A lithium battery health state estimation method and system based on efficient fuzzy clustering integration of equivalent granularity

Abstract

The present invention discloses a lithium battery health state estimation method and system based on efficient fuzzy clustering integration based on equivalent granularity, which relates to the field of battery health state prediction, and solves the problem that the existing method regards each sample as an independent individual, while ignoring the equivalent relationship between samples, and uses all sample points to learn the core main cluster structure of the data set. The present invention constructs an equivalent point set of the data set, and constructs a potential equivalent granularity set based on the potential equivalent granularity of the equivalent subset elements, divides the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set, measures the fuzzy relationship between any two core equivalent granularities, constructs a core equivalent granularity relationship matrix, identifies the main cluster structure, assigns the fuzzy equivalent granularity to the obtained main cluster structure, and at the same time, restores the label of the hidden sample point to the label of the potential equivalent granularity equivalent to it, and obtains a consensus result; the battery health state is predicted based on the consensus result.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery health state prediction, and in particular to a lithium battery health state estimation method and system based on efficient fuzzy clustering integration of equivalent granularity. Background Art

[0002] Lithium-ion batteries, with their excellent energy density, long cycle life, and fast charge and discharge capabilities, have become the core technology in key areas such as electric vehicles, mobile devices, and large-scale energy storage systems. However, as batteries undergo continuous charge and discharge cycles and are tested by environmental factors, performance degradation and aging become inevitable phenomena. Therefore, accurate assessment of the battery's state of health (SOH) is of vital importance for continuous monitoring of battery performance and prevention of potential performance degradation. An efficient and accurate SOH assessment mechanism can not only ensure the safe operation of the battery, but also play a vital role in optimizing the battery management system (BMS), extending the battery life, and enhancing user trust.

[0003] In recent years, data-driven battery health status prediction methods have attracted widespread attention in the industry and have shown great application potential and development prospects. This data-based prediction technology, with its accuracy and generalization, is gradually becoming the forefront of battery health management, heralding a new era of battery monitoring and maintenance. However, the traditional data-driven SOH prediction method has obvious defects, that is, most methods directly use known full-life cycle battery data for training to complete the SOH prediction of the target battery, which makes the useful information decentralized, making it difficult to obtain a more ideal prediction result. The degradation process of the battery is complex and changeable, and different degradation stages often show significantly different degradation trends. Therefore, if the data of the entire life cycle are placed together for training, the prediction ability of the model will often be weakened. As for clustering and integrating uncertain data, the traditional clustering integration method is not enough to effectively mine more valuable information.

[0004] Considering the complexity of these uncertain information, researchers have introduced some granular computing methods such as rough set theory and fuzzy mathematics theory into clustering ensembles, trying to capture the structural characteristics of the data set from some new ways. Related research shows that rough set theory and fuzzy mathematics theory can indeed better deal with the uncertainty problem in clustering ensembles.

[0005] Although clustering ensemble methods for uncertain data have achieved good research results, research in this area is still relatively scarce. It is worth noting that most previous studies regard each sample as an independent individual, while ignoring the equivalence relationship between samples. This will lead to a waste of resources and the inability to fully utilize local data information, thereby reducing the clustering efficiency of the model. Clustering efficiency is of course crucial while ensuring clustering accuracy. Especially for large-scale data sets, it is also meaningless if the algorithm is seriously time-consuming or even difficult to obtain the final clustering results. Therefore, how to effectively improve clustering efficiency while maintaining accuracy in clustering ensemble models that deal with uncertainty is a key challenge. In addition, existing clustering ensemble methods usually use all sample points to learn the core cluster structure of the data set. The participation of uncertain points will mislead the generation of the actual core cluster structure, which will directly affect the final clustering results of the model. Summary of the invention

[0006] The technical problems to be solved by the present invention are:

[0007] Existing methods treat each sample as an independent individual and ignore the equivalence relationship between samples, resulting in a waste of resources. In addition, existing methods use all sample points to learn the core cluster structure of the data set. The participation of uncertain points will mislead the generation of the actual core cluster structure and affect the final clustering results of the model.

