Method and system for identifying and early warning abnormal failure cells of energy storage battery

Abstract

The invention relates to a method and system for identifying and early warning abnormal failure cells of an energy storage battery, and the method comprises the steps: carrying out the estimation of the health state of a whole-station battery cell of an energy storage power station, obtaining a cell estimation value, and endowing the cell estimation value with a corresponding confidence coefficient; calculating the intrinsic health characteristics of each battery, and marking the batteries with the intrinsic health characteristics not in a set range as suspicious single batteries; comparing and analyzing the batteries in the neighbor group to calculate a root cause discrimination score, an estimated value of a local state space and a corresponding confidence coefficient; carrying out adaptive weight fusion on the single body estimation value and the estimation value of the local state space to obtain a final estimation value; and finishing graded early warning of the abnormal state of the energy storage battery based on the final estimation value. According to the invention, through an innovative integrated mechanism of "abnormal accumulation triggering-local space diagnosis-game weight fusion-grading early warning decision", multilayer intelligent discrimination and grading response of abnormal states are realized.

Description

Technical Field

[0001] This invention relates to the field of battery abnormal deterioration identification technology, and in particular to a method and system for identifying and warning of abnormal deterioration of individual cells in energy storage batteries. Background Technology

[0002] As the global energy structure shifts towards low-carbon and high-proportion renewable energy, electrochemical energy storage power stations are showing a trend towards large-scale, clustered deployment. Currently, in mainstream lithium-ion energy storage systems, a single 200MWh-class power station typically contains approximately 200,000 battery cells, while in larger-scale or multi-site clustered applications, the number of battery cells can reach millions. Against this backdrop, accurate assessment of the State of Health (SOH) and early anomaly identification in large-scale energy storage systems have become key technologies for ensuring safe system operation and improving operational efficiency.

[0003] However, existing SOH estimation techniques face significant contradictions in engineering applications. On the one hand, while high-precision estimation methods based on refined electrochemical models or complete cycle data can achieve high reliability, their algorithmic complexity is high, and their computational and storage resource consumption is significant, making it difficult to maintain high-frequency, continuous operation on hundreds of thousands of battery cells. For example, in a typical 100MW / 200MWh energy storage power station, simultaneously performing high-precision SOH calculations on all 200,000 cells in the station would cause the computational task to accumulate linearly with the number of cells to an extremely high level, thus affecting system real-time performance and operating costs. On the other hand, lightweight estimation methods based on voltage, current segments, or impedance characteristics, while having lower resource consumption, are limited by data quality, operating condition fluctuations, and the degree of model simplification, resulting in insufficient stability of the estimation results and making it difficult to meet the confidence and reliability requirements of engineering scenarios.

[0004] In terms of anomaly diagnosis, existing technologies typically rely on fixed thresholds or single health indicators for judgment, failing to effectively combine the multidimensional operating characteristics of batteries and their behavior patterns over time. When sudden changes occur in SOH or related characteristics, the system struggles to distinguish whether the cause is measurement noise, algorithm error, or actual performance degradation, leading to both false alarms and false negatives, thus affecting the effectiveness of anomaly identification.

[0005] Furthermore, most current energy storage battery monitoring systems exhibit a siloed architecture with fragmented functional modules, lacking coordination mechanisms between SOH estimation, feature analysis, anomaly detection, and alarm decision-making. Uncertainties in the output of upstream modules cannot be effectively utilized by downstream modules, and information such as battery group behavior analysis and long-term time-series feature mining is often overlooked, limiting the improvement of overall system performance.

[0006] Meanwhile, the aging process of batteries is characterized by nonlinearity, condition dependence, and time-varying nature. The state of harmonic decay (SOH) rate changes dynamically with factors such as temperature, rate capability, and operating strategy. Fixed-threshold diagnostic strategies are difficult to adapt to the characteristics of the entire battery life cycle. They are often too conservative in the early stages and may be slow to respond in the later stages of degradation, failing to achieve adaptive anomaly identification and early warning.

[0007] In summary, existing technologies still have shortcomings in terms of computing resource scheduling, adaptive diagnostic capabilities, multi-dimensional information fusion, and large-scale deployment capabilities. There is an urgent need for a new monitoring and early warning technology solution that can achieve highly reliable State of Health (SOH) assessment, robust anomaly identification, and intelligent decision-making through multi-module collaboration in resource-constrained large-scale energy storage applications, thereby improving the safety and full lifecycle operation management level of energy storage systems. Summary of the Invention

[0008] To address the aforementioned problems, the purpose of this invention is to provide a method and system for identifying and warning of abnormal battery cell deterioration.

[0009] A method for identifying and providing early warning of abnormal battery cell deterioration in energy storage systems includes:

[0010] Step 1: Estimate the health status of all battery cells in the energy storage power station to obtain individual cell estimates, and assign corresponding confidence levels to the individual cell estimates;

[0011] Step 2: Calculate the intrinsic health characteristics of each battery and mark batteries whose intrinsic health characteristics are not within the set range as suspicious cells;

[0012] Step 3: Construct a neighbor group of suspected individual cells based on the electrical connection topology of the battery cluster and the thermal management zoning model;

[0013] Step 4: Compare and analyze the batteries in the neighboring group to calculate the root cause discrimination score, the estimated value of the local state space and its corresponding confidence level;

[0014] Step 5: Based on the confidence scores of the individual unit estimates, the confidence scores of the local state space estimates, and the root cause discrimination scores, the individual unit estimates and the local state space estimates are adaptively weighted and fused to obtain the final estimate.

