Early warning strategy, program and equipment for thermal runaway battery cell in ship energy storage equipment and storage medium

Abstract

The invention discloses an early warning strategy, program and equipment for a thermal runaway battery cell in ship energy storage equipment, and a storage medium, and the method employs multi-scale step mode decomposition and Spearman rank correlation analysis to decompose and reconstruct the voltage of the battery cell, overcomes the problems of mode aliasing and boundary effect in a conventional signal decomposition method, and achieves the early warning of the thermal runaway battery cell in the ship energy storage equipment. Limitation of reconstruction according to modal frequency in the prior art is avoided; an improved Sigmoid nonlinear mapping method is provided, so that tiny voltage difference between high-risk and low-risk battery cells can be amplified in a self-adaptive manner; abnormal battery cells are dynamically screened through sparse representation in combination with an adaptive threshold value, the problem of insufficient adaptability of a fixed threshold value method in a variable operation environment is overcome, and the phenomena of false alarm and missing alarm can be remarkably reduced; the multi-dimensional characteristics such as the voltage, the temperature and the change rate of the battery cell are comprehensively considered, the relevance between the battery cell characteristics is quantified based on a mutual information method, and the potential risk of the battery cell can be more comprehensively revealed. The invention further designs a set of joint evaluation method integrating the deviation score and the mutual information value, and the reliability of high-risk battery cell judgment is further improved through multi-level screening.

Description

Technical Field

[0001] This invention belongs to the field of early warning technology for high-risk thermal runaway cells in marine lithium-ion battery systems operating in complex marine environments. Specifically, it relates to an early warning strategy, program, device, and storage medium for thermal runaway cells in marine energy storage equipment. Background Technology

[0002] In recent years, with the continuous advancement of the "dual carbon" target, the application of electric ships in the marine transportation sector has grown rapidly. Lithium-ion batteries, due to their high energy density, long lifespan, and low self-discharge rate, have become the preferred energy storage equipment for electric ships. However, the complexity of the marine environment (such as high salt spray corrosion and high humidity) intensifies the mechanical, thermal, and electrical stresses within the battery cells, accelerating the degradation process and posing a severe challenge to the operational reliability of lithium-ion battery systems. Simultaneously, ships typically operate on long-cycle missions, making frequent maintenance and replacement of energy storage equipment impossible, and potential hazards can easily trigger thermal runaway after long-term accumulation. Compared to land-based energy storage equipment, ships, due to their unique operating environment and limited personnel escape conditions, require higher timeliness and accuracy in early warning of battery cell thermal runaway.

[0003] Traditional thermal runaway early warning methods are mostly based on signals such as smoke, temperature, and escaped gases. However, these signals usually appear in the critical stage of thermal runaway, making it difficult to provide sufficient time for early risk avoidance. Although some methods for early warning of battery cells have been studied, they are limited by model accuracy and datasets, and previous model-based and data-driven methods face problems such as inaccurate model parameter updates and small sample sizes. Furthermore, the real-time performance and warning timescale of these two methods are not suitable for marine energy storage equipment. This invention adopts a signal analysis-based method, which can balance the timescale and accuracy of early warning.

[0004] Signal analysis-based methods typically rely on time-domain and frequency-domain characteristic parameters to provide early warnings by identifying characteristic parameters that do not conform to a specific distribution. This invention fully considers voltage, temperature, and their corresponding rates of change, employing a multi-dimensional feature joint evaluation approach to achieve early warning. It can quickly identify high-risk cells for thermal runaway based on voltage and temperature data in discharge data segments.

[0005] A search revealed no similar implementations to this invention in current literature and patent databases. Related patents all focus on thermal runaway early warning for battery cells in land-based energy storage equipment, failing to adequately consider the novel characteristics and challenges of thermal runaway early warning for battery cells in marine energy storage equipment. CN202411376911, "A Method for Thermal Runaway Early Warning of Power Batteries under Multiple Operating Conditions Based on a Vehicle-Cloud System," relies on cloud computing, receiving features calculated by the receiving device, and using the feature set to calculate the health of each individual battery cell to achieve early warning. This invention considers a relatively short warning timescale, making it difficult to achieve early warning for high-risk thermal runaway cells in the early stages. CN202411584891, "A Method, System, Electronic Device, and Storage Medium for Fault Diagnosis of Energy Storage Cells," is based on a clustering algorithm, constructing a clustering model for operational data to obtain original normal and abnormal clusters. Then, early warning is achieved through cluster division of real-time operational data and the accumulation of abnormal judgment counts. It is evident that existing patents differ from this invention in their objects, methods, and means. Summary of the Invention

[0006] The purpose of this invention is to provide an early warning strategy, program, device, and storage medium for thermal runaway cells in marine energy storage equipment.

