Evolution mode identification method and system based on time sequence tensor segmentation

Abstract

The invention provides an evolution mode recognition method and system based on time sequence tensor segmentation, and the method comprises the steps: collecting three-order time sequence tensor data, and taking the three-order time sequence tensor data as input data; dividing the third-order time sequence tensor into a plurality of tensor segments through a sliding window; extracting a core tensor of the tensor segment based on Tuck decomposition, and performing clustering based on a time sequence weighted clustering method to obtain a coarse-grained segmentation result; optimizing the coarse granularity segmentation result based on information gain to obtain a fine granularity segmentation result; extracting correlation characteristics of each fragment based on the maximum information coefficient; and introducing a neighbor rule, and identifying an evolution mode corresponding to each fragment by adopting an unsupervised three-way clustering algorithm. The invention further constructs an evolution mode recognition system based on time sequence tensor segmentation, a data acquisition part, a time sequence tensor segmentation and evolution mode mining part and a monitoring APP part are connected with one another to form a complete system, and effective recognition of segmentation of the time sequence tensor and the evolution mode is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of time series data pattern recognition, and in particular to an evolutionary pattern recognition method and system based on time series tensor segmentation. Background Art

[0002] In today's digital age, the generation and collection of data has shown explosive growth, among which time series data is particularly important due to its wide application in multiple fields. As a complex data structure, the third-order time series tensor usually contains multiple time series, which are collected from multiple locations (such as users, devices, etc.) through various types of sensors. The sequences collected by these users or locations contain rich potential patterns, which evolve alternately over time and exhibit complex and changeable dynamic characteristics. For example, in industrial production, the operating status data of equipment can help predict faults and optimize production processes. Therefore, accurately extracting the pattern information contained in the time series tensor is of vital importance for understanding the inherent laws behind the data, making effective decision support, and realizing intelligent system management.

[0003] To deeply explore the pattern information in time series tensors, it is usually necessary to preprocess the data by partitioning them into time subintervals. Existing time series segmentation methods can be divided into three main categories: morphology-based methods, clustering-based methods, and statistical optimization-based methods. These methods have made significant progress in time series segmentation. However, most existing time series tensor segmentation methods focus on univariate time series, and few algorithms are specifically designed for multivariate time series tensors. In addition, most segmentation methods require manual presetting of the number of segments, and the parameter tuning process increases computational complexity and reduces the algorithm's portability. Furthermore, the numerical values of the segmentation points obtained by sliding window segmentation methods are always integer multiples of the window size, resulting in deviations from the actual results. Moreover, current tensor segmentation algorithms rarely start from the tensor structure itself, ignoring the interactions between the three different dimensions of time, variable, and location.

[0004] In the field of time series data analysis, clustering is a method used to identify potential similarities between individuals and divide them into clusters based on historical data. It can effectively reflect the differences and similarities between large amounts of data. Current clustering methods for time series are significantly impacted by the number of attribute variables in the sequence. When the number of attribute variables is large and the time span is long, computational efficiency decreases significantly. Furthermore, time series tensors contain multiple dimensions, such as time, geography, and variables. Traditional two-way distance methods cannot effectively identify fragmented patterns in tensors, necessitating three-way clustering.

[0005] Another area of focus in clustering algorithms is determining the number of clusters. Typically, the number of clusters is unknown during clustering, a phenomenon known as the cluster numbering problem. This problem requires prior knowledge of the changes occurring in the study area, or it can be defined by human experts through a supervised process. When prior knowledge is unclear or expert experience is lacking, an effective approach is to determine the final number of clusters based on the clustering results for different numbers of clusters. This method is simple and effective, but it requires repeated experiments and calculations to find a good score, consuming significant computing resources. Summary of the Invention

[0006] In response to the above problems, the purpose of the present invention is to provide an evolutionary pattern recognition method and system based on time series tensor segmentation, which utilizes a time series tensor parameter adaptive segmentation algorithm based on time series weighted clustering and a time series unsupervised three-way clustering algorithm based on correlation feature extraction to achieve effective recognition of the segmentation and evolutionary patterns of time series tensors.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In one aspect, a method for recognizing an evolutionary pattern based on time series tensor segmentation is provided, the method comprising the following steps:

[0009] S1, collect third-order time series tensor data as input data;

[0010] S2. Divide the third-order time series tensor into multiple tensor segments using a sliding window;

[0011] S3. extracting the core tensor of the tensor segment based on Tucker decomposition, and clustering it based on a time series weighted clustering method to obtain a coarse-grained segmentation result;

[0012] S4. Optimizing the coarse-grained segmentation result based on information gain to obtain a fine-grained segmentation result;

[0013] S5. extracting correlation features of each segment based on the maximum information coefficient;

[0014] S6. Introduce the nearest neighbor rule and use the unsupervised three-way clustering algorithm to identify the evolutionary patterns corresponding to each fragment.

[0015] On the other hand, an evolutionary pattern recognition system based on time series tensor segmentation is provided, for implementing any of the above methods, the system comprising:

[0016] A data acquisition module is used to collect third-order time series tensor data as input data;

[0017] A time series tensor segmentation module is configured to divide the third-order time series tensor into multiple tensor segments using a sliding window; extract the core tensors of the tensor segments based on Tucker decomposition, and cluster them based on a time series weighted clustering method to obtain a coarse-grained segmentation result; and optimize the coarse-grained segmentation result based on information gain to obtain a fine-grained segmentation result;

[0018] The evolutionary pattern recognition module is used to extract the correlation of each fragment based on the maximum information coefficient; the nearest neighbor rule is introduced and an unsupervised three-way clustering algorithm is used to identify the evolutionary pattern corresponding to each fragment.

[0019] In addition, the system also includes a monitoring APP.

[0020] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0021] In an embodiment of the present invention, firstly, a time series tensor parameter adaptive segmentation algorithm based on time series weighted clustering (PASTWC) is used to cluster the tensor segments extracted by Tucker decomposition through a sliding window to obtain a coarse-grained classification result, and then fine-grained segmentation is achieved through information gain optimization; then, an unsupervised three-way clustering of time series based on correlation feature representation (UTCFTS) is used to realize the evolutionary pattern recognition of the segmented fragments. The method of the present invention can not only perform accurate three-way clustering on multivariate time series data, but also effectively identify potential evolutionary patterns. The experimental results on multiple artificial data sets and real data sets show that the algorithm proposed in the present invention has superior performance and can effectively complete the segmentation of time series tensors and the recognition of evolutionary patterns.