[0015] Step 6: Complete the graded early warning of abnormal states of energy storage batteries based on the final estimated values.

[0016] Preferably, in step 1, when the battery completes an effective full charge-discharge cycle, the estimated single-cell OH value of the battery is obtained by the ampere-hour integration method. local And assign a corresponding confidence level C. local; where an effective full-charge-discharge cycle is: a complete process from a fully charged state to the cutoff voltage, with a discharge depth exceeding 90%; for a time window T window For batteries that have not completed a valid full charge-discharge cycle, when the cycle time t ≥ T window At that time, a lightweight estimation algorithm based on fragment data obtains the estimated single-cell SOH value of the battery. local And assign a corresponding confidence level C local .

[0017] Preferably, in step 2, the intrinsic health characteristics include: capacity-resistance recombination factor, relaxation voltage recovery rate, constant current charging voltage curve slope, and incremental capacity characteristic peak shift; wherein, the capacity-resistance recombination factor F CR The calculation formula is:

[0018]

[0019] Where ΔQ is the segment capacity, I is the average current, and R is the DC internal resistance;

[0020] Relaxation voltage recovery rate η Vr The calculation formula is:

[0021]

[0022] Among them, V start V end These are the starting and ending voltages of the resting period, V. relax This is the theoretical equilibrium voltage.

[0023] Preferably, step 3: constructing a neighbor group of suspected cells based on the electrical connection topology of the battery cluster and the thermal management zoning model includes:

[0024] Step 3.1: Calculate the electrical correlation degree based on the electrical connection topology diagram of the battery cluster; wherein, the formula for calculating the electrical correlation degree is:

[0025]

[0026] Where, d e (v t ,v i ) represents the shortest path hop count in the electrical topology, v t Indicates a suspicious monomer, v i Let S represent the target single entity, ∈ be the smoothing factor, and S e (v t ,v i ) represents electrical correlation degree;

[0027] Step 3.2: Calculate the thermal management correlation degree based on the thermal management zoning model's distance between thermal management zones; whereby the formula for calculating the thermal management correlation degree is:

[0028]

[0029] Where, d t (v t ,v i ) represents the physical spatial distance, σ t This refers to the parameters related to the heat-affected zone.

[0030] Step 3.3: When S is satisfied simultaneously e (v t ,v i )>θ e and S t (v t ,v i )>θ t At that time, the monomer v i Include the neighboring groups of the suspected individual.

[0031] Preferably, step 4: performing comparative analysis on the batteries in the neighboring group to calculate the root cause discrimination score, the estimated value of the local state space, and its corresponding confidence level, includes:

[0032] Step 4.1: Calculate the correlation coefficient of the historical SOH trajectories of the target individual and its neighboring populations to achieve trajectory consistency comparison; the trajectory consistency comparison results are as follows:

[0033]

[0034] Where, ρ traj This indicates the trajectory consistency comparison result, SOH target Represents the health status history sequence of the target monomer, SOH neighbors σ represents the historical sequence of the health status of a neighborhood group. target σ represents the standard deviation of the target monomer SOH sequence. neighbors This represents the standard deviation of the SOH sequences in the neighboring population;

[0035] Step 4.2: The feature coherence comparison is achieved by evaluating the difference in feature vectors between the target individual and its neighboring groups at the current time using Mahalanobis distance; the feature coherence comparison result is as follows:

[0036]

[0037] Among them, D m This represents the feature coherence comparison result, where Ftarget represents the feature vector of the target individual at the current time, μneighbors is the mean vector of the feature vectors of the neighboring group, and Σ neighborsLet be the covariance matrix of the eigenvectors of the neighboring group;

[0038] Step 4.3: Calculate the root cause discrimination score based on the comparison results; the formula for calculating the root cause discrimination score is as follows:

[0039] P root =σ(α·(1-ρ) traj )+β·D m )

[0040] Among them, P root σ(·) represents the root cause discrimination score, σ(·) is the Si gmoi d function, and α and β are weight coefficients;

[0041] Step 4.4: Calculate the estimated value of the local state space based on the state of healthy individuals in the neighboring group, and assign the corresponding confidence level; wherein, the estimated value of the local state space is:

[0042]

[0043] Among them, SOH spatial w represents the estimate of the local state space. j Let SOH represent the j-th weight. j This represents the state of the j-th healthy individual in the neighboring group.

[0044] Preferably, step 5: adaptively weighting the individual unit estimates and the local state space estimates to obtain the final estimate based on the confidence levels corresponding to the individual unit estimates, the confidence levels corresponding to the local state space estimates, and the root cause discrimination scores, includes:

[0045] Step 5.1: In the game model, the individual estimates and local state space estimates are treated as two game participants to construct utility functions, and the fusion weight is calculated; the formula for calculating the fusion weight is:

[0046] U local =C local ·P root

[0047] U spatial =C spatial ·(1-P root )

[0048]

[0049] Among them, U local U represents the degree of confidence in one's own estimate when there is suspicion of true degradation. spatial C represents the degree of dependence on spatial consensus when the error is suspected to be an estimation error. local C represents the confidence level of the individual estimates. spatialP represents the confidence level corresponding to the local state-space estimate. root α represents the root cause discrimination score, α represents the fusion weight, and T represents the temperature parameter.