[0007] An early warning strategy for thermal runaway cells in marine energy storage equipment includes the following steps:

[0008] The system acquires operational data of the ship's energy storage equipment over a period of time and extracts data segments from the ship's energy storage equipment during the discharge phase. The data segments during the discharge phase include the voltage and temperature values ​​of each cell at each sampling time.

[0009] By using multi-scale step mode decomposition, the voltage value of each cell at each sampling time is decomposed into multiple sets of mode functions. The Spearman rank correlation coefficient between each mode function and the original voltage value is calculated. The mode function with the smallest corresponding Spearman rank correlation coefficient value is removed. The voltage of the remaining mode functions is reconstructed to obtain the denoised voltage.

[0010] Based on the nonlinear voltage mapping method of the improved Sigmoid function, the denoised voltage of each cell is adaptively mapped to obtain the mapped voltage.

[0011] Based on sparse representation theory, the reconstruction error of the mapped voltage of each cell is calculated. The deviation between the reconstruction error of the mapped voltage of each cell and the mean of the reconstruction errors of the mapped voltage of all cells is used as the deviation score of the cell. An adaptive deviation threshold is constructed, and cells with a deviation score greater than the adaptive deviation threshold are marked as cells with abnormal deviation.

[0012] Construct the multidimensional feature vector of each cell and the multidimensional mean feature vector of all cells, and calculate the mutual information value of the multidimensional feature vector of each cell and the multidimensional mean feature vector of all cells based on the mutual information method; construct an adaptive mutual information threshold, and mark the cells whose corresponding mutual information value is less than the adaptive mutual information threshold as cells with abnormal mutual information.

[0013] By combining the detection results of deviation anomalies and mutual information anomalies, an early warning can be provided for thermal runaway cells in ship energy storage equipment through a joint evaluation strategy.

[0014] Furthermore, the ship energy storage equipment includes K groups of battery cells, and the data segment for each discharge stage includes the voltage value V of each battery cell at each sampling time. j,k With temperature value T j,k V j,k T represents the voltage value of the k-th cell at the j-th sampling time. j,k This represents the temperature value of the k-th cell at the j-th sampling time; k = 1, 2, ..., K; j = 1, 2, ..., J; J is the total number of sampling times, and the time interval between adjacent sampling times is ΔT;

[0015] The method utilizes multi-scale step mode decomposition to determine the voltage value V of each cell at each sampling time. j,k Decomposed into G sets of modal functions u g,k (j); Calculate the modal functions u of each mode. g,k (j) and the original voltage value V j,k The Spearman rank correlation coefficient ρ between them g,k :

[0016]

[0017] Where R(x) represents the rank of x;

[0018] The corresponding Spearman rank correlation coefficient ρ g,k The mode function with the smallest value u g,k (j) Remove the remaining G-1 group of mode functions and reconstruct the voltage to obtain the denoised voltage.

[0019] Furthermore, the nonlinear voltage mapping method based on the improved Sigmoid function selects the average voltage of all cells after denoising, after removing the maximum and minimum values ​​at the same sampling time. As a reference voltage, the voltage of each cell after noise reduction Perform adaptive mapping to obtain the mapped voltage z j,k :

[0020]

[0021] Where α is the amplification factor, used to adjust the sensitivity of the mapping; γ is the dynamic power, used to control the dynamic changes in voltage difference.

[0022] Furthermore, based on sparse representation theory, the reconstruction error e of the mapped voltage of each cell is calculated. k Specifically:

[0023] Step 1: Initialize the iteration count t = 1; construct the mapped voltage vector Z k =[z 1,k ,z 2,k ,...,z J,k ] T Principal component analysis was used to extract the mapped voltage vector Z. k The first m principal component voltages form the initial basis vector set D. k (1);

[0024] Step 2: Using the Lasso regression algorithm, based on the objective function f, obtain the sparse coefficient vector β that minimizes the corresponding objective function value. k (t);

[0025] f = ||z k -D k (t)·β k (t)+λ·β k (t)||

[0026] β k (t) = argmin(f)

[0027] Where λ is the regularization parameter;

[0028] Step 3: Calculate the reconstruction error vector E k (t);

[0029] E k (t)=z k -D k (t)·β k (t)

[0030] Step 4: Utilize the reconstructed error vector E k (t) and the sparse coefficient vector β k (t) Update the basis vector set D k (t+1);

[0031] D k (t+1)=D k (t)+η·E k (t)·(β k (t)) T

[0032] Where η is the update step size;

[0033] Step 5: If ||D k (t+1)-D k If (t)||≤ξ, and ξ is the convergence threshold, then stop the iteration and proceed to step 6; otherwise, let t=t+1 and return to step 2.

