cell voltage is used as a variable to find and identify the voltage rising edge and voltage falling edge. The timing data corresponding to the voltage rising edge and the timing data corresponding to the voltage falling edge are selected as the charging and discharging process data of the energy storage system, respectively. The charging and discharging process data are combined to form a complete charging and discharging cycle timing sample for each battery cell.

[0024] In this embodiment of the invention, the identification of voltage rising and falling edges is based on the joint determination of the first-order differential sign and amplitude threshold, and short-term reverse fluctuations caused by sampling noise are eliminated during the identification process; time-series data segments that satisfy the condition of continuous positive differential and cumulative voltage increment exceeding the threshold are identified as charging process data, and time-series data segments that satisfy the condition of continuous negative differential and cumulative voltage drop exceeding the threshold are identified as discharging process data, and the rest intervals between charging and discharging processes are segmented and marked; the combination rule of complete charging and discharging cycle time-series samples is as follows: the charging process data segment of the same battery cell and the adjacent discharging process data segment constitute a cycle sample, and the start and end times, sample length, rate statistics and temperature statistics are recorded for the cycle sample. When there is an abnormal interruption in the charging or discharging process, the corresponding data segment is marked as an incomplete cycle and is removed or downweighted from the clustered input sample set.

[0025] S600. Cluster-based Consistency Detection and Outlier Determination: For time-series samples of complete charge-discharge cycles of each battery cell, clustering is performed based on clustering algorithms such as K-means. The Euclidean distance between each time-series sample and its cluster center is calculated. Time-series samples whose Euclidean distance exceeds the outlier detection threshold are identified as outliers, and the battery cells corresponding to the outliers are output as inconsistent battery cells. The consistency detection and outlier determination based on clustering includes the following sub-steps: S601. Construction of Clustering Input Samples and Scale Unification: Convert the complete charge-discharge cycle time series samples of each battery cell into clustering input samples. The clustering input samples are sample feature vectors formed according to preset feature dimensions. Perform time scale normalization and dimension normalization processing on samples of different cycles to make each sample comparable in length dimension and amplitude dimension, and form a consistent feature sample set for clustering. S602. Clustering Parameter Acquisition and Center Initialization: Acquire the number of cluster categories K and the cluster center initialization method, initialize K cluster centers based on the consistency feature sample set, and set the clustering iteration termination condition and the minimum intra-cluster error convergence criterion; S603. Clustering Iteration and Sample Attribution Determination: A clustering algorithm (such as K-means clustering algorithm) is used to iteratively update the sample set with consistent characteristics. The sample attribution and cluster center update are performed cyclically until the termination condition is met, and the final cluster center and the cluster category to which each sample belongs are obtained. S604. Distance Measurement and Deviation Calculation: Using Euclidean distance as the distance measurement standard, calculate the Euclidean distance between each sample feature vector and its cluster center to obtain the sample deviation and form the deviation distribution; S605. Outlier Threshold Determination and Outlier Detection: An outlier detection threshold is determined based on the distribution of deviation, wherein the outlier detection threshold is generated by a statistic of deviation, and the statistic includes at least the mean deviation and the standard deviation of deviation; samples whose deviation exceeds the outlier detection threshold are identified as outliers. S606. Result Mapping and Output: Map outliers to their corresponding cell numbers and cycle numbers, output the cell numbers corresponding to outliers as inconsistent cell numbers, and simultaneously output the sample deviation as a measure of the degree of deviation for subsequent steps of parameter optimization and alarm classification.

[0026] S700. Adaptive optimization of clustering parameters: The number of cluster categories and the outlier detection threshold are used as variables to be optimized. These variables are then optimized and updated based on the consistency detection performance evaluation index. The final inconsistent battery cell detection result is output based on the optimized result. To improve the accuracy of consistency detection, this invention uses the inconsistency detection performance (accuracy) as the objective function, and the number of clusters in the k-means clustering algorithm and the distance threshold for outlier detection as variables to be optimized, employing an artificial intelligence optimization algorithm for optimization.