[0022] This invention utilizes a data acquisition module to better capture the monitoring data required for mining and builds a monitoring app for real-time monitoring. The data acquisition module, the evolutionary pattern recognition method based on time series tensor segmentation, and the monitoring app are interconnected to form a complete system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 This is a flow chart of an evolutionary pattern recognition method based on time series tensor segmentation provided by an embodiment of the present invention;

[0025] Figure 2 Schematic diagram of the coarse-grained segmentation process provided by an embodiment of the present invention;

[0026] Figure 3 Schematic diagram of information gain and objective function provided by an embodiment of the present invention;

[0027] Figure 4 Schematic diagram of the time series weighted clustering process provided by an embodiment of the present invention;

[0028] Figure 5 is a schematic diagram of fine-grained segmentation provided by an embodiment of the present invention;

[0029] Figure 6 This is a schematic diagram of time series segment pattern recognition based on unsupervised three-way clustering provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] First, the basic theories used in the embodiments of the present invention are explained as follows:

[0032] (1) Loss function based on information entropy

[0033] In information theory, Shannon entropy (H) quantifies the uncertainty in a probability distribution, reflecting how much information can be obtained from a given quantity. Higher entropy indicates greater uncertainty, meaning that less information can be inferred about the quantity of interest.

[0034]

[0035] Where p(n) represents the probability of n-1 outcomes, and H(·) represents the entropy of the probability distribution. Given a segmentation of a time series, a loss function l based on information entropy is introduced, which measures the reduction in entropy when the time series is segmented according to a specific segmentation method. More specifically, l reflects the expected reduction in entropy when the time series X is further segmented, and is defined as follows:

[0036]

[0037] Where K+1 is the number of segments, |X k | is the length of the kth segment. X is a segment of the entire time series and H(X) is the entropy of the entire time series.

[0038] (2) Hierarchical clustering algorithm TW-FINCH

[0039] TW-FINCH is a hierarchical clustering algorithm for grouping semantically consistent graphic frames in videos. It aims to infer the boundaries of semantically consistent visual concepts in videos to achieve action segmentation in videos. It builds a graph where nodes represent members, p and q in the formula, and each edge weight W(p,q) represents the semantic proximity between two members:

[0040]

[0041] W(p,q)=Ω f (p,q)·Ω t (p,q) (4)

[0042] where Ω f (p,q) is the spatial adjacency matrix, Ω t (p,q) is the temporal adjacency matrix. TW-FINCH merges semantically close members and repeats this process until the hierarchy of partitions is completed, where different partitions are different action segments in the video.

[0043] (3) Maximum information coefficient

[0044] Mutual information refers to the degree to which the information content of one random variable decreases after the other variable is known. The greater the correlation between the two variables, the greater the mutual information value. When the two variables are independent, the mutual information value is 0. The main idea of the Maximum Information Coefficient (MIC) is to assume that there is a certain correlation between the two variables. Then, a method is used to divide the scatter plot of the joint samples of the two variables into a grid, so that most points fall into a few cells. This solves the problem of calculating the joint probability in the mutual information.

[0045] Definition 1: Given a finite ordered pair dataset Positive integers a, b and sample size num, let the data set be distributed in the coordinate system, use a, b to divide the data set D into an a*b grid G, and the variable values in D fall into the grid to obtain the corresponding probability distribution D| G By changing the grid division, different mutual information values will be obtained, among which the maximum mutual information value is as follows:

[0046] I * (D,a,b)=maxI(D| G ) (5)

[0047] Definition 2: For the convenience of comparison, the maximum mutual information value is normalized so that its value is between [0, 1]:

[0048]

[0049] Definition 3: Assuming that the number of ordered pairs in the dataset D is num, the formula for defining the MIC of two variables A and B in the set is:

[0050]

[0051] Where B(num) = num ε Represents the upper limit of the grid division. The constant ε is set according to the experience value or scale. Too large a value will lead to false correlations, while too small a value will only detect simple correlations. Usually, when B(num)=num 0.6 The effect is best when

[0052] Unlike traditional multivariate time series, this invention aims to segment and identify evolutionary patterns in time series tensors. This can also be viewed as collaborative segmentation and pattern recognition for multiple multivariate time series. A time series tensor is essentially a multidimensional array that can meaningfully represent a wide range of spatiotemporal datasets. Table 1 lists the symbols used in this invention and their descriptions.

[0053] Table 1 Symbols and definitions

[0054]

[0055]

[0056] The embodiment of the present invention provides an evolution pattern recognition method based on time series tensor segmentation, such as Figure 1 As shown, the processing flow of this method includes the following steps:

[0057] S1. Collect third-order time series tensor data as input data.

[0058] The third-order time series tensor data is monitoring-type multivariate time series data collected by sensors, including data in the fields of information retrieval data, environmental monitoring data, human activity data, financial analysis data, building energy consumption data, etc.

[0059] S2. Divide the third-order time series tensor into multiple tensor segments through a sliding window.

[0060] The collected third-order time series tensor X∈R M×N×T The sampling results of N variables for M users / locations within T time periods are presented. Based on the changing trend of the third-order time series tensor, K segmentation points are found through a sliding window, and X is divided into K+1 tensor segments along the time dimension.

[0061] Specifically, X can be regarded as a stack of multiple multivariate time series of M users / locations, denoted as X = {X d(1) ,X d (2),...,X d(m) ,...,X d(M)},in:

[0062]

[0063] represents the measured value of the variable v(n) of the user / location d(m) at time t(i), where m=1,...,M; n=1,...,N; i=1,...,T. The purpose of the present invention is to adaptively find K segmentation points {b1,b2,...,b k ,...,b K} Divide X into K+1 segments along the time dimension, denoted as X={X1,X2,...,X k ,...,X K ,X K+1}, where b k is the kth segmentation point, and X k is the kth segment, k=1,...,K. After segmentation, segment X k The corresponding time interval is s k , and s k =[b k-1 ,b k ), length is |s k |=|b k -b k-1 |.