[0008] The present invention adopts the following technical solutions to solve the above technical problems:

[0009] The present invention provides a lithium battery health state estimation method based on efficient fuzzy clustering integration of equivalent granularity, comprising the following steps:

[0010] S1. Collect the health characteristic data set of lithium battery charge and discharge cycle, and cluster the data set to obtain basic clustering result data;

[0011] S2, based on the base clustering result data, the samples are represented by dummy variables and scaled;

[0012] S3, regard the sample points that are always in the same cluster in the base clustering result as a potential equivalent granularity, construct the equivalent point set of the data set, regard the first sample point in each subset as the potential equivalent granularity of the subset, and construct the potential equivalent granularity set;

[0013] S4, dividing the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set;

[0014] S5, measure the fuzzy relationship between any two core equivalent granularities, construct a relationship matrix composed of core equivalent granularities, and identify the subject cluster structure based on the relationship matrix;

[0015] S6. Use the fuzzy relationship to assign the fuzzy equivalent granularity to the obtained main cluster structure; at the same time, restore the labels of the hidden sample points in the equivalent point set to the labels of the potential equivalent granularity equivalent to it, and obtain the final consensus result composed of all sample points;

[0016] S7. Predict the battery health status based on the consensus results.

[0017] Furthermore, S1 includes the following process:

[0018] Collect the lithium battery charge and discharge cycle health feature data set DX = {x1, x2, ..., x j ,…,x N}, where N represents the number of sample points in the data set DX. The data set DX is divided L times to obtain a set of basic clustering results Π={π 1 ,π 2 ,…,π L},in represents the i-th base clustering result in Π, is the i-th base clustering result π i The lth cluster, P i is the i-th base clustering result π i The number of clusters, the set of all clusters in the set of base clustering results Π is Where P C is the total number of all clusters in Π, and the set Π is also expressed in the form of the base clustering result matrix, that is, Π = [π ij ] L×N , where π ij Represents the cluster label assigned by the i-th partition to the j-th sample point;

[0019] Define the co-correlation matrix Α=[a jk ] N×N for:

[0020]

[0021] in,

[0022] Furthermore, S2 includes the following process:

[0023] Based on the base clustering result matrix Π=[π ij ] L×N , for any sample point x in the data set DX j , construct its dummy variable feature representation dx j for:

[0024]

[0025] in, The length of is the base clustering result π under the i-th partition i The number of clusters P i , from left to right, each element represents the 1st cluster to the Pth cluster. i Cluster, based on the DBI index, the dummy variable feature of the sample point is represented by dx j Upgraded to scaled dummy variable feature representation d * x j :

[0026]

[0027] Among them, the scaling factor Used to measure the contribution of the i-th base clustering result, DBI i It is the DBI index value score of the base clustering result under the i-th partition.

[0028] Furthermore, the sparse partitioning strategy based on third-party random points in S4 divides the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set, including the following process:

[0029] S4-1. Randomly generate a point x that is equal to any sample point x in the data set DX. j A third-party sample point x with the same feature dimension g , calculate all sample points to the random point x g The distance between them generates a distance sequence Reconstruct the sequence DX according to the ascending order of the distance values ​​in D corresponding to the sample points in DX # ={d # x1,d # x2,…,d # x N}, in sequence DX # Calculate the distance between the "current target point" and the "neighboring point";

[0030] S4-2, "neighboring points" are sample points within the set iteration step range, and iteration is performed until any sample point d is obtained. # x j Its "neighboring point" d # x m The distance is obtained by sparse distance matrix

[0031] based on Measuring any sample point x j The local density ρ j :

[0032]

[0033] in, Represents a sparse distance matrix The number of effective distances in the jth row of θ is used to ensure that the denominator is always meaningful; a larger density sample set with a density higher than the local density mean is obtained.

[0034] S4-3, according to the potential equivalent granularity set DX * , equivalent point set E and larger density sample set ρ bigger , the potential equivalent granularity set DX * Divided into core equivalent granularity set CS and fuzzy equivalent granularity set FS:

[0035] CS={d * xr|ρ bigger ∩E r ≠φ,r=1,2,…,|E|} (7)

[0036] FS=DX * -CS (8)

[0037] Among them, E r is the rth subset element of E, |E| represents the number of subset elements in the equivalent point set E, d * x r For DX * The potential equivalent granularity corresponding to the rth position in .