[0027] Preferably, the consistency detection performance evaluation index in step S700 includes at least detection accuracy, false positive rate, and false negative rate. The optimization update adopts the particle swarm optimization algorithm, which searches for the optimal combination of cluster number and outlier discrimination threshold by iteratively updating the position and velocity of particles within a preset search range. During the optimization process, the fitness function is set as a weighted objective function that maximizes detection accuracy and minimizes the sum of false positive rate and false negative rate. The iteration termination condition is reaching the maximum number of iterations or the fitness function remaining unchanged for several consecutive generations. Finally, the optimal number of clusters and outlier discrimination threshold of the objective function are selected as the optimization result. The search range is jointly determined by the battery cluster size, sample size, and historical distance distribution.

[0028] In a further preferred embodiment, when outputting the detection results in steps S600 to S700, a deviation index is further constructed based on the ratio of the Euclidean distance to the outlier discrimination threshold, and the deviation index is mapped to a multi-level alarm level. The alarm level is output together with the battery cell number, cycle number, and key influencing factor values ​​to support the graded handling and source tracing analysis of inconsistent battery cells.

[0029] In addition, during the execution of steps S400 to S700, the consistency detection database is updated on a rolling basis according to a preset time span or a preset number of cycles for the continuous operation data of the energy storage system. After each update, a complete charge and discharge cycle time series sample is regenerated and clustering and outlier determination are performed. When the same battery cell is determined to be an outlier in multiple consecutive cycle samples, the battery cell is output as a continuously inconsistent cell and its outlier duration is recorded.

[0030] Example 2: Application Case Based on Example 1 above, Example 2 uses a lithium iron phosphate energy storage system of an independent energy storage power station as an example to illustrate the engineering applicability and testing effect of the method of the present invention. The energy storage system has a rated power of 50MW and a rated capacity of 100MWh. The battery cluster is composed of several battery packs connected in series and parallel. The BMS records the individual cell terminal voltage, battery cluster current, measuring point temperature, and charge / discharge condition indicators at a sampling period of 1s, and continuously collects 30 days of operating data to form a historical operating dataset.

[0031] Data preprocessing (corresponding to step S100). The full data is timestamped and resampled, and missing segments are padded and marked using forward hold. Outliers are removed by pre-setting physical boundaries based on voltage, current, and temperature, and the terminal voltage sequence is smoothed and denoised to obtain a continuous and usable historical running dataset.

[0032] Equivalent Model and State Parameter Estimation (corresponding to step S200). A Thevenin equivalent circuit model is selected to establish a discrete state-space representation, with SOC and polarization voltage as state variables, cell current as input, and cell terminal voltage as observation. An OCV-SOC mapping table is established and the noise covariance matrix is ​​initialized. A Kalman filter framework is used to recursively estimate the SOC time series, while simultaneously performing online recursive correction of the ohmic internal resistance to output the internal resistance time series. Taking a representative cell in the battery cluster as an example, its median internal resistance is approximately 1.42 mΩ, and the root mean square of the residual between the voltage recovery process corresponding to the polarization time constant and the measured terminal voltage is 8.6 mV, meeting engineering usability requirements.

[0033] Key influencing factor screening and database construction (corresponding to steps S300~S400). Candidate influencing factors include cumulative charge-discharge cycles, years of operation, ambient temperature and humidity, and temperature rise characteristics. The Pearson correlation coefficient between candidate factors and SOC / internal resistance time series is calculated, and the cumulative charge-discharge cycles (|r|=0.46) and temperature rise characteristics (|r|=0.38) are selected as key influencing factors. The SOC time series, internal resistance time series, and key factor fields are written into a structured wide table to form a consistency detection database.

[0034] Periodic sample generation and outlier detection (corresponding to steps S500~S600). Rising and falling edges are identified in the database using the first-order difference of the individual cell voltage. Charging and discharging segments are extracted and combined to form complete charge-discharge cycle samples. The periodic samples are then standardized in terms of time scale and dimensions to construct a cluster feature sample set. K-means clustering is used, and the Euclidean distance from the sample to the cluster center is calculated. An outlier threshold is generated based on the mean plus two standard deviations. Results show that battery cell #17 exceeds the threshold in all five consecutive cycle samples. Its average internal resistance is 2.05 mΩ (approximately 44% higher than the cluster median), and its SOC estimation sequence exhibits a stable lag relative to the cluster average, classifying it as an inconsistent cell. Maintenance review revealed that the battery pack connection resistance corresponding to this cell is excessively high, and the temperature rise is also excessively high, verifying the interpretability of the detection conclusion.