[0034] Step 6: Based on the basis vector set D k (t+1) Perform steps 2 and 3 once to obtain the reconstruction error vector E. k (t+1); Let D k =D k (t+1), e k,j (t+1) represents the reconstruction error vector E. k The j-th element in (t+1).

[0035] Furthermore, the reconstructing error e of each cell k The average of the reconfiguration errors of all cells The deviation is used as the deviation score (DS) of the battery cell. k Calculate the adaptive deviation threshold V th If the deviation score of the battery cell is DS k Greater than the adaptive deviation threshold V th If so, the cell is marked as a cell with abnormal deviation.

[0036]

[0037] V th =θ·μ DS +(1-θ)·(ζ·max(DS k )+(1-ζ)·(μ DS +δ·σ DS ))

[0038] Where θ,ζ∈(0,1) are weighting coefficients; μ DS Deviation score DS for all cells k The mean; σ DS Deviation score DS for all cells k The standard deviation is δ; δ is the amplification factor.

[0039] Furthermore, the construction of the multidimensional feature vector C for each battery cell... k :

[0040] C k =[c 2,k ,c 3,k ,...,c J,k ]

[0041] ci,k =[z i,k ,T i,k ,Δz i,k ,ΔT i,k ]

[0042] Where i = 2, 3, ..., J;

[0043] For each dimension of data, construct a data matrix X. d d = 1, 2, 3, 4, data matrix X d element x in d,i,k for:

[0044]

[0045] Construct the multidimensional mean feature vector of all battery cells

[0046]

[0047] in,

[0048] The multidimensional feature vector C of each battery cell is calculated based on the mutual information method. k Multidimensional mean eigenvector of all battery cells Mutual information value I k Calculate the mutual information threshold T MIV , will correspond to I k Less than T MIV The battery cells were marked as having abnormal mutual information.

[0049]

[0050] Where φ is the adjustment factor;

[0051] Furthermore, the joint evaluation strategy is as follows:

[0052] If a battery cell is marked as having abnormal deviation in three consecutive discharge data segments, then the battery cell is determined to be a high-risk battery cell for thermal runaway.

[0053] If a battery cell is marked as having abnormal deviation in two consecutive discharge data segments, then the battery cell is determined to be a potentially high-risk battery cell.

[0054] If a battery cell has been marked as a deviation abnormality cell twice in the historical data segment, and the deviation score is abnormal again when the next discharge data segment is detected, then the battery cell is determined to be a potentially high-risk battery cell.

[0055] If a cell has been identified as a potentially high-risk cell, its mutual information value in the corresponding data segment is further evaluated; if the mutual information value is detected and the cell is marked as having abnormal mutual information, then the cell is determined to be a high-risk cell for thermal runaway.

[0056] A computer device / equipment / system includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned early warning strategy for thermal runaway cells in marine energy storage equipment.

[0057] A computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the steps of the early warning strategy for thermal runaway cells in the aforementioned ship energy storage equipment.

[0058] A computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the aforementioned early warning strategy for thermal runaway cells in marine energy storage equipment.

[0059] The beneficial effects of this invention are as follows:

[0060] This invention employs multi-scale step mode decomposition and Spearman rank correlation analysis to decompose and reconstruct cell voltage, overcoming the problems of mode aliasing and boundary effects in traditional signal decomposition methods, and avoiding the limitations of traditional reconstruction based on mode frequency. The decomposition method used has higher decomposition efficiency when dealing with non-stationary step voltages, providing high-quality input for subsequent analysis.

[0061] The improved Sigmoid nonlinear mapping method proposed in this invention can adaptively amplify minute voltage differences between high- and low-risk battery cells. Compared to linear mapping methods, this invention is more effective in capturing minute voltage differences.

[0062] This invention uses sparse representation combined with adaptive threshold to dynamically screen abnormal battery cells, overcoming the problem of insufficient adaptability of fixed threshold methods in variable operating environments. It can flexibly adjust the screening criteria according to actual operating conditions, which can significantly reduce false alarms and missed alarms.