[0064] The time interval set obtained by segmentation is:

[0065] S={s1,s2,...,s k ,...,s K ,s K+1}={[b0,b1),[b1,b2),…,[b k-1 ,b k ),…,[b K-1 ,b K ),[b K ,b K+1 ]},in

[0066] 1=b0<b1<...<b k <...<b K <b K+1 =T. Segment X k It can be expressed as in:

[0067]

[0068] is the multivariate time series of user / location d(m) in the kth segment, is the nth variable v(n) at time t(b k-1 +h), m=1,2,...,M;n=1,2,...,N;h=1,2,...,(|s k |-1).

[0069] After the segmentation algorithm has collaboratively partitioned the time series tensor data into K+1 time subintervals, three-way clustering is required on the sequence segments corresponding to all time subintervals to obtain complete pattern information in the time series tensor. Clustering these subsequences from different users and locations requires dividing them into C spatial clusters under the dual constraints of spatial continuity and non-continuous spatial repetitive geographic pattern discovery, ensuring that the data within the cluster is as similar as possible and the data between clusters is as different as possible.

[0070] S3. Extract the core tensor of the tensor segment based on Tucker decomposition, and cluster it based on the time series weighted clustering method to obtain the coarse-grained segmentation result.

[0071] like Figure 2 As shown, Figure 2 In the figure, (a) represents the feature extraction process, (b) represents the segmentation based on time-series weighted clustering, and (c) represents the coarse-grained segmentation result of the time-series tensor. The tensor segments obtained by the sliding window are decomposed using Tucker to extract the corresponding core tensors. These core tensors are then clustered using a parameter-free time-series weighted method. During the clustering process, each core tensor is initially regarded as a separate cluster. In order to facilitate fast clustering, a spatiotemporal graph representation matrix is constructed, and the result of each clustering iteration effectively represents a segmentation result. At this time, the information gain of the current segmentation result is calculated as follows: Figure 3 As shown in (a) and (b). After several iterations, all individuals are merged into a single cluster. The segmentation result with the highest objective function value (indicated by the asterisk) is selected as the final segmentation result. Figure 2 In , the fourth iteration is selected as the final coarse-grained segmentation result, as shown in Figure 2 This clustering method is very efficient and usually only requires 3 to 6 iterations to obtain the final result.

[0072] Specifically, suppose the third-order time series tensor X is calculated according to the sliding window The time dimension is divided into J windows, represented as X = {X1, X2, ..., X j ,...,X J}. The jth window is X j ∈ M×N×W ,j∈1,...,J,there are is the time series of user / location d(m) in the jth window:

[0073]

[0074] in Represents the value of the user / location d(m) variable v(n) at the jth window u time, n=1,2,...,N; m=1,2,...,M; u=0,1,2,...,(W-1); j=1,2,...,J.

[0075] In order to reduce computational overhead and capture the geometric integrity of tensor data, the present invention uses Tucker decomposition to extract features from the obtained tensor segments. A third-order tensor X j ∈ M×N×W It can be approximately expressed as:

[0076]

[0077] A in formula (11) j ∈ M×P , B j ∈ N×Q , C j ∈ I×R is a tensor X j The factor matrices are of different dimensions and P≤M, Q≤N, R≤I. is the corresponding core tensor, which can be regarded as the dimensionality reduction feature of the original tensor, and each factor matrix can be regarded as the principal component in different dimensions. The optimization goal of the present invention is to minimize the Frobenius norm error between the original tensor segment and the reconstructed tensor segment given a predefined rank of the core tensor:

[0078]

[0079] HOSVD is used as an initialization step to obtain a rough estimate of the factor matrix and core tensor. Subsequently, HOOI iteratively refines these factor matrices to obtain a more accurate and efficient decomposition. This combination leverages the global decomposition capabilities of HOSVD and the local optimization capabilities of HOOI, thereby improving the accuracy of the decomposition while reducing computational complexity. In addition, to better capture the relationship between multivariate time series at different locations, instead of imposing orthogonal constraints on the time dimension, orthogonal constraints are imposed on the variable dimension and the user / location dimension.

[0080] In addition, the predetermined rank of the core tensor (i.e. The P, Q, R in will affect the decomposition effect and computational complexity. The ideal rank should ensure that the core tensor obtained is of appropriate size and can accurately reconstruct the original tensor. Therefore, in order to adaptively determine the jth tensor segment gj The ideal rank of the core tensor of is, and the optimization objective based on formula (12) is:

[0081]

[0082] In the embodiment of the present invention, a Bayesian optimization algorithm is used to determine the optimal tensor rank. When determining the rank of the core tensor, a weight α controls the trade-off between error and rank and is set to 0.0005.

[0083] Based on formula (13), we can get the tensor segment X j The ideal rank of the core tensor is , but the ideal rank of other tensor segments may be different. In order to determine the final ideal rank of the core tensor of all tensor segments, A window is randomly selected from the window and the optimal rank of its core tensor is calculated. The one with the highest frequency among the ranks is selected as the final core tensor rank of all tensor segments during decomposition.

[0084] For the convenience of description, we will start from each tensor segment {X1,X2,....,X j ,...,X J}The core tensor obtained is recorded as {g1,g2,....,g j ,...,g J}, where when the time dimension length t of the last interval divided by the sliding window is less than the rank R of the time dimension of the determined core tensor g, the last interval tensor X J With X J-1 After connecting in the time dimension, tensor decomposition is performed to obtain the core tensor. At this time, there are a total of J-1 core tensors. At this time, record J w = J-1; when the interval tensor X J With X J-1 When there is no need to merge to solve the core tensor, there are a total of J core tensors, record J w = J. Based on this, the core tensor obtained is recorded as

[0085] Furthermore, inspired by TW-FINCH, the present invention proposes to use a space-time graph representation matrix G to encode feature space and temporal proximity, with the time series as the modulation factor. The matrix is based on the similarity matrix Ω in feature space and space-time, respectively. f and Ω tConstruct. Through G, all core tensors are connected to their nearest neighbors and divided into clusters. The initial result divides the original tensor into several segments. Similar clusters are then repeatedly merged to obtain further continuous segmentations. Each step forms a group of previously obtained clusters, and the recursion terminates when only one cluster remains. During the iteration process, the result with the largest rate of change in information gain is selected as the final segmentation result.

[0086] Specifically, in the segmentation process, first define a matrix Node V is the extracted core tensor The edge E describes the proximity of the core tensors in the feature space and time dimensions. Here, assume that G(p,q) represents g p and g q The corresponding edges between g p , g q belong And 1≤p≤J W , 1≤q≤J W .