[0035] Parameter adaptive optimization and comparison results (corresponding to step S700). With the objective function of maximizing detection accuracy and minimizing the weighted average of false positive and false negative rates, particle swarm optimization is used to search for K and the outlier threshold. Before optimization (K=3), the accuracy was 89.1%, the false positive rate was 6.4%, and the false negative rate was 4.5%; after optimization (K=4), the accuracy increased to 95.2%, the false positive rate decreased to 2.7%, and the false negative rate decreased to 2.1%. In contrast, the method using only "single-moment voltage / temperature outliers" generated false alarms for 6 individual cells during load fluctuation periods, while the present invention maintains stability based on time-series samples from a complete charge-discharge cycle.

[0036] Figure 3 This figure illustrates the Euclidean distance distribution from different battery cell periodic samples to the cluster centers and the outlier threshold. As shown, after cluster analysis of 24 battery cell periodic samples, the Euclidean distance from most normal cells to the cluster centers is concentrated between 11.8 and 16.1, with an average distance of 14.25. The outlier threshold was determined to be 19.86 by calculating the mean plus twice the standard deviation. Cell #17 has an Euclidean distance of 28.6, far exceeding the outlier threshold, indicating that its characteristic parameters deviate significantly from the normal cluster centers, and it was successfully identified as an inconsistent cell. This verifies that the Euclidean distance-based outlier detection method of this invention can effectively distinguish abnormal batteries. Figure 4 This is a schematic diagram comparing the internal resistance time-series curves of inconsistent monomer #17 and normal monomers in the same cluster. As shown in the figure, within a 30-day operating period, the average internal resistance of the normal monomers in the same cluster slowly increased from 1.42 mΩ to 1.50 mΩ, an increase of approximately 5.6%; while the internal resistance of inconsistent monomer #17 rapidly deteriorated from 1.45 mΩ to 2.05 mΩ, an increase of as much as 41.4%, which is 36.7% higher than the median value within the cluster. This significant difference in internal resistance growth indicates that monomer #17 has a serious performance degradation problem and needs timely maintenance or replacement, thus verifying the practical value and accuracy of the detection method of this invention.

[0037] The objectives of this invention have been fully and effectively achieved through the above embodiments. Those skilled in the art will understand that this invention includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments described above. Although the invention has been described with reference to what is currently considered the most practical and preferred embodiments, it should be understood that the invention is not limited to the disclosed embodiments, and any modifications that do not depart from the functional and structural principles of the invention will be included within the scope of the claims.

Claims

1. A method for detecting the consistency of individual battery cells in an energy storage system, taking into account charge-discharge timing characteristics, characterized in that, At least the following steps are included: S100. Obtain historical operating data of each battery cell in the energy storage system to form a historical operating dataset for consistency detection; S200. Based on historical operating datasets, establish an equivalent circuit model characterizing the dynamic characteristics of a single battery cell, identify model parameters, and recursively estimate and update the SOC and internal resistance parameters of the battery cell. S300. Construct a set of candidate influencing factors based on historical operation datasets, perform correlation analysis on the SOC and internal resistance time series of each battery cell, and screen candidate influencing factors with an absolute value of correlation coefficient greater than a preset threshold as key influencing factors. S400. Perform field normalization and structured storage on the historical operation dataset, and write the obtained SOC time series, internal resistance time series and key influencing factors selected into the historical operation dataset to build a battery cell consistency detection database; S500. Using the battery cell voltage as a variable, identify the voltage rising edge and voltage falling edge, determine the timing data corresponding to the voltage rising edge as the charging process data, determine the timing data corresponding to the voltage falling edge as the discharging process data, and combine them to form a complete charge and discharge cycle timing sample for each battery cell. S600. Cluster the time series samples of each battery cell's complete charge and discharge cycle, calculate the Euclidean distance between each time series sample and its cluster center, identify time series samples whose Euclidean distance exceeds the outlier detection threshold as outliers, and output the battery cell corresponding to the outlier as an inconsistent battery cell. S700. The number of cluster categories and the outlier discrimination threshold are used as variables to be optimized. The variables to be optimized are updated according to the consistency detection effect evaluation index, and the final inconsistent battery cell detection results are output based on the optimized results.

