Control system data statistical method and system based on cluster

Abstract

The invention provides a cluster-based control system data statistics method and system. The method comprises the following steps: determining a multi-dimensional time sequence state data core object of a control system; starting from any unaccessed core object, forming an initial data cluster, fitting a Gaussian mixture model by using an expectation maximization algorithm, and extracting a parameter set of the Gaussian mixture model as statistical characteristics of the cluster; iteratively calculating the Bhattacharyya distance between the Gaussian mixture models of any two data clusters, and when the distance is smaller than a preset merging threshold value, re-fitting the Gaussian mixture model for the merged new clusters until no cluster can be merged; in any data cluster, cluster generation and combination are carried out again; data points which do not belong to any cluster are regarded as noise points, a noise point set of the region is regarded as a new data set according to the space-time density, and cluster generation, merging and splitting steps are executed on the new data set to generate a new cluster; and outputting the Gaussian mixture model parameter set of all the current data clusters as a statistical result of the current working condition of the control system.

Description

Technical Field

[0001] The present application relates to the field of system control, and in particular to a cluster-based control system data statistics method and system. Background Art

[0002] In modern industrial control fields such as chemical engineering, power generation, and intelligent manufacturing, control systems are becoming increasingly complex, generating massive amounts of multidimensional state data with time-series characteristics during operation. Real-time and accurate analysis of this data is a key prerequisite for online operating condition monitoring, fault diagnosis, and adaptive control strategy adjustment. Data-driven statistical analysis methods, particularly cluster analysis, are currently the mainstream technology for identifying and characterizing different system operating conditions. For example, algorithms such as K-means (K-Means), Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and Gaussian Mixture Models (GMM) are widely used to group data points with similar characteristics into distinct data clusters. Each data cluster is typically considered to correspond to a relatively stable operating state or mode of the system. Extracting and analyzing the features of these data clusters can provide a decision-making basis for understanding and controlling complex industrial processes.

[0003] However, traditional clustering methods face many challenges when applied to dynamically changing industrial processes. The operating conditions of industrial systems often exhibit complex dynamic characteristics such as gradual drift, smooth transition, merger and split, and even the emergence of new operating conditions. The K-Means algorithm requires a preset number of clusters, which makes it difficult to adapt to the dynamic changes in the number of operating conditions, while the global parameter setting of the DBSCAN algorithm makes it difficult to handle data clusters with uneven density, different shapes and dynamic evolution. The traditional similarity measurement method based on Euclidean distance ignores the correlation between data dimensions and cannot accurately capture the true distribution form within the operating condition data. Most existing methods lack an effective management mechanism for the cluster evolution process, making it difficult to automatically handle the merger and splitting between clusters, and unable to effectively monitor outlier data points that are temporarily regarded as noise, thereby possibly ignoring potential new operating conditions gradually formed by these noise points.

[0004] Therefore, there is an urgent need for a data statistics method that can dynamically and adaptively track the entire process of operating condition evolution and automatically complete the generation, merging, splitting and initiation of new clusters, so as to provide accurate and real-time operating condition feature description for advanced adaptive control systems. Summary of the Invention

[0005] To address the above issues, this application proposes a cluster-based control system data statistics method, including:

[0006] Acquire multidimensional time series state data of the control system; for each data point, count the number of neighboring data points in a spatiotemporal neighborhood defined by a preset Mahalanobis distance threshold and a temporal proximity threshold, and define the data point as a core object when the number is greater than a core object determination threshold;

[0007] Starting from any unvisited core object, recursively expand it using the density reachability principle to form initial data clusters. Then, use the expectation-maximization algorithm to fit a Gaussian mixture model to each initial data cluster, and extract the parameter set of the Gaussian mixture model as the statistical characteristics of the cluster. Iteratively calculate the Bhattacharyya distance between the Gaussian mixture models of any two data clusters. When the distance is less than a preset merging threshold, the two data clusters are merged and the Gaussian mixture model is refitted to the merged new cluster until no clusters can be merged.