[0038] Furthermore, in S4-1, all sample points to the random point x are calculated g The distance between them is:

[0039]

[0040] Furthermore, the sparse partitioning strategy based on equivalent point sets in S4 divides the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set, including the following process:

[0041] According to the equivalent point set E, calculate the sparse distance matrix of DX in, For any sample point d * x j With d * x j Any other sample point d in the subset element of E * x m The distance between any sample point d * x j The local density ρ j for:

[0042]

[0043] The division of potential equivalent granularity includes two cases: (1) If Then, the potential equivalent granularity set is divided into the core equivalent granularity set CS and the fuzzy equivalent granularity set FS by equations (7) and (8); (2) If Then the non-zero density set ρ non-zero ={d * x j |ρ j ≠0,j∈[1,N]}, then:

[0044] CS={d * x r |ρ non-zero ∩E r ≠φ,r=1,2,…,|E} (11)

[0045] Among them, d * x r For DX * The potential equivalent granularity corresponding to the rth position in the fuzzy equivalent granularity set FS is DX * -CS.

[0046] Further, S5 includes the following steps:

[0047] According to the core equivalent granularity set CS, the γ-fuzzy operator is used Measuring the equivalent granularity of any two cores * x j and d * x k The fuzzy relationship between jk :

[0048]

[0049] Among them, |d * x j +d * x k | represents the 2-norm of the sum of two vectors, d * x j d * x k represents the inner product of two vectors,

[0050] γ-Fuzzy Operator is defined as:

[0051] Let the mapping S:[0,1] 2 →[0,1], for γ∈[-1,0], there is a γ-fuzzy operator

[0052]

[0053] Construct the relationship matrix F composed of all core equivalent granularities = [f jk ] |CS|×|CS| , where |CS| represents the number of elements in the core equivalent granularity set CS;

[0054] Based on the relationship matrix F, the DPC model is used to give the main cluster structure result of the data set π θ .

[0055] Furthermore, S6 includes the following process:

[0056] First, calculate π θ Any main cluster C i The mean vector V i :

[0057]

[0058] Where k is π θ The number of clusters in ;

[0059] Then, using the γ-fuzzy operator To evaluate the jth fuzzy equivalent granularity d * x j and the i-th mean vector V i The fuzzy relationship between:

[0060]

[0061] Among them, |FS| is the number of fuzzy equivalent granularities in FS;

[0062] Finally, the maximum membership principle is used to allocate fuzzy equivalent granularity, that is, any fuzzy equivalent granularity d * x j Assigned to the known primary cluster with which it has the largest membership relationship;

[0063] The labels of the hidden sample points in the equivalent point set E are restored to the labels of the equivalent potential granularity, and the final consensus result π composed of all sample points is obtained. * .

[0064] Furthermore, a regression model SVR is used in S7 to predict the battery health status.

[0065] The present invention also provides a lithium battery health status estimation system based on efficient fuzzy clustering integration of equivalent granularity. The system has a program module corresponding to the steps of the method described in any of the above technical solutions, and executes the steps in the above-mentioned lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity during runtime.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] The present invention discloses a lithium battery health status estimation method and system based on efficient fuzzy clustering integration of equivalent granularity. First, the original data is converted into a scaled dummy variable feature representation through a base clustering result matrix, and then a potential equivalent granularity set with an equivalent relationship is searched through the base clustering result matrix. This process expands the model from the sample level to the equivalent granularity level to achieve the purpose of initially reducing the data scale and reducing the additional running time occupied by sample points with equal utility. The present invention uses the core equivalent granularity to construct the main cluster structure of the data set, and the fuzzy equivalent granularity is assigned to the known main cluster structure. Finally, the labels of the hidden equivalent sample points are restored to obtain the final division results of all sample points.

[0068] The present invention designs two different modes of sparse strategies to discover the specific responsibilities of each potential equivalent granularity in the underlying data structure, thereby dividing the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set, further reducing the input data scale of the consensus process and further improving the clustering efficiency.

[0069] The model proposed in the present invention has excellent clustering accuracy and clustering efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A flowchart of a lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity in an embodiment of the present invention;

[0071] Figure 2 The distribution and base clustering result diagram of the example data set in the embodiment of the present invention;

[0072] Figure 3 In the embodiment of the present invention, the new sequence DX # Schematic diagram of the sparse computing method below;

[0073] Figure 4 This is a diagram showing the fitting of the predicted value and the true value of the SOH of the target battery No. 2 using the data of the battery No. 1 as the training set in the embodiment of the present invention;

[0074] Figure 5 This is a diagram showing the fitting of the predicted value and the true value of the SOH of the target battery No. 2 using the data of the battery No. 3 as the training set in the embodiment of the present invention;

[0075] Figure 6 This is a diagram showing the fitting of the predicted value and the true value of the SOH of the target battery No. 2 using the data of the battery No. 4 as the training set in an embodiment of the present invention. DETAILED DESCRIPTION