[0063] This invention comprehensively considers multi-dimensional characteristics of battery cells, such as voltage, temperature, and their rate of change, and quantifies the correlation between these characteristics based on mutual information methods, thus revealing potential risks more comprehensively. Compared to existing methods that rely on single features, using multi-dimensional features for evaluation can effectively improve the comprehensiveness and accuracy of early warning.

[0064] This invention designs a joint evaluation method that integrates deviation score and mutual information value, and further improves the reliability of high-risk cell identification through multi-level screening. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the voltage decomposition and reconstruction process in this invention.

[0066] Figure 2 This is a schematic diagram of the nonlinear voltage mapping and initial screening of cell anomalies in this invention.

[0067] Figure 3 This is a schematic diagram illustrating the process of constructing multidimensional features and calculating mutual information in this invention.

[0068] Figure 4 This is a schematic diagram of the joint assessment and high-risk judgment process in this invention. Detailed Implementation

[0069] The present invention will now be further described with reference to the accompanying drawings.

[0070] During operation, shipboard energy storage equipment must withstand complex dynamic conditions and harsh marine environments over extended periods. These environmental factors accelerate cell performance degradation and increase the likelihood of thermal runaway. Therefore, the application of current cell condition monitoring and early warning technologies in complex marine environments remains challenging, and improving the foresight and accuracy of early warning strategies has become an urgent issue to address.

[0071] The shortcomings of existing technologies are mainly reflected in the following aspects: Current thermal runaway early warning methods are mainly based on signals such as smoke, temperature, and escaped gases. These features usually appear when the cell condition has deteriorated to a critical stage, which cannot provide sufficient early warning time to avoid risks, and the early warning time scale is relatively short. Although using voltage signals for early warning can obtain early warning effects on a longer time scale, traditional signal decomposition methods are difficult to accurately filter out irrelevant interference when facing non-stationary step voltages with irrelevant interference in marine energy storage equipment, resulting in a decrease in feature extraction quality and potentially affecting the early warning effect. Existing methods often ignore the similarity between the voltages of high-risk and low-risk cells in the early stage, making it difficult to effectively quantify these subtle differences. Existing methods usually rely on only a single feature such as voltage or temperature for early warning, lacking a multi-dimensional comprehensive assessment of the cell condition, resulting in a high false alarm and false negative rate. Traditional methods rely on fixed thresholds to screen high-risk cells, but due to the complex operating environment faced by marine energy storage equipment, fixed thresholds are difficult to adapt to the changing operating environment, which may lead to frequent false alarms and false negatives.

[0072] This invention aims to fully consider the technical requirements of ship energy storage equipment in complex marine environments. Through innovative signal processing and multi-dimensional feature analysis methods, it establishes an efficient early warning strategy for thermal runaway cells. This method focuses on extracting and amplifying minute voltage differences between cells during operation, combined with dynamically adaptive screening rules and a comprehensive evaluation mechanism, to achieve accurate identification and early warning of thermal runaway risks. This solution significantly improves the safety and reliability of battery systems, providing strong technical support for the long-term stable operation of electric ships.

[0073] This scheme comprises four key parts: voltage decomposition and reconstruction, nonlinear voltage mapping and initial screening of cell anomalies, construction of multidimensional features and mutual information calculation, and joint evaluation and high-risk assessment. The steps of the scheme are as follows:

[0074] A. The data segments are divided according to the operating status, and the data segments of the discharge stage are extracted. The original voltage is decomposed using multi-scale step mode decomposition, and the decomposition parameters are adjusted through optimization algorithms to extract the mode functions. Spearman correlation is used to optimize and reconstruct the modes, filter out irrelevant interference, and finally obtain the denoised voltage as a high-quality input for subsequent analysis.

[0075] B. An adaptive mapping method based on an improved Sigmoid function is used to map the denoised voltage. By compressing the relative normal voltage and amplifying the potential abnormal voltage, the subtle voltage differences between high- and low-risk cells are effectively captured. A sparse representation is used to construct a cell operating state characterization model. The degree of deviation of the cell from the normal state is quantified by sparse linear combination, and an adaptive threshold is used to achieve preliminary screening of abnormal cells.

[0076] C. Construct a multidimensional feature vector of the battery cell (including voltage, temperature and their rate of change), and further evaluate the abnormal behavior of the battery cell during operation based on the mutual information method, so as to provide a basis for joint evaluation.

[0077] D. Based on the initial screening results of anomalies, and combined with the mutual information values ​​of multidimensional features, design joint evaluation rules to comprehensively judge the risk status of the battery cell.