[0008] The average log-likelihood of all data points in any data cluster relative to its Gaussian mixture model is calculated. When the average log-likelihood is lower than a preset splitting threshold, the data points in the cluster are regarded as independent data sets, and the clusters are regenerated and merged. The data points that do not belong to any cluster are regarded as noise points, and their spatiotemporal density is monitored within a sliding time window. When the spatiotemporal density exceeds a preset germination threshold, the noise point set in the region is used as a new data set, and the cluster generation, merging and splitting steps are performed on the new data set to generate a new cluster. The Gaussian mixture model parameter set of all current data clusters is output as the statistical result of the current working condition of the control system.

[0009] Preferably, defining the data point as a core object includes:

[0010] A preset Mahalanobis distance threshold, a temporal proximity threshold, and a core object determination threshold are set; for any data point, if the number of its neighboring data points in the spatiotemporal neighborhood defined by the Mahalanobis distance threshold and the temporal proximity threshold exceeds the core object determination threshold, the data point is defined as a core object.

[0011] Preferably, fitting a Gaussian mixture model for each initial data cluster using an expectation-maximization algorithm comprises:

[0012] The initial number of components of the Gaussian mixture model is set for the initial data cluster to be fitted, and the expectation maximization algorithm is used for iterative fitting until the change in the log-likelihood value of the model meets the preset convergence condition.

[0013] Preferably, the merging of the two data clusters comprises:

[0014] The Bhattacharyya distance between the Gaussian mixture models corresponding to the two data clusters is calculated, and when the Bhattacharyya distance is less than a preset merging threshold, merging is performed.

[0015] Preferably, the step of monitoring the spatiotemporal density within the sliding time window and taking the noise point set of the region as a new data set when the spatiotemporal density exceeds a preset germination threshold comprises:

[0016] Within a preset sliding time window, the state space is divided into grid cells, and the number of noise points falling into each grid cell is counted; when the number of noise points in any grid cell exceeds the preset germination threshold, the noise point set in the cell is processed as a new data set.

[0017] Preferably, the parameter set of the Gaussian mixture model includes the weight, mean vector and covariance matrix of each Gaussian component.

[0018] The present application also provides a cluster-based control system data statistics system, comprising:

[0019] a core object acquisition unit configured to acquire multi-dimensional time series state data of the control system; for each data point, counting the number of neighboring data points within a spatiotemporal neighborhood defined by a preset Mahalanobis distance threshold and a temporal proximity threshold; and defining the data point as a core object when the number is greater than a core object determination threshold;

[0020] The merging unit is used to start from any unvisited core object, recursively expand it based on the density reachability principle to form an initial data cluster, and use the expectation maximization algorithm to fit a Gaussian mixture model to each initial data cluster, extracting the parameter set of the Gaussian mixture model as the statistical characteristics of the cluster; it iteratively calculates the Bhattacharyya distance between the Gaussian mixture models of any two data clusters. When the distance is less than a preset merging threshold, the two data clusters are merged and the Gaussian mixture model is refitted to the merged new cluster until no clusters can be merged;

[0021] The result output unit is used to calculate the average log-likelihood of all data points in any data cluster relative to its Gaussian mixture model. When the average log-likelihood is lower than a preset splitting threshold, the data points in the cluster are regarded as independent data sets, and re-clustering and merging are performed on them; data points that do not belong to any cluster are regarded as noise points, and their spatiotemporal density is monitored within a sliding time window. When the spatiotemporal density exceeds a preset germination threshold, the noise point set in the area is used as a new data set, and the cluster generation, merging and splitting steps are performed on the new data set to generate a new cluster; and the Gaussian mixture model parameter set of all current data clusters is output as the statistical result of the current operating condition of the control system.

[0022] Preferably, defining the data point as a core object includes:

[0023] A preset Mahalanobis distance threshold, a temporal proximity threshold, and a core object determination threshold are set; for any data point, if the number of its neighboring data points in the spatiotemporal neighborhood defined by the Mahalanobis distance threshold and the temporal proximity threshold exceeds the core object determination threshold, the data point is defined as a core object.