[0076] In order to enable those skilled in the art to better understand the scheme of the present invention, exemplary implementations or embodiments of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described implementations or embodiments are only implementations or embodiments of a part of the present invention, not all of them. Based on the implementations or embodiments of the present invention, all other implementations or embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0077] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0078] Combination Figures 1 to 3 As shown, the present invention provides a lithium battery health state estimation method based on efficient fuzzy clustering integration of equivalent granularity, comprising the following steps:

[0079] S1. Collect the health characteristic data set of lithium battery charge and discharge cycle, and cluster the data set to obtain basic clustering result data;

[0080] S2, based on the base clustering result data, the samples are represented by dummy variables and scaled;

[0081] S3, regard the sample points that are always in the same cluster in the base clustering result as a potential equivalent granularity, construct the equivalent point set of the data set, regard the first sample point in each subset as the potential equivalent granularity of the subset, and construct the potential equivalent granularity set;

[0082] S4, dividing the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set;

[0083] S5, measure the fuzzy relationship between any two core equivalent granularities, construct a relationship matrix composed of core equivalent granularities, and identify the subject cluster structure based on the relationship matrix;

[0084] S6. Use the fuzzy relationship to assign the fuzzy equivalent granularity to the obtained main cluster structure; at the same time, restore the labels of the hidden sample points in the equivalent point set to the labels of the potential equivalent granularity equivalent to it, and obtain the final consensus result composed of all sample points;

[0085] S7. Based on the consensus results, the regression model SVR is used to predict the battery health status.

[0086] Preferably, S1 includes the following process:

[0087] Collect the lithium battery charge and discharge cycle health feature data set DX = {x1, x2, ..., xj, ..., x N}, where N represents the number of sample points in the data set DX, using the same single clustering algorithm under random initialization or different parameter settings or multiple base clustering algorithms {A1, A2, …, A L}, the data set DX is divided L times, and a set of basic clustering results Π={π 1 ,π 2 ,…,π L},in represents the i-th base clustering result in Π, is the i-th base clustering result π i The lth cluster, P i is the i-th base clustering result π i The number of clusters, the set of all clusters in the set of base clustering results Π is Where P C is the total number of all clusters in π. For the convenience of expression, the set π is also expressed in the form of a base clustering result matrix, that is, π = [π ij ] L×N , where π ij Represents the cluster label assigned by the i-th partition to the j-th sample point.

[0088] Define the co-correlation matrix Α=[a jk ] N×N for:

[0089]

[0090] in,

[0091] By analyzing equation (1), we can see that a jk The larger the sample point x is, the j With x k In more base clustering results, they are divided into the same cluster. In other words, x j With x k The greater the possibility of being classified into the same cluster in the final integrated result. jk =1, it means that this group of sample points belong to the same cluster in all base clustering results. This means that x j With x k They are also in the same cluster in the final partition.

[0092] Preferably, S2 includes the following process:

[0093] Based on the base clustering result matrix Π=[π ij ] L×N , for any sample point x in the data set DX j , construct its dummy variable feature representation dx j for:

[0094]

[0095] in, The length of is the base clustering result π under the i-th partition i The number of clusters P i , from left to right, each element represents the 1st cluster to the Pth cluster. i cluster, Only the πth ij The value of the element is 1, and the values ​​of the remaining elements are 0. In addition, considering that there are indeed different degrees of quality differences between the randomly generated base clustering results, the dummy variable feature of the sample point is represented by dx based on the DBI index. j Upgraded to scaled dummy variable feature representation d * x j :

[0096]

[0097] Among them, among them, The length of P i , from left to right, each element represents the 1st cluster to the Pth cluster. i cluster. Only the πth ij The value of the element is ω i , and the values ​​of the remaining elements are all 0. Scaling factor Used to measure the contribution of the i-th base clustering result, DBI i is the DBI index score of the base clustering result under the i-th partition. The smaller the value, the better the clustering effect. On the one hand, this process can effectively eliminate the influence of the size of the cluster label value on the relationship between sample points; on the other hand, it can accurately map the actual differences between the base clustering results.

[0098] Preferably, S3 includes the following process:

[0099] Suppose that a sample data set DX containing 14 sample points is given = {x1,x2,x3,x4,x5,x6,x7,x8,x9,x 10 ,x 11 ,x 12 ,x 13 ,x 14}.like Figure 2 The actual distribution of the sample data set and the four base clustering results are shown as follows: 1 ,π 2 ,π 3 ,π 4 First, based on the result matrix Π formed by the four base clustering partitions and equation (1), the co-correlation matrix Α=[a jk] 14×14 , the co-correlation matrix A is shown in Table 1.