[0050] Step 5.2: Calculate the final estimate based on the fusion weights; the formula for calculating the final estimate is:

[0051] SOH final =α·SOH local +(1-α)·SOH spatial

[0052] Among them, SOH final This represents the final estimated value, SOH. local This represents the estimated value of the monomer, SOH. spatial This represents the local state space estimate.

[0053] Preferably, step 6: completing the graded early warning of abnormal states of the energy storage battery based on the final estimated value includes:

[0054] Step 6.1: Construct a comprehensive risk assessment function using the final estimated value, root cause discrimination score, decay rate, and decay acceleration; whereby the comprehensive risk assessment function is:

[0055] R = w1·P root +w2·|SOH dev |+w3·|r norm |+w4·max(0,a norm )

[0056] Where R represents the risk assessment result, w1 represents the first weighting coefficient, w2 represents the second weighting coefficient, w3 represents the third weighting coefficient, w4 represents the fourth weighting coefficient, and SOH... dev Indicates the relative offset. SOH threshold This represents the preset health status safety threshold, r norm a represents the normalized decay rate. norm This represents the normalized decay acceleration;

[0057] Step 6.2: Generate a three-level warning using the risk assessment results. When 0.1≤R<0.3, trigger an observation-level warning; when 0.3≤R<0.7, trigger an alert-level warning; when R≥0.7, trigger an emergency-level warning.

[0058] This invention also provides a system for identifying and warning of abnormal cell deterioration in energy storage batteries, comprising:

[0059] The single-cell health status module is used to estimate the health status of all battery cells in the energy storage power station to obtain a single-cell estimate, and to assign a corresponding confidence level to the single-cell estimate;

[0060] The suspicious cell identification module is used to calculate the intrinsic health characteristics of each battery and mark batteries whose intrinsic health characteristics are not within the set range as suspicious cells;

[0061] The neighbor group building module is used to build neighbor groups of suspected individual cells based on the electrical connection topology graph and thermal management zoning model of battery clusters;

[0062] The comparison analysis module is used to perform comparison analysis on batteries in the neighbor group to calculate the root cause discrimination score, the estimated value of the local state space and its corresponding confidence level;

[0063] The fusion module is used to adaptively weight and fuse the individual estimates and the local state space estimates based on the confidence scores of the individual estimates, the confidence scores of the local state space estimates, and the root cause discrimination scores to obtain the final estimate.

[0064] The early warning module is used to provide graded early warnings of abnormal states of energy storage batteries based on the final estimated values.

[0065] The present invention also provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus, characterized in that the computer program, when executed by the processor, implements the steps in the above-described method for identifying and warning of abnormal cell failure in an energy storage battery.

[0066] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps in the above-described method for identifying and warning of abnormal cell failure in an energy storage battery.

[0067] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0068] This invention relates to a method for identifying and warning of abnormal cell deterioration in energy storage batteries. Compared with the prior art, this invention achieves multi-layer intelligent identification and hierarchical response to abnormal states through an innovative integrated mechanism of "abnormal accumulation triggering → local spatial diagnosis → game weight fusion → hierarchical early warning decision".

[0069] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0071] Figure 1 The flowchart illustrates a method for identifying and warning of abnormal cell deterioration in energy storage batteries provided by this invention. Detailed Implementation

[0072] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0073] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0074] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0075] Please see Figure 1 A method for identifying and warning of abnormal cell deterioration in energy storage batteries, comprising:

[0076] Step 1: Estimate the health status of all battery cells in the energy storage power station to obtain individual cell estimates, and assign corresponding confidence levels to the individual cell estimates;

[0077] A dual-cycle SOH hybrid update mechanism is established, the core of which lies in balancing estimation accuracy and computational cost through different strategies. The main update event is defined as the completion of a valid full charge-discharge cycle of the battery. At this point, a high-confidence single-cell SOH estimate is obtained using the ampere-hour integration method. local It also outputs its high-level confidence level C. local .

[0078] For time window T window For batteries that have not completed a valid full charge-discharge cycle, a time-triggered condition t≥T is introduced. window When this condition is met, the system initiates a lightweight estimation algorithm based on fragment data to obtain the estimated SOH value under this operating condition. local and its corresponding confidence level C local This confidence level will comprehensively consider the length, quality, and representativeness of the data segment, thereby ensuring the continuity of system monitoring and providing a unified reliability metric for data sources of different quality.

[0079] Step 2: Calculate the intrinsic health characteristics of each battery and mark batteries whose intrinsic health characteristics are not within the set range as suspicious cells;

[0080] At the same trigger moment of the SOH update, multi-dimensional intrinsic health characteristics decoupled from operating conditions are simultaneously calculated, forming a multimodal evidence chain. Core characteristics include: the capacity-resistance composite factor F. CR Relaxation voltage recovery rate η Vr The slope k of the constant current charging voltage curve CC and incremental capacity characteristic peak shift These features, through ratio calculation and differential analysis, eliminate interference from major operating conditions, are directly related to the battery aging mechanism, and are generated synchronously with the SOH estimate, providing highly robust intrinsic evidence for triggering the decision.