[0024] Preferably, fitting a Gaussian mixture model for each initial data cluster using an expectation-maximization algorithm comprises:

[0025] The initial number of components of the Gaussian mixture model is set for the initial data cluster to be fitted, and the expectation maximization algorithm is used for iterative fitting until the change in the log-likelihood value of the model meets the preset convergence condition.

[0026] Preferably, the merging of the two data clusters comprises:

[0027] The Bhattacharyya distance between the Gaussian mixture models corresponding to the two data clusters is calculated, and when the Bhattacharyya distance is less than a preset merging threshold, merging is performed.

[0028] Preferably, the step of monitoring the spatiotemporal density within the sliding time window and taking the noise point set of the region as a new data set when the spatiotemporal density exceeds a preset germination threshold comprises:

[0029] Within a preset sliding time window, the state space is divided into grid cells, and the number of noise points falling into each grid cell is counted; when the number of noise points in any grid cell exceeds the preset germination threshold, the noise point set in the cell is processed as a new data set.

[0030] Preferably, the parameter set of the Gaussian mixture model includes the weight, mean vector and covariance matrix of each Gaussian component.

[0031] The present invention has the following beneficial effects:

[0032] By introducing the Mahalanobis distance to define the spatiotemporal neighborhood of data points, the correlation between different dimensions of the data can be effectively taken into account. It can more accurately measure the true similarity between data points than the traditional Euclidean distance, thereby improving the accuracy of core object identification and initial data cluster division.

[0033] Fitting data clusters with a Gaussian mixture model can not only capture the arbitrary shape and direction of the clusters, but also provide a more refined and comprehensive statistical description of the system operating conditions in the form of a set of probability distribution parameters.

[0034] By calculating the Bhattacharyya distance between models and the average log-likelihood of data within a cluster, objective cluster merging and splitting criteria are established, which can adjust the cluster structure according to the intrinsic changes in data distribution, thereby accurately reflecting the integration and differentiation process of working conditions.

[0035] The monitoring of noise points and the new cluster initiation mechanism can discover and generate new operating condition clusters from temporarily ignored data, avoiding the omission of potential new patterns and providing a more accurate, detailed and reliable data basis for subsequent control decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of Example 1;

[0037] Figure 2 A schematic diagram of the core object;

[0038] Figure 3 Schematic diagram of initial cluster generation and Gaussian mixture model fitting;

[0039] Figure 4 Schematic diagram before and after the merger;

[0040] Figure 5 Schematic diagram before and after splitting;

[0041] Figure 6 Schematic diagram before and after germination. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0043] In a specific embodiment, the present application proposes a cluster-based control system data statistics method, such as Figure 1 Shown, including:

[0044] S1, acquiring multi-dimensional time series state data of the control system; for each data point, counting the number of neighboring data points in a spatiotemporal neighborhood defined by a preset Mahalanobis distance threshold and a temporal proximity threshold, and defining the data point as a core object when the number is greater than a core object determination threshold;

[0045] Through the distributed control system DCS or the data acquisition and supervisory control system SCADA, the measured values ​​of multiple key variables in the control system are collected in real time. For example, sensor data such as the temperature, pressure, liquid level and stirring speed of the chemical reactor are collected to form a multi-dimensional data vector stream with a time stamp. The covariance matrix of all data points in the sliding time window is calculated for the subsequent calculation of the Mahalanobis distance. For any data point p, other data points q within a certain range before and after it in time are traversed, and the Mahalanobis distance between p and q is calculated. If the absolute value of the difference between the timestamp of q and the timestamp of p is less than the preset time proximity threshold, and the Mahalanobis distance between p and q is less than the preset Mahalanobis distance threshold, then q is regarded as a neighbor of p. The total number of neighboring points of each data point is counted. If the total number exceeds the core object determination threshold, for example 5, the data point is marked as a core object, such as Figure 2 shown.