[0100] By observing the co-correlation matrix Α in Table 1, it can be found that the sample point d * x2,d * x3 and d * The similarity relationship between x4 and x4 is 1. This means that they are always in the same cluster in all base clustering divisions. In addition, the similarity relationship between the three sample points and the remaining sample points is exactly the same. This shows that the three sample points are equivalent in the overall relationship calculation and can be regarded as a potential equivalent granularity. Therefore, the sample points with the same attributes can constitute the equivalent point set E = {{d * x1},{d * x2,d * x3,d * x4},{d * x5,d * x6,d * x 11},{d * x7,d * x8,d * x9},{d * x 10 ,d * x 12 ,d * x 13 ,d * x 14}}.

[0101] It is worth noting that the sample points at this time are represented by scaled dummy variable features. That is, different from x2, x3 and x4 in the original data set, d * x2,d * x3 and d * x4 is exactly the same. For convenience, each subset element in the equivalent point set E retains the first sample point in sequence as the potential equivalent granularity of the equivalent subset element by default. Therefore, the equivalent replacement potential equivalent granularity set of the sample data set DX is DX * ={d * x1,d * x2,d * x5,d * x7,d * x 10 It can be seen that the construction of potential equivalent granularity can reduce the amount of initial input data in the consensus process from 14 to 5, and the calculation scale is reduced by more than half.

[0102] Table 1

[0103]

[0104] Preferably, the sparse partitioning strategy based on third-party random points in S4 divides the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set, including the following process:

[0105] S4-1, to calculate any sample point x in the original data set DX j The local density ρ j , local density ρ j The definition is given by x j The distance between the sample points is determined by the distance relationship with the nearest neighbors. Therefore, in order to measure the distance between sample points with the least amount of calculation, the third-party random point method is used here to sparsely search the neighbors of each sample point, that is:

[0106] If the distances of two points to the same third-party data point differ greatly, then the two points are dissimilar and are not neighbors of each other. If the distances of two points to the third-party data point differ very little, then the two points are not necessarily similar, and vice versa. Based on this principle, first randomly generate a point x that is similar to any sample point in the data set DX. j A third-party sample point x with the same feature dimension g , calculate all sample points to the random point x g The Euclidean distance between:

[0107]

[0108] Generate a distance sequence Reconstruct the sequence DX according to the ascending order of the distance values ​​in D corresponding to the sample points in DX # ={d # x1,d # x2,…,d # x N},by Figure 2 The example data set is illustrated in Figure 2. # In this case, we only need to calculate the distance between the "current target point" and the "neighboring point", such as Figure 3 shown.

[0109] S4-2, "neighboring points" are sample points within the range of the set iteration step size step = 0.4N-1, and sample points outside the iteration step size

[0110] No calculation is required. Each iteration is based on the previous measurement number, and the previously calculated sample distance does not need to be recalculated later. Iterate until any sample point d is obtained. # x jIts "neighboring point" d # x m The distance is obtained by sparse distance matrix

[0111]

[0112] By observation It can be seen that more than half of the useless distance calculations are saved in the entire process of searching for neighbors.

[0113] based on Measuring any sample point x j The local density ρ j :

[0114]

[0115] in, Represents a sparse distance matrix The number of effective distances in the jth row of , θ is taken as 0.5, to ensure that the denominator is always meaningful; a larger density sample set higher than the local density mean is obtained

[0116] S4-3, according to the potential equivalent granularity set DX * , equivalent point set E and larger density sample set ρ bigger , the potential equivalent granularity set DX * Divided into core equivalent granularity set CS and fuzzy equivalent granularity set FS:

[0117] CS={d * x r |ρ bigger ∩E r ≠φ,r=1,2,…,|E|} (8)

[0118] FS=DX * -CS (9)

[0119] Among them, E r is the rth subset element of E, |E| represents the number of subset elements in the equivalent point set E, d * x r For DX * The potential equivalent granularity corresponding to the rth position in .