[0078] The specific implementation method includes the following steps:

[0079] Step 1: Acquire operational data of the ship's energy storage equipment over a period of time, and extract data segments during the discharge phase of the equipment. Each data segment during the discharge phase includes the voltage value V of each cell at each sampling time. j,k With temperature value T j,k ;

[0080] The ship energy storage equipment includes K sets of battery cells, V j,kT represents the voltage value of the k-th cell at the j-th sampling time. j,k This represents the temperature value of the k-th cell at the j-th sampling time; k = 1, 2, ..., K; j = 1, 2, ..., J; J is the total number of sampling times, and the time interval between adjacent sampling times is ΔT;

[0081] For each data segment of the discharge stage, perform the following data processing steps:

[0082] Step 1: Decompose and reconstruct the voltage:

[0083] The voltage value V of each cell at each sampling time is obtained by using multi-scale step mode decomposition. j,k Decomposed into G sets of modal functions u g,k (j);

[0084] To avoid unstable decomposition quality caused by manually setting parameters, the multi-scale step mode decomposition adopts a genetic algorithm, using the total variance explained rate as the fitness function, to optimize the hyperparameters required for multi-scale step mode decomposition, including bandwidth penalty, jump weight, penalty scaling factor, minimum jump amplitude threshold, and number of modes G.

[0085] To avoid information distortion that may be caused by frequency-based mode selection, this invention uses the Spearman rank correlation coefficient to evaluate the correlation between each mode obtained from MJMD decomposition and the original voltage, thus determining the interference components that need to be removed. It should be noted that existing research shows that the Spearman correlation between each mode obtained using similar methods and the original signal exhibits a monotonically decreasing characteristic. The mode with the lowest correlation is usually noise or irrelevant interference, while other modes contain physically meaningful information; further removal would lead to the loss of effective signal information. Therefore, only the mode with the lowest correlation is removed, and the remaining modes are superimposed to obtain the denoised voltage. The correlation calculation method is as follows:

[0086] Calculate the modal functions u g,k (j) and the original voltage value V j,k The Spearman rank correlation coefficient ρ between them g,k :

[0087]

[0088] Where R(x) represents the rank of x;

[0089] The corresponding Spearman rank correlation coefficient ρ g,k The mode function with the smallest value u g,k (j) Remove the remaining G-1 group of mode functions and reconstruct the voltage to obtain the denoised voltage.

[0090] Step 2: Nonlinear mapping of voltage and initial screening of cell anomalies:

[0091] Based on a nonlinear voltage mapping method using an improved Sigmoid function, the average voltage of all cells after denoising, after removing the maximum and minimum values ​​at the same sampling time, is selected. As a reference voltage, the voltage of each cell after noise reduction Perform adaptive mapping to obtain the mapped voltage z j,k ;

[0092]

[0093] Where α is the amplification factor, used to adjust the sensitivity of the mapping; γ is the dynamic power, used to control the dynamic changes in voltage difference;

[0094] Normally, cell malfunctions are a low-probability event in a system, and the voltage of the vast majority of cells is normal. Therefore, to effectively represent the overall voltage trend and avoid interference from extreme values, the average voltage of all cells at the same sampling time, after removing the maximum and minimum values, is selected as the reference value to ensure the robustness of the mapping process. Under normal operating conditions, Generally, the voltage tends to stabilize, and its value is usually less than 1. Therefore, after voltage mapping, it also tends to a certain constant. The role of the balance coefficient of 1 is to prevent the occurrence of zero difference and to allow the input term to deviate from 0, preventing excessive compression of the mapped voltage.

[0095] Based on sparse representation theory, a set of basis vectors D is constructed to approximately represent the cell voltage. k Calculate the mapped voltage z of each cell. j,k Reconstruction error e k ,

[0096] Step 4.1: Initialize the iteration count t = 1; construct the mapped voltage vector Z k =[z 1,k ,z 2,k ,...,z J,k ] T Principal component analysis was used to extract the mapped voltage vector Z. k The first m principal component voltages form the initial basis vector set D. k (1);

[0097] Step 4.2: Using the Lasso regression algorithm, based on the objective function f, obtain the sparse coefficient vector β that minimizes the corresponding objective function value. k (t);

[0098] f = ||z k -D k (t)·β k (t)+λ·β k(t)||

[0099] β k (t) = argmin(f)

[0100] Where λ is the regularization parameter;

[0101] Step 4.3: Calculate the reconstruction error vector E k (t);

[0102] E k (t)=z k -D k (t)·β k (t)

[0103] Step 4.4: Utilize the reconstructed error vector E k (t) and the sparse coefficient vector β k (t) Update the basis vector set D k (t+1);

[0104] D k (t+1)=D k (t)+η·E k (t)·(β k (t)) T

[0105] Where η is the update step size;

[0106] Step 4.5: If ||D k (t+1)-D k If (t)||≤ξ, where ξ is the convergence threshold, then stop the iteration and proceed to step 4.6; otherwise, let t=t+1 and return to step 4.2.