[0087] In order to obtain the space-time graph representation matrix G, we first construct a similarity matrix in the feature space The edge weight is calculated as follows:

[0088]

[0089] where dis(g p ,g q ) is g p and g q In order to adapt to the time weighting matrix later, Normalized in the range [0,1], expressed as

[0090]

[0091] Then the matrix Converted to the final spatial similarity matrix Ω f :

[0092]

[0093] Secondly, construct another time similarity matrix The edge weight is calculated as follows:

[0094]

[0095] Ω t (p,q) represents g p and g q The time difference between them. Timing diagram matrix Ω tis used as the modulation factor of the feature space map, where Ω t (p,q) provides a weighting mechanism. Finally, Ω f and Ω t Combined to get the time-series modulation graph matrix

[0096] Ω(p,q)=Ω f (p,q)·Ω t (p,q) (18)

[0097] Based on Ω, the final spatiotemporal graph representation matrix is obtained Encodes the feature space and time distances. For any node in Ω, its shortest edge is set to 1, and all its other edges are set to 0:

[0098]

[0099] Clusters are formed by iteratively connecting each node to its nearest neighbors through the spatiotemporal graph representation matrix G. For one iteration, a simple clustering result is Figure 4 As shown. The obtained space-time graph representation matrix G is as follows Figure 4 As shown in (a) in the figure, each row contains only a few elements with a value of 1, and the other elements are set to 0, which means that the shortest edge of each node is set to 1, and all other edges are set to 0. Figure 4 (b) illustrates multiple clusters by connecting all nodes to their nearest neighbors. For example, nodes 10, 11, and 12 are grouped together in one cluster because they are closest to each other. Each node represents a core tensor extracted from a tensor segment, so the clustering result can be mapped back to the original tensor for segmentation. Specifically, each tensor segment has a cluster label, so segmentation can be achieved by connecting tensor segments with the same label, such as Figure 4 As shown in (c).

[0100] After the first iteration, the mean of all clusters obtained is subjected to modified TW-FINCH clustering. Each iterative clustering step actually corresponds to a segmentation result. The process terminates when only one cluster remains. Finally, the segmentation result with the largest rate of change in information gain is selected as the final coarse-grained segmentation:

[0101]

[0102]

[0103] where l (e) is the information gain of the e-th iteration. The k-th segment of the e-th iteration can be expressed as: The corresponding time interval is s (e) (k). yes User / Place d (m) The probability of the nth variable appearing. Based on formula (20), the coarse-grained segmentation point set can be obtained

[0104] S4. Optimize the coarse-grained segmentation results based on information gain (information entropy) to obtain fine-grained segmentation results.

[0105] Since the segmentation results after clustering are based on a sliding window, the position of each segmentation point is a multiple of the window size and needs to be optimized within its left and right windows.

[0106] In order to obtain fine-grained segmentation results, the coarse-grained segmentation point set Is further optimized. For the nth segment point The two adjacent windows on the left and right are merged into one interval The tensor segments contained in this time interval are represented as: Optimized segmentation point b k It is obtained by maximizing the information gain l(k):

[0107]

[0108] Where l(k) is the segment point b k The information gain when b k The tensor It is divided into two parts, the left one is The one on the right is They are X kl and X kr User / Place d (m) The probability of the nth variable appearing. Based on formula (24), we can get the set of fine-grained segmentation points {b1, b2, ..., b k ,...,b K}.

[0109] Figure 5 It is the process of optimizing the segmentation points. Figure 5 In (a), in order to optimize the coarse-grained segmentation point b1, the time interval of the two windows around point b1 is selected, with a total length of 2W. Then a greedy search is used. Figure 5In (b), assuming that the interval contains 12 time points, each time point is used as a segmentation point to calculate the corresponding information gain. Since the number of segmentation points is fixed, the point with the largest information gain is the optimal choice. Initially, the segmentation point is point b1, that is, the sixth point is selected. However, after fine-grained segmentation, the third point is selected as the new segmentation point because it produces the highest information gain. Finally, the fine-grained segmentation result is as follows Figure 5 As shown in (c).

[0110] The overall process of the time series tensor parameter adaptive segmentation algorithm based on time series weighted clustering (PASTWC) provided by the present invention is shown in Table 2.

[0111] Table 2 PASTWC algorithm process

[0112]

[0113] Next, the Unsupervised Three-way Clustering of Time Series (UTCFTS) algorithm based on correlation feature extraction is used to effectively identify the evolution pattern of the segmented fragments. The overall process is as follows: Figure 6 As shown, features of each fragment are first extracted based on correlation, and then unsupervised clustering is performed on the extracted features to identify the pattern category corresponding to each fragment.

[0114] S5. Extract the correlation of each segment based on the maximum information coefficient.

[0115] Among them, correlation includes variable correlation, time correlation, and user / location correlation.

[0116] (1) Variable correlation extraction

[0117] For location d(m), there are K+1 segments, and (K+1)*M subsequence segments need to be clustered.

[0118] Among them, the kth segment of location d(m) can be expressed as:

[0119]

[0120] where b k-1 and b k are two adjacent segment points, Containing N different variables, such a multivariate time series segment can be regarded as a subsequence segment containing multiple single variables, containing the interactive information of the time dimension and the variable dimension. Considering that the covariance coefficient is only sensitive to linear relationships, it is difficult to describe the large number of nonlinear relationships between variables. MIC is used to measure the correlation between different variables. The maximum mutual information coefficient matrix of this multivariate time series is: It can be calculated by the following formula:

[0121]

[0122] yes The variables n and The maximum mutual information coefficient between is a symmetric matrix, that is The elements of the upper triangular matrix except the diagonal elements completely contain the relationship between any two different variables. Therefore, the internal correlation between any two different variables can be expressed as the maximum mutual information sequence To describe, L=(N-1)N / 2. We can know The length of depends on the number of variables N, not the length of the subsequence segments. That is, two multivariate time series of different lengths but with the same number of variables, after variable correlation extraction, will have sequences with the same maximum mutual information coefficient. This allows for the use of fast metrics such as Euclidean distance to measure the similarity between different subsequence segments.

[0123] (2) Time correlation extraction

[0124] According to formulas (5)(6)(7)(29)(30), multivariate time series of different lengths can also be converted into maximum mutual information coefficient sequences of the same length in this way, so that they can be compared and analyzed under the same standard. After constructing the covariance sequence of each multivariate time series (MTS), principal component analysis (PCA) can be used to analyze the composition of these maximum mutual information coefficient sequences, thereby extracting features that can represent the original multivariate time series.

