Spinning timing data fuzzy hierarchical clustering analysis method fusing time domain characteristics

Abstract

The purpose of this invention is to address the issue of accuracy in processing spinning time-series data streams, which are characterized by high noise levels and distinct time-domain features generated during the operation of spinning workshops, by employing a fuzzy hierarchical clustering method that integrates time-domain characteristics. This method reduces the impact of noise during the classification process and considers both time-domain features and noise effects. The technical solution of this invention is to provide a fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics. This invention proposes a fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain features. It iterates between the DTS feature matrix and the MTS feature matrix, considering the time-domain characteristics and noise effects in the spinning time-series data. Without increasing time complexity, it incorporates the time-frequency characteristics and noise effects in the spinning time-series data during the iteration process. Compared with the latest methods, this invention can more accurately process newly generated time-series data in spinning manufacturing.

Description

Technical Field

[0001] This invention relates to a method for analyzing time-series data generated during spinning manufacturing, specifically a fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics, belonging to the field of time-series data analysis. Background Technology

[0002] Mechanization, automation, and large-scale production are inevitable choices for textile enterprises to reduce costs, improve product quality, and enhance their competitiveness. Adjusting production plans based on changes in the workshop environment and production processes can improve the efficiency and quality of large-scale production.

[0003] With the development of data perception and artificial intelligence technologies, some scholars have tried to dynamically formulate production plans based on time-series data collected during the textile process. However, the analysis of time-series data faces two problems: (1) Noise has a great influence on time-series signals, which may lead to signal distortion and increased error, affecting the quality and accuracy of data classification; (2) Time-series signals in the short cycle of spinning process have periodic and repetitive characteristics, which interfere with the analysis of time-series data and make it difficult to guarantee the accuracy and comprehensiveness of data analysis.

[0004] Some scholars have proposed time-series data analysis methods based on fuzzy hierarchical clustering, which can reflect the fuzzy relationships between data, more closely reflect the actual situation, and, more importantly, consider the impact of factors such as noise on the clustering results, thus improving the stability and reliability of the clustering results. For example, Li Fangwei's team proposed a hierarchical evaluation model that combines qualitative and quantitative methods, and combined it with fuzzy hierarchical clustering criteria to make the index weights more consistent with the actual environment, resulting in more accurate evaluation results. Jerry W. Sangma's team proposed a fuzzy hierarchical clustering method that uses variable clustering methods to cluster multiple nominal data streams. It uses normalized cosine similarity to calculate the fuzzy affinity of the data streams to different clusters. Based on the degree of change in node entropy, it updates the hierarchical clustering structure by merging and / or splitting nodes, improving the accuracy of clustering. Summary of the Invention

[0005] The purpose of this invention is to provide a fuzzy hierarchical clustering time series data analysis method that integrates time domain characteristics, so as to analyze time series data collected in the textile process.

[0006] To achieve the above objectives, the technical solution of the present invention provides a fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics, characterized by comprising the following steps:

[0007] Step 1: Use sliding window technology to collect historical runtime sequence data of the device and record the corresponding time point t. Each sliding window is regarded as a data matrix with m×m dimensions, and each data matrix corresponds to a time point t.

[0008] Step 2: Construct the DTS feature matrix and MTS feature matrix based on the historical runtime sequence data obtained in Step 1, including the following steps:

[0009] Step 201: Calculate the spinning time series data DTS factor d mn :

[0010]

[0011] In formula (1): It is the spinning time series data in historical time series data D1; a i b j They are matrices Independent row vectors in the array; ||·|| 2 This indicates the calculation of Euclidean distance; The formula for calculating the impact factor of time-domain data is given, where t m It is a matrix The time of yarn unwinding, t n It is a matrix The doffing time, where k is an exponential parameter;

[0012] Step 202: After the DTS factor of any two spinning time series data in the historical runtime data is calculated, the DTS feature matrix is ​​generated;

[0013] Step 203: Randomly initialize the MTS feature matrix;

[0014] Step 3: Update the MTS feature matrix based on the MTS feature matrix and DTS feature matrix obtained in the previous step:

[0015]

[0016] In equation (2): u ij d represents the membership degree of data node i to each cluster j; ij σ represents the MTS factor between data node i and cluster j; σ is a parameter that controls the width of the Gaussian kernel function, which is the average of all values ​​in the DTS feature matrix;

[0017] Step 4: Update the DTS feature matrix, including the following steps:

[0018] Step 401: Calculate the new DTS factor based on the updated MTS feature matrix:

[0019]