[0120] Preferably, the sparse partitioning strategy based on the equivalent point set in S4 divides the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set, including the following process:

[0121] The sample points contained in each subset element of the equivalent point set E are all equivalent and always belong to the same cluster in all base clustering partitions. Therefore, from a global perspective, this just shows that any subset element E r The sample points in are each other's "most loyal neighbors". According to this feature, Figure 2 The sparse distance matrix based on this strategy in the example for:

[0122]

[0123] For any sample point d * x j With d * x j Any other sample point d in the subset element of E * x m (“most loyal neighbors”), by analyzing the sparse distance matrix It can be found that only a small amount of calculation is needed to obtain any sample point d * x j The distance relationship between neighbors. Define any sample point d * x j The local density ρ j for:

[0124]

[0125] In particular, when When "[]" The value of is 0;

[0126] The division of potential equivalent granularity includes two cases: (1) If Then, the potential equivalent granularity set is divided into the core equivalent granularity set CS and the fuzzy equivalent granularity set FS by equations (8) and (9); (2) If Then the non-zero density set ρ non-zero ={d * x j |ρ j ≠0,j∈[1,N]}, then:

[0127] CS={d * xr|ρ non-zero ∩E r ≠φ,r=1,2,…,|E} (11)

[0128] Among them, d * x r For DX *The potential equivalent granularity corresponding to the rth position in the fuzzy equivalent granularity set FS is DX * -CS.

[0129] Preferably, S5 comprises the following steps:

[0130] Since the core equivalent granularity has relatively high stability and local density, a clearer and more accurate main cluster structure can be found based on the core equivalent granularity than on the original data set composed of all sample points.

[0131] According to the core equivalent granularity set CS, the γ-fuzzy operator is used Measuring the equivalent granularity of any two cores * x j and d * x k The fuzzy relationship between jk :

[0132]

[0133] Among them, |d * x j +d * x k | represents the 2-norm of the sum of two vectors, d * x j d * x k represents the inner product of two vectors,

[0134] γ-Fuzzy Operator is defined as:

[0135] Let the mapping S:[0,1] 2 →[0,1], for γ∈[-1,0], there is a γ-fuzzy operator

[0136]

[0137] Construct the relationship matrix F composed of all core equivalent granularities = [f jk ] |CS|×|CS| , where |CS| represents the number of elements in the core equivalent granularity set CS;

[0138] γ is a parameter in the fuzzy operator, and its value belongs to [-1,0]. Once γ takes a certain value in the adjustable range [-1,0], then the sample point d * x j and d * x k The corresponding operation relationship between them is determined accordingly; based on the relationship matrix F, the DPC model is used to give the main cluster structure result π of the data setθ .

[0139] The DPC model has three important advantages in exploring the underlying cluster structure of a dataset. First, the DPC model can identify cluster centers through decision diagrams without pre-specifying the number of clusters. Second, it can handle non-convex datasets well and does not require an iterative process. In addition, compared with the γ-fuzzy operator The combined DPC model has a stronger ability to handle fuzzy relationships.

[0140] Preferably, S6 includes the following process:

[0141] First, calculate π θ Any main cluster C i The mean vector V i :

[0142]

[0143] Where k is π θ The number of clusters in ;

[0144] Then, using the γ-fuzzy operator To evaluate the jth fuzzy equivalent granularity d * x j and the i-th mean vector V i The fuzzy relationship between:

[0145]

[0146] Among them, |FS| is the number of fuzzy equivalent granularities in FS;

[0147] Finally, the maximum membership principle is used to allocate fuzzy equivalent granularity, that is, any fuzzy equivalent granularity d * x j Assigned to the known primary cluster with which it has the largest membership relationship;

[0148] The labels of the hidden sample points in the equivalent point set E are restored to the labels of the equivalent potential granularity, and the final consensus result π composed of all sample points is obtained. * .

[0149] According to the above description of the efficient fuzzy clustering ensemble (EFCEG) model based on equivalent granularity, the overall implementation process of EFCEGs can be summarized as Algorithm 1. For the convenience of description, the EFCEGs algorithms embedded in two different potential equivalent granularity partitioning strategies (based on third-party random points and sparse partitioning strategies based on equivalent point sets) are called EFCEG-A and EFCEG-B respectively.

[0150]

[0151]

[0152] The present invention proposes a lithium battery health status estimation method (algorithm) based on an efficient fuzzy clustering integration of equivalent granularity, which is the underlying technical core of the present invention. Various products can be derived based on the algorithm.

[0153] Based on the method proposed in the present invention, a lithium battery health status estimation system based on efficient fuzzy clustering integration of equivalent granularity is developed using a programming language. The system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity during runtime.