[0125] For the multivariate time series dataset obtained previously Each subsequence Can be converted into the corresponding maximum mutual information coefficient sequence In this way, the entire data set can be represented as a multivariate maximum mutual information coefficient time series

[0126] Next, PCA analysis is performed on the maximum mutual information coefficient to obtain the eigenvector matrix V C and eigenvalue Λ C , and the principal component Y C . Eigenvalue Λ C Indicates the contribution of each eigenvector to the variance of the data set, and the eigenvector V C It indicates the main direction of change of the data.

[0127] In PCA, the eigenvectors corresponding to the first λ largest eigenvalues are usually selected to form a new feature space, which can retain the main information of the data. For the convenience of representation, suppose there is a time series X∈R n*m , through PCA, the original multivariate time series X∈R n*m Transformed into a new multivariate time series Y∈R n*λ , its dimension is reduced from n*m to n*λ, where λ is the number of principal components selected to be retained. The transformed sequence Y can be expressed as follows:

[0128] Y n×λ =X n×m V m×λ (31)

[0129] Applied to the algorithm of the present invention, after the feature extraction of the multivariate maximum mutual information coefficient time series, each original multivariate time series is converted into a feature vector, which consists of the first λ rows of the V matrix, representing the Finally, the entire data set D is converted into a feature matrix F, whose dimension is U*λ, where U is the number of sequences in the data set, U=M*(K+1). In the process of feature extraction of the present invention, the covariance matrix Σ C Perform principal component analysis to obtain the eigenvector matrix V C and eigenvalue Λ C . Eigenvector matrix V C Each column of represents an eigenvector, which defines the new eigenspace. C It represents the variance contribution of the data in the direction of each eigenvector, that is, the amount of information.

[0130] In order to select the most important eigenvectors, we usually sort them according to the size of the eigenvalues and select the eigenvectors corresponding to the first lambda largest eigenvalues. These most important eigenvectors can best represent the information of the original data. For each multivariate time subseries in the data set The corresponding maximum mutual information coefficient sequence can be projected into the new feature space composed of these λ most important eigenvectors.

[0131]

[0132] Here V C (:,1:nk) represents the matrix consisting of the first λ columns selected from VC. Through this formula, each multivariate time series is converted into a λ-dimensional feature vector Finally, the dataset D containing U subsequences can be represented by a feature matrix F through dimensionality reduction, which contains each multivariate time series in the dataset The eigenvector of

[0133]

[0134] In this process, each multivariate time series is effectively represented as a λ-dimensional feature vector, with dimensions from (b k -b k-1 )×N is reduced to 1×λ, which is usually much smaller than the original dimension.

[0135] (3) User / location relevance

[0136] Since multiple time series come from different users or locations, the impact of time series data between different locations also needs to be considered. Data is usually positively correlated only within a certain spatial / location range. Therefore, sequences with close geographical locations will have a greater correlation. For different users, sequences collected from users with similar behaviors will be closer. Therefore, when geographic coordinates are missing, the degree of sequence similarity can be used to determine the proximity of users and spatial dimensions. Specifically, for the complete time series X collected from different locations or users in the time series tensor, d(m) and For variable n, the Maximum Information Coefficient (MIC) is used for measurement, which can find the degree of correlation between the two time series on variable n without assuming a linear relationship, where n = 1, ..., N.

[0137]

[0138] because For a multivariate time series, we can define the overall user or location m, The degree of correlation between them is:

[0139]

[0140] According to formula (35), for the time series X of the mth location d(m) You can get an M-dimensional feature To reflect the association relationship of different time series in the location or user dimension.

[0141] S6. Introduce the nearest neighbor rule and use the unsupervised three-way clustering algorithm to identify the evolutionary patterns corresponding to each fragment.

[0142] For a long time, clustering-based pattern recognition has generally been affected by initialization and requires a priori given number of clusters. In order to construct a clustering algorithm that does not require initialization and automatically identifies the number of pattern types, the present invention uses the concept of entropy. First, let the proportional coefficient α c is the probability that a sample belongs to the cth class (the input here is the sample feature, hereinafter referred to as the sample. From the above, we know that there are a total of U = (K + 1) * M samples. For the sake of convenience, we will refer to them as U samples below), c = 1, ..., C. Then Becomes the average value of the information, also the proportional coefficient α c The corresponding entropy value.

[0143] When α c = 1 / C, the entropy reaches its maximum. Therefore, this term is added to the objective function J(z,A) of the k-means algorithm as a penalty. Then, a model is constructed to estimate α by minimizing the entropy. c , to obtain α c The maximum information. Minimize Equivalent to maximization Therefore, a new objective function is proposed as follows:

[0144]

[0145] To determine the number of clusters, we next consider another entropy term and let the membership z uc is the membership degree of the u-th sample to the c-th class. uc and the proportionality factor α c Combined to get the term z uc lnα c The overall objective function is as follows:

[0146]

[0147] The Lagrangian of the equation is:

[0148]

[0149] Membership degree z uc The update equation is as follows:

[0150]

[0151] Cluster center a c The update equation is as follows:

[0152]

[0153] From (39), we can see that the membership value is 1 or 0, so we can take z uc xu To represent the samples in the cth cluster, the update of the cluster center in formula (40) is not just a simple selection of the cluster average value, but the selection of the cluster average value and samples that are closest to other samples as the new cluster center.

[0154] By comparing the proportionality factor α c By finding the partial derivative of the Lagrangian function, we can get the update equation at the t+1th iteration as follows:

[0155]

[0156] For the cth mixing ratio currently being updated All categories need to be considered Therefore, the weighted average sum of all proportional coefficients is introduced into Equation (41) This ensures that the algorithm is updated When , the distribution of all categories in the entire data set can be taken into account, thus more accurately reflecting the probability of each data point belonging to each category. if Less than the sum of the weighted averages of all proportional coefficients The new mixing ratio That is, in the next iteration, the smaller ratio will decrease and the larger ratio will increase, and then a race will occur.