[0020] In equation (3), d kj The original DTS factor, u ik u jkThe value of s is determined by the updated MTS feature matrix from step 3. ij It is the new DTS factor;

[0021] Step 402: Update the DTS feature matrix based on the new DTS factor obtained in step 401;

[0022] Step 5: Merge the two closest clusters and update the DTS feature matrix, which includes the following steps:

[0023] Step 501: Update the unclassified cluster set, merge the two closest clusters to form a new cluster, add this new cluster to the unclassified cluster set, and delete the two closest clusters;

[0024] Step 502: There are two types of clusters in the fuzzy time-series spinning data hierarchy tree, namely single-point clusters and merged clusters. The DTS factor between single-point clusters is calculated using formula (1).

[0025] DTS factor d for single-point clusters and merged clusters mn The calculation method further includes the following steps:

[0026] Step 5021: Calculate the cluster centers of the merged clusters:

[0027]

[0028] In equation (4), It is the i-th data instance in the merged cluster. Indicates the cluster center of the currently merged cluster;

[0029] Step 5022: Calculate the DTS factor d for single-point clusters and merged clusters. mn :

[0030]

[0031] In equation (5): This refers to the spinning time series data within a single-point cluster. Indicates the cluster center of the spinning time-series data in the current merged cluster; a i b j They are matrices Independent row vectors in t; m yes The time for the yarn to fall off; t n yes The average doffing time.

[0032] Merged clusters and their DTS factors d mn The calculation method further includes the following steps:

[0033] Step 5121: Calculate the cluster centers of the merged clusters:

[0034]

[0035] In equation (6): These are the cluster centers of the two merged clusters. These are the i-th data instance in the merged cluster, and n is the number of data instances.

[0036] Step 5122: Calculate the DTS factor d of the two merged clusters. m,n :

[0037]

[0038] In equation (7): These are the cluster centers of the two merged clusters; a i b j They are matrices Independent row vectors in t; m yes Average doffing time; t n yes The average doffing time.

[0039] Step 503: After the DTS factor calculation is completed, update the DTS feature matrix.

[0040] Step 6: Repeat steps 3, 4 and 5 until all clusters are merged to obtain a hierarchical tree model of fuzzy time-series spinning data;

[0041] Step 7: Use sliding window technology to acquire real-time runtime timing data of the equipment. Input the acquired real-time runtime timing data as new samples into the fuzzy time-series spinning data hierarchical tree model. The fuzzy time-series spinning data hierarchical tree model determines the type of the new sample, including the following steps:

[0042] Step 701: Calculate the DTS factor for the new sample and the already divided clusters;

[0043] Step 702: Find the cluster with the smallest DTS factor, calculate the Gaussian kernel density K(x) of this cluster, and then calculate the Gaussian kernel density K(x) after adding the new sample to this cluster. ′ ):

[0044] like If a new sample is considered to belong to a cluster, only the MTS feature matrix corresponding to that cluster is updated.

[0045] like The new sample is considered to belong to a new cluster point and is added to the fuzzy time-series spinning data hierarchical tree model, updating all DTS feature matrices and MTS feature matrices.

[0046] Preferably, when using sliding window technology to collect historical or real-time runtime sequence data of the device, each sliding window contains m data streams, and for each data stream, m data instances are extracted from different time points.

[0047] Preferably, in step 203, each data node is considered as an independent cluster, and the MTS feature matrix is ​​initialized as follows: A n,n , i≠j, and n is the number of data nodes.

[0048] Preferably, in step 6, the obtained fuzzy time-series spinning data hierarchical tree model is segmented as needed.

[0049] Preferably, in step 701, the DTS factor of the new sample and the already divided clusters is calculated using formula (1) or formula (5).

[0050] In the clustering process, this invention introduces a time factor into the distance matrix. Through continuous iteration of the membership matrix and distance matrix, it minimizes the impact of noise and the enrichment of signal features within certain time periods in the time-series data, thereby improving the accuracy of data analysis. When analyzing time-series data generated in spinning manufacturing using the method disclosed in this invention, it can minimize the impact of noise and the enrichment of signal features within certain time periods, ensuring the accuracy of time-series data stream analysis. This enables accurate analysis of time-series data generated in spinning manufacturing, providing a foundation for decision-making processes such as equipment condition monitoring. Attached Figure Description

[0051] Figure 1 This illustrates the sample model obtained through a sliding window;

[0052] Figure 2 The obtained fuzzy hierarchical tree model is illustrated. Detailed Implementation

[0053] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0054] This embodiment discloses a fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics. It analyzes the time-series data generated during the spinning process using a fuzzy hierarchical tree, specifically including the following steps:

[0055] Step 1: Data Collection

[0056] The target dataset was collected from the Litai Korla and Kuitun parks in Xinjiang. For the core components of each device, appropriate sensors were installed to read and record the corresponding data, which was then transmitted to a local server for classification. The server can be selected according to actual needs. The collected dataset contains 4,500 time series, each with a length of 140, totaling approximately 630,000 data points.