[0107] Step 4.6: Based on the basis vector set D k (t+1) Execute steps 4.2 and 4.3 once to obtain the reconstruction error vector E. k (t+1); Let D k =D k (t+1), e k,j (t+1) represents the reconstruction error vector E. k The j-th element in (t+1);

[0108] The reconstruction error e of each cell k The average of the reconfiguration errors of all cells The deviation is used as the deviation score (DS) of the battery cell. k Calculate the adaptive threshold V th If the deviation score of the battery cell is DS k Greater than the adaptive threshold V thIf so, the cell is marked as a cell with abnormal deviation.

[0109]

[0110] V th =θ·μ DS +(1-θ)·(ζ·max(DS k )+(1-ζ)·(μ DS +δ·σ DS ))

[0111] Where θ,ζ∈(0,1) are weighting coefficients; μ DS Deviation score DS for all cells k The mean; σ DS Deviation score DS for all cells k The standard deviation of ; δ is the magnification factor;

[0112] Step 3: Construction of multidimensional features and calculation of mutual information:

[0113] Judging high-risk cells based solely on abnormal cells screened by DS (Data Sensing) may lead to missed or false alarms due to the use of a single voltage feature. To improve the accuracy of the final early warning, a mutual information method based on multi-dimensional features is used to quantify the correlation between multi-dimensional feature parameters of different cells and further evaluate the cell status. In actual operation, the mean characteristics of all cells within the same data segment can characterize the behavior of cells under normal conditions. Therefore, the cell status can be judged by calculating the mutual information between each cell and the mean feature. Voltage and temperature are basic parameters reflecting the operating status and internal thermal management of cells. However, analyzing only static voltage and temperature may not be sufficient to accurately capture the dynamic changes of early cell anomalies. Therefore, voltage change rate and temperature change rate are introduced as dynamic features, forming a four-dimensional feature including voltage, temperature, voltage change rate, and temperature change rate for mutual information calculation.

[0114] Constructing the multidimensional feature vector C for each battery cell k Multidimensional mean eigenvector of all battery cells For each dimension of data, the number of groups B is designed based on the Freedman-Diaconis rule. d The multidimensional feature vector C of each battery cell is calculated based on the mutual information method. k Multidimensional mean eigenvector of all battery cells Mutual information value I k Cells with significantly outlier mutual information values ​​are marked as cells with abnormal mutual information.

[0115] Constructing the multidimensional feature vector C for each battery cell k :

[0116] C k =[c 2,k ,c 3,k ,...,c J,k ]

[0117] c i,k =[z i,k ,T i,k ,Δz i,k ,ΔT i,k ]

[0118] Where i = 2, 3, ..., J;

[0119] For each dimension of data, construct a data matrix X. d d = 1, 2, 3, 4, data matrix X d element x in d,i,k for:

[0120]

[0121] Construct the multidimensional mean feature vector of all battery cells

[0122]

[0123] in,

[0124] The multidimensional feature vector C of each battery cell is calculated based on the mutual information method. k Multidimensional mean eigenvector of all battery cells Mutual information value I k ;

[0125] Because of the dimensional inconsistency between the dynamic rate of change and the original voltage and temperature, the method of discarding the first time-instance data of voltage and temperature is adopted to ensure that the data length of different feature dimensions is consistent, and to avoid additional interference that may be introduced by methods such as zero padding or difference smoothing.

[0126] Before calculating mutual information, the four-dimensional feature sequences need to be grouped to estimate the probability distribution. To address the impact of sequence length differences between different data segments, this invention designs the number of groups B based on the Freedman-Diaconis rule. d :

[0127]

[0128] Where, x d,max With x d,min For data matrix X d Maximum and minimum elements in the middle; IQR(X) d ) is the data matrix Xd Interquartile spacing; This indicates rounding up to the nearest integer.

[0129] For cell k, each dimension d falls into the p-th position. d The marginal probability of grouping is defined as:

[0130]

[0131] Where bin(·) is the sample at the corresponding p d Number of groups.