[0081] Establish an abnormal cumulative memory value A for each individual i And introduce the forgetting factor λ and the dynamic threshold θ i When an anomaly is found in a monomer during an evaluation, update A. i (t+1)=λA i (t)+Δ i (t), where Δ i (t) represents the anomaly intensity. If A i Exceeding the threshold θ in the short term i The system immediately triggers a diagnosis; if the growth continues for multiple cycles, its priority is adaptively increased, eventually entering a high-precision diagnostic process. This mechanism balances short-term sensitive response with long-term trend recognition, enabling intelligent early screening of abnormally deteriorating monomers.

[0082] Step 3: Construct a neighbor group of suspected individual cells based on the electrical connection topology of the battery cluster and the thermal management zoning model;

[0083] Dynamic Neighbor Network Construction: Targeting Tagged Suspicious Individuals v t Based on the battery cluster electrical connection topology diagram G e (V,E) and thermal management zoning model M t The neighbor network N is dynamically constructed through multiple constraints. t Define the electrical correlation degree S. e (v t ,v i Correlation between thermal management and S t (v t ,v i ), where S e Based on the shortest path calculation of electrical topology, S t Calculation based on thermal management zone distance. When S is satisfied... e (v t ,v i )>θ e And S t (v t ,v i )>θ t At that time, the monomer v i The dynamic neighbor network N of the target unit t This network ensures that neighboring cells and the target cell maintain strong correlation in both electrical connectivity and thermal management, forming a local reference benchmark with clear engineering and physical significance.

[0084] Step 4: Compare and analyze the batteries in the neighboring group to calculate the root cause discrimination score, the estimated value of the local state space and its corresponding confidence level;

[0085] Within the constructed local health state space, a dual comparison analysis is performed between the suspected individual and its neighboring network. Trajectory consistency comparison is achieved by calculating the correlation coefficient ρ between the target individual and its neighboring groups' historical SOH trajectories. traj accomplish:

[0086] ρ traj =Cov(SOH) target SOH neighbors ) / (σ target ·σ neighbors )

[0087] Feature co-orientation matching evaluates the Mahalanobis distance D between the target individual and its neighboring groups of multidimensional feature vectors at the current time. m accomplish:

[0088]

[0089] This dual comparison mechanism comprehensively assesses the degree of individual anomaly from two dimensions: temporal evolution and instantaneous state, providing a quantitative basis for subsequent root cause identification.

[0090] Based on the results of the dual comparison analysis, a quantitative model for root cause identification is established. The root cause identification score P is defined. root for:

[0091] P root =σ(α·(1-ρ) traj )+β·D m )

[0092] Where σ(·) is the Sigmoid function, α and β are weighting coefficients, and ρ traj D is the trajectory correlation coefficient. m The characteristic Mahalanobis distance is P. root ∈[0,1], the closer the value is to 0, the more likely it is an estimation error, and the closer it is to 1, the more likely it is a true degradation.

[0093] Simultaneously, based on the state of healthy individuals in the neighboring group, a health state value SOH based on the local state space is generated. spatial and its confidence level:

[0094]

[0095] C spatial =f(σ neighbors ,N,ρ traj )

[0096] Where w j σ is the weight coefficient of neighboring individual j, N is the number of healthy individuals in the neighboring network, and σ is the weight coefficient of neighboring individual j. neighbors This represents the standard deviation of the SOH values ​​of the neighboring population. This mechanism achieves fine-grained identification of root causes through continuous quantization and provides downstream fusion modules with spatial reference values ​​with confidence levels, ensuring consistency with the upstream data interface.

[0097] Step 5: Based on the confidence scores of the individual unit estimates, the confidence scores of the local state space estimates, and the root cause discrimination scores, the individual unit estimates and the local state space estimates are adaptively weighted and fused to obtain the final estimate.

[0098] A game theory-based weighted decision system receives SOH from upstream. local C local SOH spatial C spatial and P root To construct a lightweight game theory model to calculate the optimal fusion weights.

[0099] The game weight α obtained from the above steps can be used to generate the final estimated value SOH.final Then update it to the historical database to complete the state estimation for the current period:

[0100] SOH final =α·SOH local +(1-α)·SOH spatial

[0101] On the other hand, when high-precision full-charge and discharge cycle measurements are obtained, the system initiates a parameter optimization mechanism. Using the SOH value as a benchmark, it backtracks to analyze the deviations of fusion results from adjacent cycles, and optimizes the temperature parameter T based on statistical analysis, thereby achieving continuous improvement of system parameters.

[0102] Step 6: Complete the graded early warning of abnormal states of energy storage batteries based on the final estimated values.

[0103] Receive multi-source decision information from upstream processing units, including root cause discrimination score P. root Final SOH estimate final Key parameters such as SOH decay kinetics and fusion weight α are identified. A unified information fusion framework is established to standardize and integrate the aforementioned multi-source data, ensuring that data from different sources and time scales can consistently support decision analysis.

[0104] Based on a pre-defined multi-level early warning rule base, the system comprehensively evaluates the fused information. According to the evaluation results, the system generates differentiated early warning levels, including observation, alert, and emergency levels, with each level accompanied by corresponding operation and maintenance guidance suggestions. The final decision is output to the operation and maintenance management system through a standardized interface, forming a complete closed loop from status monitoring to operation and maintenance execution, achieving accurate identification and intelligent operation and maintenance of abnormal states of energy storage batteries.