[0057] like Figure 1 As shown, the j-th sliding window (W) j It contains m = 5 data streams, namely (S1, S2, S3, S4, and S5). For each data stream, five data instances are extracted from different time points from time t1 to t5. Therefore, as... Figure 1 In the example shown, the sliding window (W) j It can be represented as a data matrix with a dimension of (5×5).

[0058] Step 2: Data Processing

[0059] The sliding window technique is used to process the historical runtime sequence data and real-time time series data collected in step 1. Specifically, the sliding window method collects a data set D in the data stream. i Each sliding window contains w data instances, so it can be viewed as an m×m dimension data matrix, with each data matrix corresponding to a time point t. w In this case, the size of the sliding window is defined as m×m.

[0060] Step 3: Construct the DTS feature matrix and MTS feature matrix based on the historical time series data D1 obtained in Step 2. The specific steps are as follows:

[0061] Step 301: Treat each data node in the historical time series data D1 as a single-point cluster N. i At this point, the set of unclassified clusters is the union of all single-point clusters.

[0062] Step 302: Calculate the DTS (Distance-Time-Spinning Feature Matrix) factor d of the spinning time series data. mn :

[0063]

[0064] in: It is the spinning time series data in historical time series data D1; a i b j They are matrices Independent row vectors in ||ai -b j || 2 It is a i With b j Euclidean distance between quantities; The formula for calculating the impact factor of time-domain data is given, where t m It is a matrix The time of yarn unwinding, t n It is a matrix The doffing time is k, which is an exponential parameter. A larger k will make the time domain effect more ambiguous. Here we choose k=2.

[0065] Step 303: Based on the DTS factor d obtained in step 302 mn Obtain the initial DTS feature matrix.

[0066] Step 304: Randomly initialize the MTS (Membership-Time-Spinning Feature Matrix) as follows:

[0067] Since we haven't specified the number of clusters, we can treat each data node as an independent cluster. Therefore, the initial MTS feature matrix is: A n,n , i≠j, and n is the number of data nodes.

[0068] For example, when n=3, the initial MTS feature matrix is:

[0069]

[0070] Step 4: Update the MTS feature matrix based on the DTS feature matrix and the MTS feature matrix. The specific formula is as follows:

[0071]

[0072] Where: u ij d represents the membership degree of data node i to each cluster j; ij σ represents the MTS factor between data node i and cluster j; σ is a parameter that controls the width of the Gaussian kernel function, which is generally taken as the average of all values ​​in the DTS feature matrix.

[0073] Step 5: Update the DTS feature matrix, which includes the following steps:

[0074] Step 501: Calculate the new DTS factor based on the updated MTS feature matrix. The specific formula is as follows:

[0075]

[0076] In the formula, u iku jk The value of s is determined by the updated MTS feature matrix from step 4. ij That is, the new DTS factor. Represent matrices respectively Independent row vectors in t k yes The time of yarn unwinding, t n yes The time of yarn unwinding, d kj This is the existing DTS factor.

[0077] Step 502: Update the DTS feature matrix based on the new DTS factor obtained in step 501.

[0078] Step 6: Merge the two closest clusters and update the DTS feature matrix, which includes the following steps:

[0079] Step 601: Update the unclassified cluster set, merge the two closest clusters to form a new cluster, add this new cluster to the unclassified cluster set, and delete the two closest clusters;

[0080] Step 602: There are two types of clusters in the hierarchical tree of fuzzy time-series spinning data: single-point clusters and merged clusters. The calculation method of the DTS factor between single-point clusters is the same as in step 302.

[0081] DTS factor d for single-point clusters and merged clusters mn The calculation method further includes the following steps:

[0082] Step 6021: Calculate the cluster centers of the merged clusters:

[0083]

[0084] In the formula, It is the i-th data instance in the merged cluster. Indicates the cluster center of the currently merged cluster;

[0085] Step 6022: Calculate the DTS factor d for single-point clusters and merged clusters. mn :

[0086]

[0087] In the formula: This refers to the spinning time series data within a single-point cluster. Indicates the cluster center of the spinning time-series data in the current merged cluster; a i b j They are matrices Independent row vectors in ||a i -b j ||2 It is a i With b j Euclidean distance between them; The formula for calculating the impact factor of time-domain data is given, where t m yes The time of yarn unwinding, t n yes The average doffing time.