[0132] Similarly, for the mean vector in dimension d, the q-th... d The marginal probability of grouping is defined as:

[0133]

[0134] The joint probability of cell k and the mean vector in the four dimensions is:

[0135]

[0136] Where p = (p1, p2, p3, p4) and q = (q1, q2, q3, q4) represent the grouping combinations of cell k and mean vector in four dimensions; the denominator represents the total number of samples under all combinations; and the numerator represents the number of samples that simultaneously satisfy cell k and mean vector in the specified numbered group.

[0137] The mutual information between cell k and the mean vector can be expressed as:

[0138]

[0139] Calculate the mutual information threshold T MIV :

[0140]

[0141] Where φ is the adjustment factor;

[0142] The corresponding I k Less than T MIV The battery cells were marked as having abnormal mutual information.

[0143] Step 4: Joint Assessment and High-Risk Judgment. After completing the initial screening of abnormal cells and the mutual information calculation of cells, this invention proposes a joint assessment method to accurately screen cells with a high risk of thermal runaway.

[0144] If a battery cell is marked as having abnormal deviation in three consecutive discharge data segments, then the battery cell is determined to be a high-risk battery cell for thermal runaway.

[0145] If a battery cell is marked as having abnormal deviation in two consecutive discharge data segments, then the battery cell is determined to be a potentially high-risk battery cell.

[0146] If a battery cell has been marked as a deviation abnormality cell twice in the historical data segment, and the deviation score is abnormal again when the next discharge data segment is detected, then the battery cell is determined to be a potentially high-risk battery cell.

[0147] If a cell has been identified as a potentially high-risk cell, its mutual information value in the corresponding data segment is further evaluated; if the mutual information value is detected and the cell is marked as having abnormal mutual information, then the cell is determined to be a high-risk cell for thermal runaway.

[0148] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An early warning strategy for thermal runaway cells in marine energy storage equipment, characterized in that: The system acquires operational data of the ship's energy storage equipment over a period of time and extracts data segments from the ship's energy storage equipment during the discharge phase. The data segments during the discharge phase include the voltage and temperature values ​​of each cell at each sampling time. By using multi-scale step mode decomposition, the voltage value of each cell at each sampling time is decomposed into multiple sets of mode functions. The Spearman rank correlation coefficient between each mode function and the original voltage value is calculated. The mode function with the smallest corresponding Spearman rank correlation coefficient value is removed. The voltage of the remaining mode functions is reconstructed to obtain the denoised voltage. Based on the nonlinear voltage mapping method of the improved Sigmoid function, the denoised voltage of each cell is adaptively mapped to obtain the mapped voltage. Based on sparse representation theory, the reconstruction error of the mapped voltage of each cell is calculated. The deviation between the reconstruction error of the mapped voltage of each cell and the mean of the reconstruction errors of the mapped voltage of all cells is used as the deviation score of the cell. An adaptive deviation threshold is constructed, and cells with a deviation score greater than the adaptive deviation threshold are marked as cells with abnormal deviation. Construct the multidimensional feature vector of each cell and the multidimensional mean feature vector of all cells, and calculate the mutual information value of the multidimensional feature vector of each cell and the multidimensional mean feature vector of all cells based on the mutual information method; construct an adaptive mutual information threshold, and mark the cells whose corresponding mutual information value is less than the adaptive mutual information threshold as cells with abnormal mutual information. By combining the detection results of deviation anomalies and mutual information anomalies, an early warning can be provided for thermal runaway cells in ship energy storage equipment through a joint evaluation strategy.

2. The early warning strategy for thermal runaway cells in ship energy storage equipment according to claim 1, characterized in that: The shipboard energy storage equipment includes K groups of battery cells, and the data segment for each discharge stage includes the voltage value V of each battery cell at each sampling time. j,k With temperature value T j,k V j,k T represents the voltage value of the k-th cell at the j-th sampling time. j,k This represents the temperature value of the k-th cell at the j-th sampling time; k = 1, 2, ..., K; j = 1, 2, ..., J; J is the total number of sampling times, and the time interval between adjacent sampling times is ΔT; The method utilizes multi-scale step mode decomposition to determine the voltage value V of each cell at each sampling time. j,k Decomposed into G sets of modal functions u g,k (j); Calculate the modal functions u of each mode. g,k (j) and the original voltage value V j,k The Spearman rank correlation coefficient ρ between them g,k : Where R(x) represents the rank of x; The corresponding Spearman rank correlation coefficient ρ g,k The mode function with the smallest value u g,k (j) Remove the remaining G-1 group of mode functions and reconstruct the voltage to obtain the denoised voltage.