[0105] This invention achieves multi-layered intelligent identification and graded response to abnormal states through an innovative integrated mechanism of "anomaly accumulation triggering → local spatial diagnosis → game-theoretic weight fusion → graded early warning decision-making." Specifically, the anomaly accumulation triggering mechanism effectively distinguishes between short-term disturbances and persistent anomalies by accumulating long-term memory of multimodal features and adaptive threshold adjustment; the local spatial diagnosis mechanism focuses on high-risk cells to accurately locate the root cause of anomalies; the game-theoretic weight fusion mechanism achieves reliable weighting among multi-source information; and the graded early warning mechanism completes risk stratification and operation and maintenance closure. While ensuring the accuracy of state estimation, this system concentrates high-precision computing resources on the most critical battery cells, significantly reducing the average computational load of the system. Ultimately, it achieves early identification, accurate location, and intelligent early warning of abnormally deteriorating cells in energy storage batteries, providing crucial support for the safe and economical operation of energy storage power stations.

[0106] The working principle of the present invention will be further explained below with reference to specific embodiments:

[0107] Step 1: Construct an anomaly accumulation triggering unit. The unit adopts a hybrid update strategy with event-driven as the main approach and time cycle as the auxiliary approach to complete the low-power estimation and update of the state of health (SOH) of all battery cells in the energy storage power station, monitor the multi-dimensional characteristic indicators of the cell health status, establish an anomaly accumulation memory value for each cell, and combine the forgetting factor and dynamic threshold to perform adaptive triggering judgment, realize intelligent identification of abnormal cells, and trigger the downstream high-precision diagnostic process.

[0108] Step 1 includes the following specific steps:

[0109] Step 1.1 Establish a dual-cycle SOH hybrid update mechanism. The system defines the main update event as the completion of a valid full-charge-discharge cycle, specifically the complete process of discharging from a fully charged state to the cutoff voltage with a discharge depth exceeding 90%. When such an event is detected, the system performs high-precision SOH estimation to obtain the estimated SOH value of each individual cell. local And assign it a high confidence level C local For time window T window For batteries that have not completed an effective full charge-discharge cycle, the system will reach a state where t≥T. window A lightweight estimation algorithm based on segment data is automatically triggered under certain conditions. This algorithm extracts effective features from arbitrary charge / discharge segments and estimates the SOH through pre-established mapping relationships. local Its confidence level C local The calculation takes into account the duration t of the data segment. duration Signal-to-noise ratio (SNR) and operating condition matching degree (M) profile Factors such as C are specifically represented as C. local =f(t) duration ,SNR,M profile ), where f(·) is a weighting function that comprehensively considers various influencing factors.

[0110] Step 1.2 Parallel computation of multidimensional features: At the same time as each SOH update, the system computes four core intrinsic health features in parallel. Capacity-resistance composite factor F CR Through calculation The factor is obtained, where ΔQ is the segment capacity, I is the average current, and R is the DC internal resistance. This factor effectively eliminates the influence of current magnitude. Relaxation voltage recovery rate η Vr Calculated by monitoring the voltage recovery process after the load is discharged, defined as... Where V start V end These are the starting and ending voltages of the resting period, V. relax This represents the theoretical equilibrium voltage. The slope k of the constant current charging voltage curve. CC Constant current charging data within a specific voltage range are selected, and linear regression is used to calculate... Incremental capacity characteristic peak shift The dV / dQ curve is obtained by numerically differentiating the charging curve, the position of the characteristic peak is identified, and its offset from the reference position is calculated.

[0111] Step 1.3 Abnormal Accumulation Memory and Triggering Mechanism: The system maintains an abnormal accumulation memory value A for each battery cell. i The initial value is 0. When an abnormal feature is detected, press A. i (t+1)=λA i (t)+Δ i (t) Update the memory value, where the forgetting factor λ∈(0,1) controls the decay rate of historical influence, Δ i (t) represents the anomaly intensity of the current period, calculated as a weighted sum of the deviations of each feature: Dynamic threshold θ i Based on the historical performance of the monomer, adaptive adjustment is made when A i More than θ i The diagnostic process is triggered immediately upon A's arrival. i When the number of consecutive cycles continues to increase, the system automatically raises the diagnostic priority of that individual to ensure that potential anomalies are dealt with in a timely manner.

[0112] Step 2: Construct a local spatial diagnostic unit. For suspicious cells that trigger identification, a dynamic neighbor network is constructed based on the battery cluster electrical connection topology and thermal management partitioning. A dual comparison of trajectory consistency and feature synergy is performed within the local state space to accurately identify the root cause of anomalies and generate a health status value (SOH) based on the local state space. spatial .

[0113] Step 2 includes the following specific steps:

[0114] Step 2.1 Dynamic Neighbor Network Construction, targeting the suspicious individual v marked in Step 1 t The system is based on the electrical connection topology diagram G of the battery cluster. e (V,E) and thermal management zoning model M t Construct a dynamic neighbor network. Electrical correlation S e (v t ,v i The calculation of ) is based on the shortest path distance in the topological graph, defined as Where d e (v t ,v i ) represents the shortest path hop count in the electrical topology, and ε is the smoothing factor. Thermal management correlation S t (v t ,v i Based on the thermal management zone distance calculation, considering the correlation between cooling duct distance and temperature field, it is defined as... Where d t (v t ,v i ) represents the physical spatial distance, σ t This is a parameter for the heat-affected zone. When S is simultaneously satisfied... e (v t ,v i )>θ e and S t (v t ,v i )>θ t At that time, the monomer v i The dynamic neighbor network N of the target unit t This ensures that neighbors are highly relevant in both electrical and thermal management dimensions.