[0046] S2: Starting from any unvisited core object, recursively expand using the density reachability principle to form initial data clusters, and fit a Gaussian mixture model to each initial data cluster using the expectation-maximization algorithm. The parameter set of the Gaussian mixture model is extracted as the statistical characteristics of the cluster; the Bhattacharyya distance between the Gaussian mixture models of any two data clusters is iteratively calculated. When the distance is less than a preset merging threshold, the two data clusters are merged, and the Gaussian mixture model is refitted to the merged new cluster until no clusters can be merged;

[0047] Similar to the cluster expansion of the DBSCAN algorithm, a core object that has not yet been assigned to any cluster is randomly selected as a seed point, and all neighboring points in its spatiotemporal neighborhood are added to a queue to be processed. A data point is taken from the queue. If the point is also a core object, all unvisited points in its neighborhood are also added to the queue. Repeat this process until the queue is empty, and all visited points together constitute an initial data cluster. Apply the expectation maximization algorithm to all data points contained in the initial data cluster, and iteratively execute the expectation step and the maximization step to finally converge and obtain the parameters of the Gaussian mixture model that describes the distribution of the data cluster, including the weight, mean vector and covariance matrix of each Gaussian component. For any two data clusters and , using their respective Gaussian mixture model parameters, such as mean vector and covariance matrix, calculate the Bhattacharyya distance between the two model distributions. The Bhattacharyya distance measures the degree of overlap between the two probability distributions. Build a matrix containing the Bhattacharyya distances between all cluster pairs and find the pair of clusters with the smallest distance. If the minimum distance is less than the preset merging threshold, all data points contained in the two clusters are merged into a larger data set, and the expectation maximization algorithm is performed again on this new data set to fit a new Gaussian mixture model to represent the merged new cluster, such as Figure 3 、4 This process is repeated until the Bhattacharyya distance between any two clusters is greater than the merging threshold.

[0048] S3, calculate the average log-likelihood of all data points in any data cluster relative to its Gaussian mixture model. When the average log-likelihood is lower than the preset splitting threshold, the data points in the cluster are regarded as independent data sets, and the clusters are regenerated and merged. The data points that do not belong to any cluster are regarded as noise points, and their spatiotemporal density is monitored within the sliding time window. When the spatiotemporal density exceeds the preset germination threshold, the noise point set in the area is used as a new data set, and the cluster generation, merging and splitting steps are performed on the new data set to generate a new cluster. The Gaussian mixture model parameter set of all current data clusters is output as the statistical result of the current working condition of the control system.

[0049] For each existing data cluster, all the data points within it are traversed, and the log-likelihood probability of each data point is calculated using the Gaussian mixture model corresponding to the cluster. Then, these log-likelihood probability values ​​are averaged to obtain an indicator that measures the fit of the model to the data. The average log-likelihood measures the fit of the model to the data. If the data points in a cluster are poorly consistent with their model as a whole, that is, the average log-likelihood is very small, it means that a single model can no longer effectively describe the complex structure within the cluster, and the cluster may have begun to split. If the average value is lower than a preset split threshold, it means that the current model can no longer well represent the overall distribution of all data points in the cluster, and may contain two or more independent sub-conditions, such as Figure 5 At this point, all the data points in the cluster are released as a new, independent data set to be processed. The cluster analysis is re-performed on this data set starting from the aforementioned core object determination, initial cluster generation, and cluster merging steps to discover the possible substructures within it.

[0050] All data points that are not classified into any cluster are temporarily marked as noise points, and a sliding time window of fixed length is maintained, for example, the latest 1000 data points. Within this window, grid division or kernel density estimation methods are used to continuously monitor the spatial distribution density of these noise points. When it is found that the number of noise points in a local area continues to accumulate over a certain period of time, and its spatiotemporal density, such as the value calculated by kernel density estimation, exceeds the preset germination threshold, it is considered that a new working condition may be forming. At this time, all noise points in the high-density area are extracted to form a new data set, such as Figure 6 As shown, the aforementioned complete set of processing procedures from core object identification, initial cluster generation, cluster merging to cluster splitting is fully executed on the data set, thereby identifying and establishing data clusters representing new working conditions.

[0051] After completing all the aforementioned generation, merging, splitting, and germination steps, the Gaussian mixture model parameter sets corresponding to all currently stable data clusters—that is, the weights, mean vectors, and covariance matrices of each cluster's Gaussian components—are packaged into a structured data format, such as a JSON or XML file. This file is output to the upper-level adaptive control module or condition monitoring system as a quantitative statistical description of the overall operating conditions of the current control system, providing a basis for subsequent control law adjustments, performance evaluation, and fault warnings.