[0157] For some clusters in the iterative process 1<k≤C (t) For example, if α c <0 or α c ≤1 / U, it is considered an illegal proportional coefficient, and the clusters corresponding to these coefficients are discarded, and the number of clusters C of the current iteration number is increased. (t) Updated to

[0158]

[0159] Where |{}| represents the cardinality of the set {}. After updating the number of clusters c, the remaining mixing ratio and the corresponding Need to be renormalized to:

[0160]

[0161] Next, in formula (37) and The γ and β parameters of the two items are updated. The parameter update equations of γ and β are:

[0162]

[0163]

[0164] The algorithm can automatically reduce the number of clusters according to the parameter γ and obtain the corresponding parameter estimate. In addition, if and If the difference between is small, β must be larger to enhance its competitiveness. and If the difference between them is large, β will become smaller to maintain stability. Since β can jump at any time, when the number of clusters C is stable, let β = 0, which means that C will no longer decrease.

[0165] In order to speed up the clustering speed and clustering accuracy, the nearest neighbor rule is introduced to accelerate the convergence of clusters. In the first iteration, all sample points will be used as a cluster center. In the merging process, according to the calculation results of formula (35), for each sample x u ,u=1,...,U, that is choose The sequence X of locations m',m'∈[1,K] with a maximum mutual information coefficient greater than 2 / 3 of the other locations d(m') The τ fragments smaller than a given threshold in the τ fragments constitute the neighbor set

[0166] According to the number of neighbors of each sample, the spatial points are sorted in descending order and a list O is created. The point with the most neighbors in list O is selected as the first cluster center, and all its neighbors are assigned to this cluster. For subsequent points in list O, if a point does not belong to any cluster and more than 50% of its spatiotemporal neighbors also do not belong to any cluster, then the point is considered the center of a new cluster.

[0167]

[0168] This method is used to speed up the first iteration. In addition, after each iteration, for each sample x u ,Right now According to the MIC coefficient obtained in formula (35), find the original location or user time series X to which it belongs d(m) Closest location and user timing The closest fragment Make a judgment, i,j∈1,...,K+1. Assume Belongs to cluster c in the current iteration i , the cluster center is a i , Belongs to cluster c in the current iteration j , the cluster center is a j :

[0169]

[0170] In this way, it is equivalent to optimizing the clusters according to the correspondence between the closest fragments of different users or locations after each iteration, taking the association information of different locations or users into consideration.

[0171] The overall process of the unsupervised three-way clustering algorithm for time series based on correlation feature extraction (UTCFTS) is shown in Table 3.

[0172] Table 3 UTCFTS algorithm flow

[0173]

[0174]

[0175] In summary, the present invention addresses the challenges existing in the prior art and proposes a parameter adaptive segmentation algorithm based on temporal-weighted clustering for time series tensor (PASTWC). The algorithm first divides the third-order time series tensor into multiple tensor segments through a sliding window. Then, combining Bayesian search and Tucker decomposition, the tensor segments obtained by the sliding window are decomposed into core tensors as input. This feature extraction method based on data compression can capture the potential inherent structure of the tensor in a low-dimensional form. Then, coarse-grained segmentation is achieved based on the improved temporal-weighted clustering algorithm (TW-FINCH), without the need to manually set the number of segments. Considering that the segmentation result is only an integer multiple of the sliding window size, PASTWC achieves fine-grained optimization within the neighborhood tensor segment intervals on both sides of the coarse-grained segmentation point based on the idea of information gain.

[0176] In view of the many shortcomings of existing clustering methods, the present invention proposes an unsupervised tri-clustering algorithm for timeseries based on correlation feature representation (UTCFTS) to identify potential time series patterns in time series tensors. First, the correlation characteristics of the variable dimensions and the time dimension in different multivariate time series are extracted. In order to measure the nonlinear relationship between different variables, the maximum information criterion (MIC) is introduced to replace the original covariance coefficient. The extracted features are combined with the Uk-means algorithm, and the improved MIC is used to set spatial rules to assist in the three-way clustering, so that accurate three-way clustering can be performed on multivariate time series data, and potential evolution patterns can be effectively identified.

[0177] The main contributions of the present invention are as follows: (1) Starting from the tensor structure itself, an adaptive time series tensor segmentation algorithm is proposed, which realizes the collaborative segmentation of the time series tensor and does not require any manual setting of parameters during the segmentation process; (2) Considering that the segmentation point position based on the sliding window segmentation is only an integer multiple of the sliding window size, the segmentation point position is carefully optimized based on the information entropy, thereby realizing fine-grained segmentation of the time series tensor; (3) A correlation-based feature extraction method is used to comprehensively extract the features of the variable dimension and time dimension in the sub-segments obtained by segmentation, and the high-dimensional, redundant multivariate time series is converted into a simple vector representation, thereby reducing the computational cost of clustering and improving the execution efficiency; (4) An improved MIC is used to judge the correlation of different user or different location dimensions, which can effectively solve the spatial dimension information measurement under the lack of geographic information, and based on this, spatial rules are constructed to assist in the realization of three-way clustering; (5) The existing advantages of the Uk-means algorithm are retained, the extracted features are used as input, irrelevant features are eliminated, and the optimal number of clusters is automatically found to realize the recognition of evolutionary patterns.

[0178] Accordingly, an embodiment of the present invention further provides an evolutionary pattern recognition system based on time series tensor segmentation, the system comprising:

[0179] A data acquisition module is used to collect third-order time series tensor data as input data;

[0180] A time series tensor segmentation module is configured to divide the third-order time series tensor into multiple tensor segments using a sliding window; extract the core tensors of the tensor segments based on Tucker decomposition, and cluster them based on a time series weighted clustering method to obtain a coarse-grained segmentation result; and optimize the coarse-grained segmentation result based on information gain to obtain a fine-grained segmentation result;

[0181] The evolutionary pattern recognition module is used to extract the correlation of each fragment based on the maximum information coefficient; the nearest neighbor rule is introduced and an unsupervised three-way clustering algorithm is used to identify the evolutionary pattern corresponding to each fragment.

[0182] The system of this embodiment can be used to perform Figure 1 The technical solution of the method embodiment shown has similar implementation principles and technical effects, which will not be repeated here.

[0183] Furthermore, the data acquisition module includes a sensor component, a serial port server, a management workstation, an interface server, an information center storage server, and a platform core switch.

[0184] Among them, the sensor component is connected to the management workstation through the serial port server, the management workstation is connected to the interface server, the interface server is connected to the information center storage server through a one-way isolation gateway, the information center storage server is connected to the platform core switch through optical fiber, and the platform core switch is connected to multiple cloud computing nodes and web servers.