[0115] Step 2.2 Dual Comparison Analysis: Within the constructed local health state space, the system performs trajectory consistency comparison and feature synergy comparison. Trajectory consistency comparison is achieved by calculating the correlation coefficient of the target individual's historical SOH trajectory with its neighboring groups. Specifically, the calculation is as follows: Covariance calculation is based on the nearest N... hist The SOH sequence for each historical period. Feature cooperability alignment assesses the difference in feature vectors between the target individual and its neighboring groups at the current time using Mahalanobis distance, calculated as follows:

[0116]

[0117] The eigenvector F contains multidimensional features such as the capacitance-resistance composite factor and the relaxation voltage recovery rate, μ neighbors and Σ neighbors These are the mean vector and covariance matrix of the eigenvectors of the neighboring group, respectively.

[0118] Step 2.3 Root cause discrimination and SOH estimation based on local state space: Based on the double comparison results, the system calculates the root cause discrimination score P. root =σ(α·(1-ρ) traj )+β·D m ), where σ(·) is the Sigmoid function, and the weight coefficients α and β are dynamically adjusted according to the importance of the features. Simultaneously, SOH estimates are generated based on the states of healthy individuals in the neighboring population. weight w j Taking into account factors such as electrical correlation, thermal management correlation, and historical consistency, the confidence level C based on the local state-space estimate is... spatial Calculated as C spatial =g(σ neighbors ,N,ρ traj ), where σ neighborsLet S be the standard deviation of the SOH values ​​of the neighboring population, and N be the number of effective neighbors. The function g(·) comprehensively evaluates the quality and consistency of the reference population. This mechanism achieves fine-grained discrimination of root causes through continuous quantification and provides a reliable spatial reference benchmark for downstream fusion.

[0119] Step 3: Construct game weight fusion units based on individual unit estimates SOH local Compared with the local state space-based estimate SOH spatial By combining quantifiable confidence indicators, a game theory model is used for adaptive weight fusion to output the optimal final estimate, SOH. final .

[0120] Step 3 includes the following specific steps:

[0121] Step 3.1 Based on game theory weight decision-making, the system receives the individual's own SOH estimate from the upstream. local and its confidence level C local SOH based on local state space SOH estimation spatial and its confidence level C spatial and root cause discrimination score P root Based on these inputs, a lightweight game theory model is constructed to calculate the optimal fusion weights.

[0122] In the game theory model, SOH local and SOH spatial Considering two players in the game, their utility functions are defined as follows:

[0123] U local =C local ·P root

[0124] U spatial =C spatial ·(1-P root )

[0125] Among them, U local This reflects the degree of confidence in one's own estimates when there is suspicion of genuine degradation. spatial This reflects the degree of dependence on spatial consensus when the error is suspected to be an estimation error. The optimal fusion weights are calculated using the Softmax function:

[0126]

[0127] In the formula, T is the temperature parameter, which controls the degree of certainty in the decision. When T is small, the system tends to choose the estimate with higher utility; when T is large, the system adopts a more balanced fusion strategy between the two.

[0128] Step 3.2, the final SOH generation and parameter optimization, is implemented as follows: The final SOH estimate is generated based on the game weight α:

[0129] SOH final =α·SOH local +(1-α)·SOH spatial

[0130] The estimated value and its corresponding fusion weight α are updated to the historical database as the authoritative state record for the current period.

[0131] When the system detects a high-precision full-charge-discharge cycle, it initiates a parameter optimization mechanism. The ampere-hour integral result (SOH) of this cycle is used. high_precision As a baseline, the fusion results from the previous K periods are reviewed, and the estimation bias Δ = SOH is calculated. high_precision -SOH final Based on statistical analysis of the bias, the gradient descent method is used to optimize the temperature parameter T.

[0132]

[0133] Where J is the loss function, which comprehensively considers both short-term estimation accuracy and long-term stability, and η is the learning rate. Through this feedback mechanism based on neighboring high-precision cycles, the system can adaptively adjust the fusion strategy, achieving continuous parameter improvement while ensuring estimation reliability.

[0134] Step 4: Construct a tiered early warning decision-making unit, integrating root cause qualitative conclusions and SOH... final It integrates multi-source information such as weights and generates tiered alarm signals and operation and maintenance suggestions based on a preset tiered early warning rule base.

[0135] Step 4 includes the following specific steps:

[0136] Step 4.1, the fusion and integration of multi-source decision information, is implemented as follows: The system receives multi-source decision information from the upstream processing unit, including the root cause discrimination score P. root Final SOH estimate final SOH decay kinetics (including decay rate) and decaying acceleration (and key parameters such as fusion weight α).

[0137] To establish a unified information fusion framework, the first step is to standardize the multi-source data. Root cause discrimination score P. root The fusion weight α is directly used as the normalized input in the [0,1] interval. The SOH decay kinetics characteristics are processed using time normalization, where the decay rate is calculated as follows: Δt daysTo determine the actual number of days between two valid estimates, ensuring comparability of data from different update cycles, the decay acceleration is calculated based on a sliding window. Where τ is a fixed evaluation period. The final SOH estimate is converted into a relative offset. By aligning timestamps and standardizing data types, we ensure that all parameters are consistent and coordinated within a unified decision-making framework.