2. The method according to claim 1, characterized in that, In step S100, the historical operating data includes at least the number of each battery cell in the battery cluster, sampling timestamp, terminal voltage, branch current or cell current, temperature measurement point number and measurement point temperature, charge and discharge condition identifier, charge and discharge rate information, SOC related original records and operating environment related data. The acquired historical operating data is preprocessed, including at least the following: resampling and aligning the data from different sampling sources with timestamps according to a uniform sampling period; filling missing segments with interpolation or forward hold and marking the filling interval; removing or correcting outliers based on the physical and statistical boundaries of voltage, current, and temperature; and performing noise reduction and smoothing on the time series data.

3. The method according to claim 1, characterized in that, In step S200, the equivalent circuit model is used to describe the dynamic response of the terminal voltage of a single battery cell during the charging and discharging process. It includes at least a battery open-circuit voltage source, an ohmic internal resistance branch for characterizing the ohmic voltage drop, and at least one RC network polarization branch for characterizing the electrochemical polarization effect. The RC network polarization branch is composed of a polarization resistor and a polarization capacitor connected in series, and together with the ohmic internal resistance branch, it determines the transient and slow-change characteristics of the terminal voltage.

4. The method according to claim 3, characterized in that, In step S200, the equivalent circuit model is constructed as a discrete state-space expression with the state of charge (SOC) and polarization voltage of the battery cell as state variables, the battery cell current as input variable, and the battery cell terminal voltage as output variable. The discrete state-space expression is discretized within the sampling period to obtain the state equation and the observation equation. The state equation is used to describe the dynamic process of the SOC changing with the integral of the input current and the polarization voltage decaying with the RC time constant and changing under current excitation. The observation equation is used to describe the terminal voltage and is composed of the open-circuit voltage term, the ohmic voltage drop term, and the polarization voltage superposition term.

5. The method according to claim 1, 3, or 4, characterized in that, In step S200, the model parameter identification and the recursive estimation and time-series update of SOC and internal resistance parameters include the following sub-steps: S201. Data Segment Selection and Initialization: Select voltage and current timing data that include the charging and discharging segments from the historical operation dataset; S202. Equivalent Circuit Model Definition and Discretization: Define an equivalent circuit model, which includes at least an ohmic internal resistance branch and an RC network polarization branch used to characterize the polarization effect; Discretize the model based on the sampling period, and establish discrete state equations and observation equations, wherein the state variables include at least the state of charge (SOC) and polarization voltage, the input is the cell current, and the observation is the cell terminal voltage. S203. Open-circuit voltage mapping and initial value setting: Establish the mapping relationship between SOC and open-circuit voltage, set the initial values ​​of SOC, polarization voltage and model parameters, and set the initial matrix for process noise covariance and observation noise covariance to meet the initialization conditions of recursive estimation. S204. Model parameter identification: Using the prediction residual of minimizing the terminal voltage of a single battery cell as the objective function, the parameters of the equivalent circuit model are identified and updated. The model parameters include at least the ohmic internal resistance and the RC network parameters, and preset physical constraints are applied to each identified model parameter. S205. SOC Recursive Estimation and Time Series Update: SOC is predicted based on discrete state equations, and SOC is corrected and updated using the measured values ​​of battery cell terminal voltage based on observation equations to form a SOC time series. The recursive estimation uses a filtering framework to correct the prediction error online. S206. Recursive estimation and time series update of internal resistance parameter: Based on the recursive estimation of SOC, the ohmic internal resistance is recursively corrected based on the observation residuals and the sensitivity of model parameters, and physical constraints and smoothing constraints are applied to the internal resistance estimate, and the internal resistance time series is output. S207. Result validity verification: Perform boundary verification and consistency verification on the SOC time series and internal resistance time series, and eliminate abnormal estimation points that do not meet the voltage, current and temperature boundary conditions.