[0185] Furthermore, the system also includes a monitoring APP, which includes: a client, a server and a system management background.

[0186] The client is used for user registration and login, online query, modification and logout; the server is used for registration and login verification as well as data transmission, addition, modification and deletion; the system management background is used to manage the database.

[0187] The evolutionary pattern recognition system based on time series tensor segmentation constructed in the present invention consists of three main parts: data acquisition, time series tensor segmentation and evolutionary pattern mining, and monitoring APP, which are interconnected to form a complete system, providing a useful reference for research and development in related fields.

[0188] In the experimental verification part, the present invention first introduces the evaluation indicators of segmentation accuracy, then uses multiple artificial synthetic data sets with real segmentation information (ground truth) and real time series tensor data sets to verify the segmentation accuracy of the proposed method, and finally uses real multivariate spatiotemporal data sets to evaluate the effectiveness of the collaborative segmentation of the proposed method.

[0189] The embodiments of the present invention were validated using multiple time series tensor datasets. The experimental data included multivariate time series datasets from various industries, including information retrieval, environmental monitoring, human activity recognition, financial analysis, and building energy consumption. Detailed information about the experimental data is shown in Table 4.

[0190] Table 4 Experimental dataset

[0191]

[0192]

[0193] In the process of evaluating the accuracy of segmentation results, choosing appropriate evaluation indicators is particularly important. Here, two indicators will be used to evaluate the segmentation algorithm.

[0194] (1)F1. Assume that the segmentation point set obtained by the experiment is {b k |k=1,2,...,K}, the set of true segmentation points is When the dataset used to evaluate the segmentation method itself has real segmentation information, the more widely used evaluation indicator is F1:

[0195]

[0196] Directly using F1 as the evaluation index will tend to penalize reasonable segmentation solutions close to the true segmentation points. In order to solve this problem, the present invention defines the true segmentation point interval set For each segment position, where is the center, 0.025T is the radius, and T is the length of the time series. If the segmentation point set obtained by the experiment is {b k |k=1,2,...,K} falls on If the true segmentation position interval does not contain any segmentation position obtained by the algorithm, it is marked as a false negative (FN).

[0197] (2) Average p-lag. In some special cases, such as when the experimental segmentation points obtained by the two segmentation methods to be evaluated fall within the reasonable segmentation range at the same time, F1 will no longer be able to judge the pros and cons of the two. To solve this problem, we mainly judge the accuracy of the segmentation results by calculating the average absolute value distance between the experimental segmentation point position and its corresponding nearest true segmentation point position. In addition, in order to have an intuitive understanding of the performance of the algorithm on different data sets, the average absolute value distance is divided by the effective radius to obtain the average p-lag:

[0198]

[0199] Where GK is the number of segmentation points marked as true positives (TP). The lower the average p-lag, the more accurate the segmentation result. In this way, the average p-lag can complement F1 to evaluate the accuracy and effectiveness of the segmentation result.

[0200] In order to verify the effectiveness of the proposed algorithm, the algorithm proposed in this paper is compared with various time series segmentation algorithms, including:

[0201] (1) DMM: A temporal segmentation algorithm based on minimum coding loss.

[0202] (2) TTW-FINCH: Directly extend TW-FINCH to the time series tensor level, where the input is the raw data of each time frame.

[0203] (3) GGS: A classic segmentation method based on Gaussian fitting, which requires manual input of the number of segments. Here, when processing time series tensors, the tensor is directly spread out as input and the actual number of segments is directly input.

[0204] (4)IGTTS: Extend the IGTS algorithm to the time series tensor level.

[0205] The comparison results of different algorithms are shown in Table 5-Table 6:

[0206] Table 5 F1 scores of PASTWC and other algorithms on different datasets

[0207]

[0208] Table 6 Average p-lag of PASTWC and other algorithms on different datasets

[0209]

[0210] Tables 5 and 6 show the segmentation index evaluation of different segmentation methods. Obviously, compared with other comparison methods, PASTWC achieves the highest F1 and the lowest average p-lag times. In addition, the average values of the segmentation effect indicators of PASTWC on multiple datasets are better than those of other comparison methods. It achieves the best effect in overall comprehensive performance and has high segmentation effectiveness and accuracy.

[0211] Furthermore, the proposed pattern recognition algorithm is evaluated based on 12 public MTS datasets, 10 of which are from UEA datasets and 2 are constructed from human activity datasets. The specific datasets are shown in Table 7:

[0212] Table 7 Description of the dataset

[0213]

[0214] The experimental evaluation indicators are as follows:

[0215] (1) Rand Index (RI);

[0216]

[0217] Rand Index (RI) measures the similarity between two clustering results. It evaluates the quality of clustering by comparing the relationship between sample pairs of true labels and predicted labels.

[0218] A true positive (TP) indicates that two elements belong to the same category in the true classification and are also assigned to the same cluster in the clustering result. A false positive (FP) indicates that two elements do not belong to the same category in the true classification but are assigned to the same cluster in the clustering result. A true negative (TN) indicates that two elements do not belong to the same category in the true classification and are assigned to different clusters in the clustering result. A false negative (FN) indicates that two elements belong to the same category in the true classification but are assigned to different clusters in the clustering result. The RI value range is [0, 1], where 1 indicates a perfect match and 0 indicates a complete mismatch.

[0219] (2) Normalized mutual information (NMI);

[0220] Normalized Mutual Information (NMI) is a metric based on information theory that measures the degree of information sharing between true and predicted labels. It quantifies the similarity between clustering results and true labels by calculating mutual information. NMI is a normalized version of mutual information. Mutual information measures the amount of information shared between two random variables (such as true and predicted labels).

[0221]

[0222] where t is the true label, and is the experimental label obtained from the experiment, is t and The joint probability of , p(t) and They are t and The marginal probability of H(T) and T and The entropy of the variables represents the amount of information shared between them. The NMI value range is [0,1], where 1 indicates perfect agreement and 0 indicates no information sharing.

[0223] The comparison algorithm is as follows:

[0224] MTSC-FCM-CD: Builds complex networks based on the similarity between FCMs, leveraging the network to build long-range global information. It also applies community detection to capture clusters of arbitrary shapes in the network, enabling clustering of multivariate time series.