[0088] Merged clusters and their DTS factors d mn The calculation method further includes the following steps:

[0089] Step 6121: Calculate the cluster centers of the merged clusters:

[0090]

[0091] In the formula: These are the cluster centers of the two merged clusters. These are the i-th data instance in the merged cluster, and n is the number of data instances.

[0092] Step 6122: Calculate the DTS factor d of the two merged clusters. m,n :

[0093]

[0094] In the formula: These are the cluster centers of the two merged clusters; a i b j They are matrices Independent row vectors in ||a i -b j || 2 It is a i With b j Euclidean distance between them; The formula for calculating the impact factor of time-domain data is given, where t m yes Average doffing time, t n yes The average doffing time.

[0095] Step 603: After the DTS factor calculation is completed, update the DTS feature matrix.

[0096] Step 7: Repeat steps 4, 5, and 6 until all clusters are merged. Do not select splitting (splitting can also be done as needed), such as... Figure 2 As shown.

[0097] Step 8: Determine the type of the new sample (time series data added in real time) by taking the following steps:

[0098] Step 801: Calculate the DTS factor of the new sample and the already divided clusters. The formula is the DTS factor calculation formula in step 302 or the DTS factor d of the single-point cluster and the merged cluster described in step 602. mn The calculation method;

[0099] Step 802: Find the cluster with the smallest DTS factor, calculate the Gaussian kernel density K(x) of this cluster, and then calculate the Gaussian kernel density K(x) after adding the new sample to this cluster. ′ );

[0100] like If the new sample is considered to belong to a cluster (same category), only the MTS feature matrix corresponding to that cluster will be updated;

[0101] like The new sample is then considered to belong to a new cluster (different category), added to the hierarchical tree, and all DTS feature matrices and MTS feature matrices are updated.

[0102] In this embodiment, the value of ε was determined to be 0.4 through experiments, and it can be adjusted according to specific circumstances.

[0103] Based on the above method, this embodiment uses the Python programming language and the TensorFlow platform to train and build... Figure 2 The model shown should have firmware specifications higher than the following: Core i7 CPU @ 2.20GHz and 8GB RAM. Model verification includes the following:

[0104] The performance of the established prediction model is evaluated using statistical parameter accuracy (A). Furthermore, to assess the model's performance in detecting false positives and false negatives of energy efficiency anomalies, this invention also employs evaluation metrics such as precision (P), recall (R), and F-score (F1), defined as follows:

[0105] True Yang (TP): Predicted to be normal, actually normal.

[0106] False positive (FP): The predicted value is normal, but the actual value is an outlier.

[0107] False negative (FN): The predicted value is an outlier, but the actual value is normal.

[0108] True Negative (TN): Predicted as an outlier, actually an outlier.

[0109]

[0110]

[0111]

[0112]

[0113] A higher P-value indicates fewer false positives that identify actual normal data as abnormal. A higher R-value indicates fewer false negatives that identify actual abnormal data as normal. F1 is a combined metric of Precision and Recall, which comprehensively reflects the model's performance in handling false positives and false negatives. Model training is complete when the model reaches the set precision or maximum number of iterations, and the model is then run to monitor energy consumption throughout the spinning process. The classification results are shown in Table 1.

[0114] Table 1. ECG5000_TEST Classification Results

[0115]

[0116] The above technical solution establishes a fuzzy hierarchical tree for analyzing time-series data generated during spinning. This method utilizes a sliding window for data collection, taking into account the characteristics of time-series data. During clustering, a time factor is introduced into the distance matrix. Through continuous iteration of the membership matrix and distance matrix, the influence of noise and signal feature enrichment within certain time periods is minimized, thereby improving the accuracy of data analysis.