[0052] In an optional embodiment, defining the data point as a core object includes:

[0053] A preset Mahalanobis distance threshold, a temporal proximity threshold, and a core object determination threshold are set; for any data point, if the number of its neighboring data points in the spatiotemporal neighborhood defined by the Mahalanobis distance threshold and the temporal proximity threshold exceeds the core object determination threshold, the data point is defined as a core object.

[0054] The center point of a data-dense area is identified by setting three key parameters. Mahalanobis distance is a spatial distance metric that takes into account the distribution of data, temporal proximity defines temporal correlation, and the core object determination threshold specifies the minimum number of neighboring points required to become a core. These three key parameters can effectively identify data points that exhibit clustering characteristics in both time and space, and avoid misjudging isolated points in time and space as meaningful core objects. For example, the Mahalanobis distance threshold can be set to 3, the temporal proximity threshold to 60 seconds, and the core object determination threshold to 5. For the data point of device A at 10:00, all other data points within 60 seconds before and after 10:00 and with a spatial Mahalanobis distance of less than 3 from device A are searched. If 6 such neighboring data points are found, because 6 is greater than the core object determination threshold of 5, this data point of device A is defined as a core object, marking the center of a potential dense cluster.

[0055] In an optional embodiment, fitting a Gaussian mixture model for each initial data cluster using an expectation-maximization algorithm includes:

[0056] The initial number of components of the Gaussian mixture model is set for the initial data cluster to be fitted, and the expectation maximization algorithm is used for iterative fitting until the change in the log-likelihood value of the model meets the preset convergence condition.

[0057] The Gaussian mixture model can capture shapes and density distributions that are more complex than a single Gaussian distribution, such as a cluster containing two parallel production lines. Setting the initial number of components is the first step, which can be determined based on prior knowledge or a preliminary assessment of data characteristics. For example, an initial data cluster may contain a group of devices facing the same direction, but they are clearly divided into two height layers, high and low. In this case, the initial number of components of the Gaussian mixture model can be set to 2. The expectation maximization algorithm begins iterative calculations, continuously adjusting the mean, covariance and weight parameters of the two Gaussian components to maximize the fit to this set of data. Assume that the set convergence condition is that the change in the log-likelihood value between two iterations is less than 0.001. When the algorithm iterates to the 50th time, the change in the log-likelihood value is 0.0008, at which point the iteration stops, and a model consisting of two Gaussian components that can accurately describe the two devices is obtained.

[0058] In an optional embodiment, merging the two data clusters includes:

[0059] The Bhattacharyya distance between the Gaussian mixture models corresponding to the two data clusters is calculated, and when the Bhattacharyya distance is less than a preset merging threshold, merging is performed.

[0060] The Bhattacharyya distance is a measure of the similarity between two probability distributions. The smaller the value, the closer the two distributions are. By calculating the Bhattacharyya distance between the Gaussian mixture models representing two data clusters, we can quantitatively assess their degree of statistical overlap. For example, if two independent clusters C1 and C2 are identified and Gaussian mixture models GMM1 and GMM2 are fitted to them, respectively, and the merging threshold is set to 0.15, the calculated Bhattacharyya distance between GMM1 and GMM2 is 0.1, which is less than the preset merging threshold of 0.15. This indicates that the two clusters have very similar distributions in terms of states such as position and velocity, and C1 and C2 can be merged into a larger data cluster.

[0061] In an optional embodiment, monitoring the spatiotemporal density within the sliding time window and taking the noise point set of the region as a new data set when the spatiotemporal density exceeds a preset germination threshold includes:

[0062] Within a preset sliding time window, the state space is divided into grid cells, and the number of noise points falling into each grid cell is counted; when the number of noise points in any grid cell exceeds the preset germination threshold, the noise point set in the cell is processed as a new data set.