3. The early warning strategy for thermal runaway cells in marine energy storage equipment according to claim 2, characterized in that: The nonlinear voltage mapping method based on the improved Sigmoid function selects the average voltage of all cells after denoising, after removing the maximum and minimum values ​​at the same sampling time. As a reference voltage, the voltage of each cell after noise reduction Perform adaptive mapping to obtain the mapped voltage z j,k : Where α is the amplification factor, used to adjust the sensitivity of the mapping; γ is the dynamic power, used to control the dynamic changes in voltage difference.

4. The early warning strategy for thermal runaway cells in marine energy storage equipment according to claim 3, characterized in that: The reconstruction error e of the mapped voltage of each cell is calculated based on sparse representation theory. k Specifically: Step 1: Initialize the iteration count t = 1; construct the mapped voltage vector Z k =[z 1,k ,z 2,k ,...,z J,k ] T Principal component analysis was used to extract the mapped voltage vector Z. k The first m principal component voltages form the initial basis vector set D. k (1); Step 2: Using the Lasso regression algorithm, based on the objective function f, obtain the sparse coefficient vector β that minimizes the corresponding objective function value. k (t); f=||z k -D k (t)·β k (t)+λ·β k (t)|| β k (t)=argmin(f) Where λ is the regularization parameter; Step 3: Calculate the reconstruction error vector E k (t); E k (t)=z k -D k (t)·β k (t) Step 4: Utilize the reconstructed error vector E k (t) and the sparse coefficient vector β k (t) Update the basis vector set D k (t+1); D k (t+1)=D k (t)+η·E k (t)·(β k (t)) T Where η is the update step size; Step 5: If ||D k (t+1)-D k If (t)||≤ξ, and ξ is the convergence threshold, then stop the iteration and proceed to step 6; otherwise, let t=t+1 and return to step 2. Step 6: Based on the basis vector set D k (t+1) Perform steps 2 and 3 once to obtain the reconstruction error vector E. k (t+1); Let D k =D k (t+1), e k,j (t+1) represents the reconstruction error vector E. k The j-th element in (t+1).

5. The early warning strategy for thermal runaway cells in marine energy storage equipment according to claim 4, characterized in that: The reconstruction error e of each cell k The average of the reconfiguration errors of all cells The deviation is used as the deviation score (DS) of the battery cell. k Calculate the adaptive deviation threshold V th If the deviation score of the battery cell is DS k Greater than the adaptive deviation threshold V th If so, the cell is marked as a cell with abnormal deviation. V th =θ·μ DS +(1-θ)·(ζ·max(DS k )+(1-ζ)·(μ DS +d·s DS )) Where θ,ζ∈(0,1) are weighting coefficients; μ DS Deviation score DS for all cells k The mean; σ DS Deviation score DS for all cells k The standard deviation is δ; δ is the amplification factor.

6. The early warning strategy for thermal runaway cells in marine energy storage equipment according to claim 5, characterized in that: The construction of multidimensional feature vector C for each battery cell k : C k =[c 2,k ,c 3,k ,...,c J,k ] c i,k =[z i,k ,T i,k ,Δz i,k ,ΔT i,k ] Where i = 2, 3, ..., J; For each dimension of data, construct a data matrix X. d d = 1, 2, 3, 4, data matrix X d element x in d,i,k for: Construct the multidimensional mean feature vector of all battery cells in, The multidimensional feature vector C of each battery cell is calculated based on the mutual information method. k Multidimensional mean eigenvector of all battery cells Mutual information value I k Calculate the mutual information threshold T MIV , will correspond to I k Less than T MIV The battery cells were marked as having abnormal mutual information. Where φ is the adjustment factor; 7. The early warning strategy for thermal runaway cells in marine energy storage equipment according to claim 1, characterized in that: The joint evaluation strategy is as follows: If a battery cell is marked as having abnormal deviation in three consecutive discharge data segments, then the battery cell is determined to be a high-risk battery cell for thermal runaway. If a battery cell is marked as having abnormal deviation in two consecutive discharge data segments, then the battery cell is determined to be a potentially high-risk battery cell. If a battery cell has been marked as a deviation abnormality cell twice in the historical data segment, and the deviation score is abnormal again when the next discharge data segment is detected, then the battery cell is determined to be a potentially high-risk battery cell. If a cell has been identified as a potentially high-risk cell, its mutual information value in the corresponding data segment is further evaluated; if the mutual information value is detected and the cell is marked as having abnormal mutual information, then the cell is determined to be a high-risk cell for thermal runaway.

8. A computer device / equipment / system, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that: When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images