Claims

1. A fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics, characterized in that, Includes the following steps: Step 1: Use sliding window technology to collect historical runtime sequence data of the device and record the corresponding time point t. Each sliding window is regarded as a data matrix with m×m dimensions, and each data matrix corresponds to a time point t. Step 2: Construct the DTS feature matrix and MTS feature matrix based on the historical runtime sequence data obtained in Step 1, including the following steps: Step 201: Calculate the spinning time series data DTS factor d mn : In formula (1): It is the spinning time series data in historical time series data D1; a i b j They are matrices Independent row vectors in the array; ||·|| 2 This indicates the calculation of Euclidean distance; The formula for calculating the impact factor of time-domain data is given, where t m It is a matrix The time of yarn unwinding, t n It is a matrix The doffing time, where k is an exponential parameter; Step 202: After the DTS factor of any two spinning time series data in the historical runtime data is calculated, the DTS feature matrix is ​​generated; Step 203: Randomly initialize the MTS feature matrix; Step 3: Update the MTS feature matrix based on the MTS feature matrix and DTS feature matrix obtained in the previous step: In equation (2): u ij d represents the membership degree of data node i to each cluster j; ij σ represents the MTS factor between data node i and cluster j; σ is a parameter that controls the width of the Gaussian kernel function, which is the average of all values ​​in the DTS feature matrix; Step 4: Update the DTS feature matrix, including the following steps: Step 401: Calculate the new DTS factor based on the updated MTS feature matrix: In equation (3), d kj The original DTS factor, u ik u jk The value of s is determined by the updated MTS feature matrix from step 3. ij It is the new DTS factor; Step 402: Update the DTS feature matrix based on the new DTS factor obtained in step 401; Step 5: Merge the two closest clusters and update the DTS feature matrix, which includes the following steps: Step 501: Update the unclassified cluster set, merge the two closest clusters to form a new cluster, add this new cluster to the unclassified cluster set, and delete the two closest clusters; Step 502: There are two types of clusters in the fuzzy time-series spinning data hierarchy tree, namely single-point clusters and merged clusters. The DTS factor between single-point clusters is calculated using formula (1). DTS factor d for single-point clusters and merged clusters mn The calculation method further includes the following steps: Step 5021: Calculate the cluster centers of the merged clusters: In equation (4), It is the i-th data instance in the merged cluster. Indicates the cluster center of the currently merged cluster; Step 5022: Calculate the DTS factor d for single-point clusters and merged clusters. mn : In equation (5): This refers to the spinning time series data within a single-point cluster. Indicates the cluster center of the spinning time-series data in the current merged cluster; a i b j They are matrices Independent row vectors in t; m yes The time for the yarn to fall off; t n yes Average doffing time; Merged clusters and their DTS factors d mn The calculation method further includes the following steps: Step 5121: Calculate the cluster centers of the merged clusters: In equation (6): These are the cluster centers of the two merged clusters. These are the i-th data instance in the merged cluster, and n is the number of data instances. Step 5122: Calculate the DTS factor d of the two merged clusters. m,n : In equation (7): These are the cluster centers of the two merged clusters; a i b j They are matrices Independent row vectors in t; m yes Average doffing time; t n yes Average doffing time; Step 503: After the DTS factor calculation is completed, update the DTS feature matrix; Step 6: Repeat steps 3, 4 and 5 until all clusters are merged to obtain a hierarchical tree model of fuzzy time-series spinning data; Step 7: Use sliding window technology to acquire real-time runtime timing data of the equipment. Input the acquired real-time runtime timing data as new samples into the fuzzy time-series spinning data hierarchical tree model. The fuzzy time-series spinning data hierarchical tree model determines the type of the new sample, including the following steps: Step 701: Calculate the DTS factor for the new sample and the already divided clusters; Step 702: Find the cluster with the smallest DTS factor, calculate the Gaussian kernel density K(x) of this cluster, and then calculate the Gaussian kernel density K(x) after adding the new sample to this cluster. ′ ): like If a new sample is considered to belong to a cluster, only the MTS feature matrix corresponding to that cluster is updated. like The new sample is considered to belong to a new cluster point and is added to the fuzzy time-series spinning data hierarchical tree model, updating all DTS feature matrices and MTS feature matrices.

2. The fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics as described in claim 1, characterized in that, When using sliding window technology to collect historical or real-time runtime sequence data of a device, each sliding window contains m data streams, and for each data stream, m data instances are extracted from different time points.

3. The fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics as described in claim 1, characterized in that, In step 203, each data node is considered as an independent cluster, and the MTS feature matrix is ​​initialized as follows: A n,n , i≠j, and n is the number of data nodes.

4. The fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics as described in claim 1, characterized in that, In step 6, the obtained fuzzy time-series spinning data hierarchical tree model is segmented as needed.

5. The fuzzy hierarchical clustering analysis method for spinning time-series data that integrates time-domain characteristics as described in claim 1, characterized in that, In step 701, the DTS factor of the new sample and the already divided clusters is calculated using formula (1) or formula (5).

Images