[0063] To discover emerging patterns that have not yet formed stable clusters from discrete data points in the noise, by setting up a grid in time and space and conducting continuous monitoring, we can capture the abnormal clustering of noise points in local areas. Clustering may indicate the formation of a new group or the beginning of a new event. For example, the data is divided into 10 by 10 grid cells, and a 5-minute sliding time window and a germination threshold of 5 points are set. Within the time window from 2:00 pm to 2:05 pm, six data points previously considered to be noise appeared in a specific grid cell. Because 6 exceeded the germination threshold of 5, it was determined that this was no longer random noise, but a meaningful clustering event. Therefore, these six points were used as a new initial data set and sent to the first step of the clustering process to attempt to identify and track this newly emerged target group.

[0064] In an optional embodiment, the parameter set of the Gaussian mixture model includes the weight, mean vector and covariance matrix of each Gaussian component.

[0065] The three parameters together define the complete characteristics of each Gaussian component in the Gaussian mixture model, enabling a detailed mathematical description of complex data distributions. The weight represents the proportion of the component in the entire mixture model, the mean vector indicates the center position of the component in the multidimensional space, and the covariance matrix describes the shape, size, and direction of the component.

[0066] The present application also provides a cluster-based control system data statistics system, comprising:

[0067] a core object acquisition unit configured to acquire multi-dimensional time series state data of the control system; for each data point, counting the number of neighboring data points within a spatiotemporal neighborhood defined by a preset Mahalanobis distance threshold and a temporal proximity threshold; and defining the data point as a core object when the number is greater than a core object determination threshold;

[0068] The merging unit is used to start from any unvisited core object, recursively expand it based on the density reachability principle to form an initial data cluster, and use the expectation maximization algorithm to fit a Gaussian mixture model to each initial data cluster, extracting the parameter set of the Gaussian mixture model as the statistical characteristics of the cluster; it iteratively calculates the Bhattacharyya distance between the Gaussian mixture models of any two data clusters. When the distance is less than a preset merging threshold, the two data clusters are merged and the Gaussian mixture model is refitted to the merged new cluster until no clusters can be merged;

[0069] The result output unit is used to calculate the average log-likelihood of all data points in any data cluster relative to its Gaussian mixture model. When the average log-likelihood is lower than a preset splitting threshold, the data points in the cluster are regarded as independent data sets, and re-clustering and merging are performed on them; data points that do not belong to any cluster are regarded as noise points, and their spatiotemporal density is monitored within a sliding time window. When the spatiotemporal density exceeds a preset germination threshold, the noise point set in the area is used as a new data set, and the cluster generation, merging and splitting steps are performed on the new data set to generate a new cluster; and the Gaussian mixture model parameter set of all current data clusters is output as the statistical result of the current operating condition of the control system.

[0070] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. In addition, the various different implementations of the embodiments of the present invention can also be arbitrarily combined, as long as they do not violate the ideas of the embodiments of the present invention, and they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A cluster-based control system data statistics method, characterized in that: include: Acquire multi-dimensional time series state data of the control system; For each data point, the number of neighboring data points in the spatiotemporal neighborhood defined by a preset Mahalanobis distance threshold and a temporal proximity threshold is counted. When the number is greater than the core object determination threshold, the data point is defined as a core object. Starting from any unvisited core object, recursively expand it using the density reachability principle to form initial data clusters. Then, use the expectation-maximization algorithm to fit a Gaussian mixture model to each initial data cluster, and extract the parameter set of the Gaussian mixture model as the statistical characteristics of the cluster. Iteratively calculate the Bhattacharyya distance between the Gaussian mixture models of any two data clusters. When the distance is less than a preset merging threshold, the two data clusters are merged and the Gaussian mixture model is refitted to the merged new cluster until no clusters can be merged. Calculate the average log-likelihood of all data points in any data cluster relative to its Gaussian mixture model. When the average log-likelihood is lower than a preset splitting threshold, treat the data points in the cluster as independent data sets and re-generate and merge them into new clusters. Treat data points that do not belong to any cluster as noise points and monitor their spatiotemporal density within a sliding time window. When the spatiotemporal density exceeds a preset germination threshold, treat the noise point set in the region as a new data set and perform cluster generation, merging, and splitting steps on the new data set to generate a new cluster. Output the Gaussian mixture model parameter sets of all current data clusters as the statistical results of the current working condition of the control system.