[0154] Example 1

[0155] The present invention is mainly aimed at the experimental lithium battery data set, which contains four batteries, No. 1, No. 2, No. 3, and No. 4 (the health characteristic parameters involved are the charge capacity variance, charge cut-off voltage, internal resistance difference, and temperature mean of each charge and discharge cycle). Here, No. 2 battery is used as the target data set, and the remaining three batteries are used as training sets. In addition, according to the four classic performance indicators MAE, RMSE, MAPE, and R 2 To evaluate the final SOH prediction effect of the original linear SVR model and the two improved models (EFCEG-A+SVR and EFCEG-B+SVR) proposed in this invention. The smaller the values ​​of MAE, RMSE and MAPE, the better the prediction effect. On the contrary, 2 The larger the value, the better.

[0156] from Figures 4 to 6 The experimental results show that when the data of different battery models are used as the training set, the prediction capabilities of the proposed EFCEG-A+SVR and EFCEG-B+SVR models are significantly improved compared with the original linear SVR model. 2 The average value of has reached more than 0.998, and even exceeded 0.999 in the best case. The MAE, RMSE and MAPE index values ​​have all been roughly reduced by more than four times. It is particularly worth noting that when a method has shown excellent performance in the initial stage, the difficulty of further improving the model performance will increase significantly, and the space for performance improvement is relatively limited. Therefore, the proposed method can be used in the overall R 2The indicator exceeds 0.99, which is extremely challenging in itself. In addition, it is observed that the EFCEG-A+SVR and EFCEG-B+SVR models can maintain stable prediction performance under various experimental configurations, which further proves that these models not only have excellent robustness but also have strong generalization ability. Through a series of carefully designed experimental configurations, it is fully demonstrated that the EFCEG-A+SVR and EFCEG-B+SVR regression models not only have excellent prediction performance, but also their core ideas have wide applicability. When encountering situations similar to the research content of the present invention, it can be considered to combine this method with other regression models to fundamentally improve the prediction ability of the model and learn more useful knowledge.

[0157] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity, characterized in that: The steps include: S1. Collect the health characteristic data set of lithium battery charge and discharge cycle, and cluster the data set to obtain basic clustering result data; S2, based on the base clustering result data, the samples are represented by dummy variables and scaled; S3, regard the sample points that are always in the same cluster in the base clustering result as a potential equivalent granularity, construct the equivalent point set of the data set, regard the first sample point in each subset as the potential equivalent granularity of the subset, and construct the potential equivalent granularity set; S4, dividing the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set; S5, measure the fuzzy relationship between any two core equivalent granularities, construct a relationship matrix composed of core equivalent granularities, and identify the subject cluster structure based on the relationship matrix; S6. Use the fuzzy relationship to assign the fuzzy equivalent granularity to the obtained main cluster structure; at the same time, restore the labels of the hidden sample points in the equivalent point set to the labels of the potential equivalent granularity equivalent to it, and obtain the final consensus result composed of all sample points; S7. predicting the battery health status based on the obtained consensus result; S1 includes the following processes: Collect the lithium battery charge and discharge cycle health feature data set DX = {x1, x2, ..., x j , …, x N }, where N represents the number of sample points in the data set DX. The data set DX is divided L times to obtain a set of basic clustering results Π={π 1 , π 2 ,…,π L },in represents the i-th base clustering result in Π, is the i-th base clustering result π i The lth cluster, P i is the i-th base clustering result π i The number of clusters, the set of all clusters in the set of base clustering results Π is Where P C is the total number of all clusters in π, and the set π is also expressed in the form of the base clustering result matrix, that is, π = [π ij ] L×N , where π ij Represents the cluster label assigned by the i-th partition to the j-th sample point; Define the co-correlation matrix A = [a jk ] N×N for: in, S2 includes the following processes: Based on the base clustering result matrix Π=[π ij ] L×N , for any sample point x in the data set DX j , construct its dummy variable feature representation dx j for: in, The length of is the base clustering result π under the i-th partition i The number of clusters P i , from left to right, each element represents the 1st cluster to the Pth cluster. i Cluster, based on the DBI index, the dummy variable feature of the sample point is represented by dx j Upgraded to scaled dummy variable feature representation d * x j : Among them, the scaling factor Used to measure the contribution of the i-th base clustering result, DBI i It is the DBI index value score of the base clustering result under the i-th partition.