[0225] Corclust: Calculates empirical correlations (such as Pearson sample correlation coefficient) between pairs of spatial points to identify spatial and spatiotemporal neighbors with similar characteristics and performs iterative clustering based on sorting and selection to achieve spatiotemporal clustering.

[0226] TICC: The Toeplitz Inverse Covariance Clustering (TICC) algorithm clusters time series data, defining each cluster as a dependency network or Markov random field and describing the interdependencies across different variables and time.

[0227] Mc2PCA: assumes a common projection axis as the prototype of each cluster in K-Means clustering, and uses the reconstruction error of MTS data to redistribute the clustering results and monitor the convergence of the transformation process.

[0228] U-kmeans-DTW: The original algorithm of Uk-means, where the input is still the original data, and the distance measurement uses the DTW algorithm to measure the effect of feature extraction.

[0229] UTCFTS-A: A variant of UTCFTS, which directly uses the CSPCA algorithm as input without MIC-assisted clustering, and is used to measure the effectiveness of feature extraction and nearest neighbor rules based on MIC improvements.

[0230] The comparison results of different algorithms are shown in Table 8:

[0231] Table 8 Comparison of RandIndex and NormalizedMutual Information of different algorithms

[0232]

[0233]

[0234] It can be seen that the UTCFTS algorithm proposed in this invention has achieved good results on multiple data sets. The evaluation indicators also show that the algorithm can better identify different types of patterns and has obvious advantages over other comparison algorithms.

[0235] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for evolutionary pattern recognition based on time series tensor segmentation, characterized in that: The following steps are involved: S1, collect third-order time series tensor data as input data; S2. Divide the third-order time series tensor into multiple tensor segments using a sliding window; S3. extracting the core tensor of the tensor segment based on Tucker decomposition, and clustering it based on a time series weighted clustering method to obtain a coarse-grained segmentation result; S4. Optimizing the coarse-grained segmentation result based on information gain to obtain a fine-grained segmentation result; S5. extracting correlation features of each segment based on the maximum information coefficient; S6. Introduce the nearest neighbor rule and use the unsupervised three-way clustering algorithm to identify the evolutionary patterns corresponding to each fragment.

2. The method for evolutionary pattern recognition based on time series tensor segmentation according to claim 1, characterized in that: In step S1, the third-order time series tensor data is monitoring-type multivariate time series data collected by sensors, including information retrieval data, environmental monitoring data, human activity data, financial analysis data, and building energy consumption data.

3. The method for evolutionary pattern recognition based on time series tensor segmentation according to claim 1, characterized in that: The step S2 specifically includes: The collected third-order time series tensor X∈R M×N×T The sampling results of N variables for M users / locations in T time periods; According to the changing trend of the third-order time series tensor, K segmentation points are found through the sliding window, and X is divided into K+1 tensor segments along the time dimension.

4. The method for evolutionary pattern recognition based on time series tensor segmentation according to claim 1, characterized in that: The step S3 specifically includes: For the tensor segments obtained by sliding window, the corresponding core tensor is extracted using Tucker decomposition; In the clustering process, each core tensor is iteratively connected to its nearest neighbors through the space-time graph representation matrix to form a cluster. Each iterative clustering corresponds to a segmentation result. After multiple iterations, the segmentation result with the largest information gain growth rate is selected as the final coarse-grained segmentation result.

5. The method for evolutionary pattern recognition based on time series tensor segmentation according to claim 1, characterized in that: The step S4 specifically includes: The time interval of two windows around the coarse-grained segmentation point is selected, and greedy search is used to calculate the corresponding information gain for each time point in the interval as a segmentation point. The point with the largest information gain is selected as the optimized segmentation point to obtain the fine-grained segmentation result.

6. The method for evolutionary pattern recognition based on time series tensor segmentation according to claim 1, characterized in that: In step S5, the correlation includes variable correlation, time correlation, and user / location correlation.

7. The method for evolutionary pattern recognition based on time series tensor segmentation according to claim 1, characterized in that: The step S6 specifically includes: Set the entropy coefficient α c is the probability that a sample belongs to the cth category, and the membership degree z uc is the degree of membership of the u-th sample to the c-th category; Given the initial number of clusters C (0) =U, initial cluster center Initial entropy coefficient Where U is the number of samples; Cluster merging is performed using the nearest neighbor rule: the spatiotemporal points are sorted in descending order according to the number of nearest neighbors of each sample to create a list; the point with the most neighbors in the list is selected as the first cluster center, and all its neighboring points are assigned to this cluster; for subsequent points in the list, if a point does not belong to any cluster and more than 50% of its neighboring points also do not belong to any cluster, then this point is considered the center of the new cluster; The initial clustering result obtained based on the nearest neighbor rule is iterated multiple times according to the set objective function, and the entropy coefficient and membership are updated, and the cluster center and the members of each cluster are further updated; among them, by discarding Clusters to adjust and when Stop iteration and output the evolution pattern corresponding to each fragment.

8. An evolutionary pattern recognition system based on time series tensor segmentation, the system being used to implement the method according to any one of claims 1 to 7, characterized in that: The system comprises: A data acquisition module is used to collect third-order time series tensor data as input data; A time series tensor segmentation module is configured to divide the third-order time series tensor into multiple tensor segments using a sliding window; extract the core tensors of the tensor segments based on Tucker decomposition, and cluster them based on a time series weighted clustering method to obtain a coarse-grained segmentation result; and optimize the coarse-grained segmentation result based on information gain to obtain a fine-grained segmentation result; The evolutionary pattern recognition module is used to extract the correlation of each fragment based on the maximum information coefficient; the nearest neighbor rule is introduced and an unsupervised three-way clustering algorithm is used to identify the evolutionary pattern corresponding to each fragment.

9. The system according to claim 8, characterized in that The data acquisition module includes a sensor component, a serial port server, a management workstation, an interface server, an information center storage server, and a platform core switch; Among them, the sensor component is connected to the management workstation through the serial port server, the management workstation is connected to the interface server, the interface server is connected to the information center storage server through a one-way isolation gateway, the information center storage server is connected to the platform core switch through optical fiber, and the platform core switch is connected to multiple cloud computing nodes and web servers.

10. The system according to claim 8, wherein: The system also includes a monitoring APP, which includes: a client, a server and a system management background; The client is used for user registration and login, online query, modification and logout; the server is used for registration and login verification as well as data transmission, addition, modification and deletion; the system management background is used to manage the database.

Images