2. The method according to claim 1, characterized in that The data points are defined as core objects, including: A preset Mahalanobis distance threshold, a temporal proximity threshold, and a core object determination threshold are set; for any data point, if the number of its neighboring data points in the spatiotemporal neighborhood defined by the Mahalanobis distance threshold and the temporal proximity threshold exceeds the core object determination threshold, the data point is defined as a core object.

3. The method according to claim 1, characterized in that The method of fitting a Gaussian mixture model to each initial data cluster using the expectation maximization algorithm includes: The initial number of components of the Gaussian mixture model is set for the initial data cluster to be fitted, and the expectation maximization algorithm is used for iterative fitting until the change in the log-likelihood value of the model meets the preset convergence condition.

4. The method according to claim 1, wherein The merging of the two data clusters comprises: The Bhattacharyya distance between the Gaussian mixture models corresponding to the two data clusters is calculated, and when the Bhattacharyya distance is less than a preset merging threshold, merging is performed.

5. The method according to claim 1, wherein The temporal and spatial density is monitored within the sliding time window. When the temporal and spatial density exceeds a preset germination threshold, the noise point set of the region is used as a new data set, including: Within a preset sliding time window, the state space is divided into grid cells, and the number of noise points falling into each grid cell is counted; when the number of noise points in any grid cell exceeds the preset germination threshold, the noise point set in the cell is processed as a new data set.

6. The method according to claim 1, characterized in that The parameter set of the Gaussian mixture model includes the weight, mean vector and covariance matrix of each Gaussian component.

7. A cluster-based control system data statistics system, characterized in that: include: A core object acquisition unit is used to acquire multi-dimensional time series state data of the control system; For each data point, the number of neighboring data points in the spatiotemporal neighborhood defined by a preset Mahalanobis distance threshold and a temporal proximity threshold is counted. When the number is greater than the core object determination threshold, the data point is defined as a core object. The merging unit is used to start from any unvisited core object, recursively expand it based on the density reachability principle to form an initial data cluster, and use the expectation maximization algorithm to fit a Gaussian mixture model to each initial data cluster, extracting the parameter set of the Gaussian mixture model as the statistical characteristics of the cluster; it iteratively calculates the Bhattacharyya distance between the Gaussian mixture models of any two data clusters. When the distance is less than a preset merging threshold, the two data clusters are merged and the Gaussian mixture model is refitted to the merged new cluster until no clusters can be merged; The result output unit is used to calculate the average log-likelihood of all data points in any data cluster relative to its Gaussian mixture model. When the average log-likelihood is lower than a preset splitting threshold, the data points in the cluster are regarded as independent data sets, and cluster generation and merging are performed on them again; data points that do not belong to any cluster are regarded as noise points, and their spatiotemporal density is monitored within a sliding time window. When the spatiotemporal density exceeds a preset germination threshold, the noise point set in the region is used as a new data set, and the cluster generation, merging and splitting steps are performed on the new data set to generate a new cluster; Output the Gaussian mixture model parameter sets of all current data clusters as the statistical results of the current working condition of the control system.

8. The system according to claim 7, characterized in that The data points are defined as core objects, including: A preset Mahalanobis distance threshold, a temporal proximity threshold, and a core object determination threshold are set; for any data point, if the number of its neighboring data points in the spatiotemporal neighborhood defined by the Mahalanobis distance threshold and the temporal proximity threshold exceeds the core object determination threshold, the data point is defined as a core object.

9. The system according to claim 7, wherein: The method of fitting a Gaussian mixture model to each initial data cluster using the expectation maximization algorithm includes: The initial number of components of the Gaussian mixture model is set for the initial data cluster to be fitted, and the expectation maximization algorithm is used for iterative fitting until the change in the log-likelihood value of the model meets the preset convergence condition.

10. The system according to claim 7, wherein: The merging of the two data clusters comprises: The Bhattacharyya distance between the Gaussian mixture models corresponding to the two data clusters is calculated, and when the Bhattacharyya distance is less than a preset merging threshold, merging is performed.

Images