6. The method according to claim 1, characterized in that, In step S300, the set of candidate influencing factors includes one or more of the following: ambient temperature and humidity, cooling medium temperature or air-cooled operation status, charge / discharge rate, SOC operating window, individual cell temperature rise characteristic, cumulative charge / discharge cycles, years of operation and resting time; the correlation analysis uses Pearson correlation coefficient to calculate the correlation between each candidate influencing factor and the SOC time series and internal resistance time series, and the candidate influencing factors whose absolute value of the correlation coefficient is greater than a preset threshold are identified as key influencing factors.

7. The method according to claim 1, characterized in that, In step S400, the field regularization and structured storage of the historical operation dataset includes: establishing an associated index for battery cluster-level and individual-level data according to the battery individual number, forming a wide table structure with the individual number-time stamp as the primary key, and writing the SOC time series, internal resistance time series and values ​​of key influencing factors into the wide table structure.

8. The method according to claim 1, characterized in that, In step S500, the identification of voltage rising edge and voltage falling edge is based on the joint determination of voltage first-order differential sign and amplitude threshold, and short-term reverse fluctuations caused by sampling noise are eliminated during the identification process. The time series data segments that satisfy the condition that the differential is continuously positive and the cumulative voltage increment exceeds the threshold are determined as charging process data, and the time series data segments that satisfy the condition that the differential is continuously negative and the cumulative voltage drop exceeds the threshold are determined as discharging process data. The rest intervals between the charging and discharging processes are divided and marked.

9. The method according to claim 8, characterized in that, In step S500, the combination rule for the complete charge and discharge cycle time series sample is as follows: the charging process data segment of the same battery cell and the adjacent discharge process data segment constitute a cycle sample, and the start and end time, sample length, rate statistics and temperature statistics are recorded for the cycle sample. When there is an abnormal interruption in the charging or discharging process, the corresponding data segment is marked as an incomplete cycle and removed or downweighted from the clustered input sample set.

10. The method according to claim 1, characterized in that, In step S600, the cluster-based consistency detection and outlier determination includes the following sub-steps: S601. Construction of Clustering Input Samples and Scale Unification: Convert the complete charge-discharge cycle time series samples of each battery cell into clustering input samples, and perform time scale normalization and dimension normalization processing on samples of different cycles to form a consistent feature sample set for clustering; S602. Clustering Parameter Acquisition and Center Initialization: Acquire the number of cluster categories K and the cluster center initialization method, initialize K cluster centers based on the consistency feature sample set, and set the clustering iteration termination condition and the minimum intra-cluster error convergence criterion; S603. Clustering Iteration and Sample Attribution Determination: The clustering algorithm is used to iteratively update the sample set with consistent characteristics, and the sample attribution and cluster center update are performed cyclically until the termination condition is met, so as to obtain the final cluster center and the cluster category to which each sample belongs; S604. Distance Measurement and Deviation Calculation: Calculate the Euclidean distance between each sample feature vector and its cluster center to obtain the sample deviation and form the deviation distribution; S605. Determination of outlier threshold and outlier identification: Determine the outlier identification threshold based on the distribution of deviation amount, and identify samples whose deviation amount exceeds the outlier identification threshold as outliers; S606. Result Mapping and Output: Map outliers to their corresponding cell numbers and cycle numbers, and output the cell numbers corresponding to the outliers as inconsistent cell numbers.

11. The method according to claim 1, characterized in that, In step S700, the consistency detection effect evaluation index includes at least detection accuracy, false detection rate, and false negative rate; the optimization update adopts the particle swarm optimization algorithm, which searches for the optimal combination of clustering category number and outlier discrimination threshold by iteratively updating the position and velocity of particles within a preset search range. During the optimization process, the fitness function is set as a weighted objective function that maximizes the detection accuracy and minimizes the sum of false detection rate and false negative rate. The iteration termination condition is reaching the maximum number of iterations or the fitness function remaining unchanged for several consecutive generations. Finally, the optimal category number and outlier discrimination threshold of the objective function are selected as the optimization result.

12. The method according to claim 1, characterized in that, During the execution of steps S400~S700, the consistency detection database is updated on a rolling basis according to a preset time span or a preset number of cycles for the continuous operation data of the energy storage system. After each update, a complete charge and discharge cycle time series sample is regenerated and clustering and outlier determination are performed. When the same battery cell is determined to be an outlier in multiple consecutive cycle samples, the battery cell is output as a continuously inconsistent cell and its outlier duration is recorded.

Images