[0138] Step 4.2 Intelligent hierarchical early warning decision generation: Based on a preset multi-level early warning rule base, a comprehensive risk assessment function R = w1·P is constructed. root +w2·|SOH dev |+w3·|r norm |+w4·max(0,a norm The weighting coefficients w1 to w4 are dynamically adjusted according to the importance of each parameter.

[0139] The system generates three levels of early warning based on the risk assessment result R: When 0.1≤R<0.3, an observation-level early warning is triggered, suggesting "increasing the monitoring frequency to twice the normal value and recording the trend of operating parameter changes"; when 0.3≤R<0.7, an alert-level early warning is triggered, suggesting "prioritizing inspection during the next planned maintenance and assessing balanced maintenance needs"; when R≥0.7, an emergency-level early warning is triggered, suggesting "immediately conducting on-site inspection and considering reducing operating power or temporarily shutting down operation".

[0140] Early warning decisions are output to the operation and maintenance management system via a standardized RESTful API interface. The transmission format uses a unified JSON data packet, which includes fields such as warning level, target unit identifier, risk assessment value, timestamp, and specific operation and maintenance recommendations. The system also establishes a feedback mechanism to record the warning response and subsequent verification results, forming a complete decision-making closed loop from status monitoring, risk assessment, warning generation to operation and maintenance execution, enabling accurate identification and intelligent operation and maintenance management of abnormal states of energy storage batteries.

[0141] The present invention also provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor. The transceiver, the memory, and the processor are connected via the bus. The computer program, when executed by the processor, implements the steps in the aforementioned method for identifying and warning of abnormal cell failure in an energy storage battery. Compared with the prior art, the beneficial effects of the electronic device provided by the present invention are the same as those of the method for identifying and warning of abnormal cell failure in an energy storage battery described above, and will not be elaborated upon here.

[0142] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps in the above-described method for identifying and warning of abnormally deteriorated cells in an energy storage battery. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present invention are the same as the beneficial effects of the above-described method for identifying and warning of abnormally deteriorated cells in an energy storage battery, and will not be elaborated here.

[0143] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying and warning of abnormal cell deterioration in energy storage batteries, characterized in that, include: Step 1: Estimate the health status of all battery cells in the energy storage power station to obtain individual cell estimates, and assign corresponding confidence levels to the individual cell estimates; Step 2: Calculate the intrinsic health characteristics of each battery and mark batteries whose intrinsic health characteristics are not within the set range as suspicious cells; Step 3: Construct a neighbor group of suspected individual cells based on the electrical connection topology of the battery cluster and the thermal management zoning model; Step 4: Compare and analyze the batteries in the neighboring group to calculate the root cause discrimination score, the estimated value of the local state space and its corresponding confidence level; Step 5: Based on the confidence scores of the individual unit estimates, the confidence scores of the local state space estimates, and the root cause discrimination scores, the individual unit estimates and the local state space estimates are adaptively weighted and fused to obtain the final estimate. Step 6: Complete the graded early warning of abnormal states of energy storage batteries based on the final estimated values.

2. The method for identifying and warning of abnormal cell deterioration in an energy storage battery according to claim 1, characterized in that, In step 1, when the battery completes an effective full charge-discharge cycle, the estimated single-cell SOH value of the battery is obtained by the ampere-hour integration method. local And assign a corresponding confidence level C. local ; where an effective full-charge-discharge cycle is: a complete process from a fully charged state to the cutoff voltage, with a discharge depth exceeding 90%; for a time window T window For batteries that have not completed a valid full charge-discharge cycle, when the cycle time t ≥ T window At that time, a lightweight estimation algorithm based on fragment data obtains the estimated single-cell SOH value of the battery. local And assign a corresponding confidence level C local .

3. The method for identifying and warning of abnormal cell deterioration in an energy storage battery according to claim 2, characterized in that, In step 2, the intrinsic health characteristics include: capacity-resistance recombination factor, relaxation voltage recovery rate, constant current charging voltage curve slope, and incremental capacity characteristic peak shift; wherein, the capacity-resistance recombination factor F CR The calculation formula is: Where ΔQ is the segment capacity, I is the average current, and R is the DC internal resistance; Relaxation voltage recovery rate η Vr The calculation formula is: Among them, V start V end These are the starting and ending voltages of the resting period, V. relax This is the theoretical equilibrium voltage.

4. The method for identifying and warning of abnormal cell deterioration in an energy storage battery according to claim 3, characterized in that, Step 3: Constructing a neighbor group of suspected individual cells based on the electrical connection topology of the battery cluster and the thermal management zoning model, including: Step 3.1: Calculate the electrical correlation degree based on the electrical connection topology diagram of the battery cluster; wherein, the formula for calculating the electrical correlation degree is: Where, d e (v t ,v i ) represents the shortest path hop count in the electrical topology, v t Indicates a suspicious monomer, v i Represents the target monomer, ε is the smoothing factor, and S e (v t ,v i ) represents electrical correlation degree; Step 3.2: Calculate the thermal management correlation degree based on the thermal management zoning model's distance between thermal management zones; wherein, the formula for calculating the thermal management correlation degree is: Where, d t (v t ,v i ) represents the physical spatial distance, σ t This refers to the parameters related to the heat-affected zone. Step 3.3: When S is satisfied simultaneously e (v t ,v i )>θ e and S t (v t ,v i )>θ t At that time, the monomer v i Include the neighboring groups of the suspected individual.