2. The lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity according to claim 1 is characterized in that: The sparse partitioning strategy based on third-party random points in S4 divides the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set, including the following process: S4-1. Randomly generate a point x that is equal to any sample point x in the data set DX. j A third-party sample point x with the same feature dimension g , calculate all sample points to the random point x g The distance between them generates a distance sequence Reconstruct the sequence DX according to the ascending order of the distance values ​​in D corresponding to the sample points in DX # ={d # x1,d # x2,…,d # x N }, in sequence DX # Calculate the distance between the "current target point" and the "neighboring point"; S4-2, "neighboring points" are sample points within the set iteration step range, and iteration is performed until any sample point d is obtained. # x j Its "neighboring point" d # x m The distance is obtained by sparse distance matrix based on Measuring any sample point x j The local density ρ j : in, Represents a sparse distance matrix The number of effective distances in the jth row of θ is used to ensure that the denominator is always meaningful; a larger density sample set with a density higher than the local density mean is obtained. S4-3, according to the potential equivalent granularity set DX * , equivalent point set E and larger density sample set ρ bigger , the potential equivalent granularity set DX * Divided into core equivalent granularity set CS and fuzzy equivalent granularity set FS: CS={d * x r |r bigger ∩E r ≠φ,r=1,2,…,|E|} (7) FS=DX * -CS (8) Among them, E r is the rth subset element of E, |E| represents the number of subset elements in the equivalent point set E, d * x r For DX * The potential equivalent granularity corresponding to the rth position in .

3. The lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity according to claim 2 is characterized in that: S4-1 calculates all sample points to random points x g The distance between them is:

4. The lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity according to claim 1 is characterized in that: The sparse partitioning strategy based on equivalent point sets in S4 divides the potential equivalent granularity set into a core equivalent granularity set and a fuzzy equivalent granularity set, including the following process: According to the equivalent point set E, calculate the sparse distance matrix of DX in, For any sample point d * x j With d * x j Any other sample point d in the subset element of E * x m The distance between any sample point d * x j The local density ρ j for: The division of potential equivalent granularity includes two cases: (1) If Then, the potential equivalent granularity set is divided into the core equivalent granularity set CS and the fuzzy equivalent granularity set FS by equations (7) and (8); (2) If Then the non-zero density set ρ non-zero ={d * x j |ρ j ≠0, j∈[1,N]}, then: CS={d * x r |r non-zero ∩E r ≠φ,r=1,2,…,|E|} (11) Among them, d * x r For DX * The potential equivalent granularity corresponding to the rth position in the fuzzy equivalent granularity set FS is DX * -CS.

5. The lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity according to claim 2 or claim 4 is characterized in that: S5 includes the following steps: According to the core equivalent granularity set CS, the γ-fuzzy operator is used Measuring the equivalent granularity of any two cores * x j and d * x k The fuzzy relationship between jk : Among them, |d * x j +d * x k | represents the 2-norm of the sum of two vectors, d * x j d * x k represents the inner product of two vectors, γ-Fuzzy Operator is defined as: Let the mapping S: [0, 1] 2 →[0, 1], for γ∈[-1, 0], there is a γ-fuzzy operator Construct the relationship matrix F composed of all core equivalent granularities = [f jk ] |CS1×|Cs| , where |CS| represents the number of elements in the core equivalent granularity set CS; Based on the relationship matrix F, the DPC model is used to give the main cluster structure result of the data set π θ .

6. The lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity according to claim 5 is characterized in that: S6 includes the following processes: First, calculate π θ Any main cluster C i The mean vector V i : Where k is π θ The number of clusters in ; Then, using the γ-fuzzy operator To evaluate the jth fuzzy equivalent granularity d * x j and the i-th mean vector V i The fuzzy relationship between: Among them, |FS| is the number of fuzzy equivalent granularities in FS; Finally, the maximum membership principle is used to allocate fuzzy equivalent granularity, that is, any fuzzy equivalent granularity d * x j Assigned to the known primary cluster with which it has the largest membership relationship; The labels of the hidden sample points in the equivalent point set E are restored to the labels of the equivalent potential granularity, and the final consensus result π composed of all sample points is obtained. * .

7. The lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity according to claim 6 is characterized in that: The regression model SVR is used in S7 to predict the battery health status.

8. A lithium battery health status estimation system based on efficient fuzzy clustering integration of equivalent granularity, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 7, and executes the steps of the lithium battery health status estimation method based on efficient fuzzy clustering integration of equivalent granularity when running.

Images