5. The method for identifying and warning of abnormal cell deterioration in an energy storage battery according to claim 4, characterized in that, Step 4: Comparative analysis is performed on the batteries in the neighboring group to calculate the root cause discrimination score, the estimated value of the local state space, and the corresponding confidence level, including: Step 4.1: Calculate the correlation coefficient of the historical SOH trajectories of the target individual and its neighboring populations to achieve trajectory consistency comparison; the trajectory consistency comparison results are as follows: Where, ρ traj This indicates the trajectory consistency comparison result, SOH target Represents the sequence of health status values ​​of the target monomer, SOH neighbors σ represents the sequence of health status values ​​of the neighbors surrounding the target unit. target σ represents the standard deviation of the target monomer SOH sequence. neighbors This represents the standard deviation of the SOH sequences in the neighboring population; Step 4.2: The feature coherence comparison is achieved by evaluating the difference in feature vectors between the target individual and its neighboring groups at the current time using Mahalanobis distance; the feature coherence comparison result is as follows: Among them, D m This represents the feature coherence comparison result, where Ftarget represents the feature vector of the target individual at the current time, μneighbors is the mean vector of the feature vectors of the neighboring group, and Σ neighbors Let be the covariance matrix of the eigenvectors of the neighboring group; Step 4.3: Calculate the root cause discrimination score based on the comparison results; the formula for calculating the root cause discrimination score is as follows: P root =σ(α·(1-ρ traj )+β·D m ) Among them, P root σ(·) represents the root cause discrimination score, σ(·) is the Si gmoi d function, and α and β are weight coefficients; Step 4.4: Calculate the estimated value of the local state space based on the state of healthy individuals in the neighboring group, and assign the corresponding confidence level; wherein, the estimated value of the local state space is: Among them, SOH spatial w represents the estimate of the local state space. j Let SOH represent the j-th weight. j This represents the state of the j-th healthy individual in the neighboring group.

6. The method for identifying and warning of abnormal cell deterioration in an energy storage battery according to claim 5, characterized in that, Step 5: Based on the confidence scores of the individual unit estimates, the confidence scores of the local state space estimates, and the root cause discrimination scores, the individual unit estimates and the local state space estimates are adaptively weighted and fused to obtain the final estimate, including: Step 5.1: In the game model, the individual estimates and local state space estimates are treated as two game participants to construct utility functions, and the fusion weight is calculated; the formula for calculating the fusion weight is: U local =C local ·P root U spatial =C spatial ·(1-P root ) Among them, U local U represents the degree of confidence in one's own estimate when there is suspicion of true degradation. spatial C represents the degree of dependence on spatial consensus when the error is suspected to be an estimation error. local C represents the confidence level of the individual estimates. spatial P represents the confidence level corresponding to the local state-space estimate. root α represents the root cause discrimination score, α represents the fusion weight, and T represents the temperature parameter. Step 5.2: Calculate the final estimate based on the fusion weights; the formula for calculating the final estimate is: SOH final α·SOH local +(1-α)·SOH spatial Among them, SOH final This represents the final estimated value, SOH. local This represents the estimated value of the monomer, SOH. spatial This represents the local state space estimate.

7. The method for identifying and warning of abnormal cell deterioration in an energy storage battery according to claim 6, characterized in that, Step 6: Based on the final estimated value, complete the graded early warning of abnormal states of the energy storage battery, including: Step 6.1: Construct a comprehensive risk assessment function using the final estimated value, root cause discrimination score, decay rate, and decay acceleration; whereby the comprehensive risk assessment function is: R=w1·P root +w2·|SOH dev |+w3·|r norm |+w4·max(0,a norm ) Where R represents the risk assessment result, w1 represents the first weighting coefficient, w2 represents the second weighting coefficient, w3 represents the third weighting coefficient, w4 represents the fourth weighting coefficient, and SOH... dev Indicates the relative offset. SOH threshold This represents the preset health status safety threshold, r norm a represents the normalized decay rate. norm This represents the normalized decay acceleration; Step 6.2: Generate a three-level warning using the risk assessment results. When 0.1≤R<0.3, trigger an observation-level warning; when 0.3≤R<0.7, trigger an alert-level warning; when R≥0.7, trigger an emergency-level warning.

8. A system for identifying and warning of abnormal cell deterioration in energy storage batteries, characterized in that, include: The single-cell health status module is used to estimate the health status of all battery cells in the energy storage power station to obtain a single-cell estimate, and to assign a corresponding confidence level to the single-cell estimate; The suspicious cell identification module is used to calculate the intrinsic health characteristics of each battery and mark batteries whose intrinsic health characteristics are not within the set range as suspicious cells; The neighbor group building module is used to build neighbor groups of suspected individual cells based on the electrical connection topology graph and thermal management zoning model of battery clusters; The comparison analysis module is used to perform comparison analysis on batteries in the neighbor group to calculate the root cause discrimination score, the estimated value of the local state space and its corresponding confidence level; The fusion module is used to adaptively weight and fuse the individual estimates and the local state space estimates based on the confidence scores of the individual estimates, the confidence scores of the local state space estimates, and the root cause discrimination scores to obtain the final estimate. The early warning module is used to provide graded early warnings of abnormal states of energy storage batteries based on the final estimated values.

9. An electronic device comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus, characterized in that, When the computer program is executed by the processor, it implements the steps in the method for identifying and warning of abnormal battery cell failure as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps in the method for identifying and warning of abnormal cell failure of an energy storage battery as described in any one of claims 1-7.

Images