A concentrator intelligent fusion terminal based on multi-sensor monitoring maintenance

Abstract

The application relates to the technical field of fault monitoring, in particular to a concentrator intelligent fusion terminal based on multi-sensor monitoring and maintenance; a suspected fault time is obtained according to the discrete characteristics of data distribution in a state data sequence of any dimension; a target time is obtained according to the data difference characteristics of the suspected fault time and an adjacent time; state representation points in a multi-dimensional space are constructed according to the data of all dimensions of the target time and clustering is carried out, and the fault representation credibility of a state cluster is obtained according to the distribution characteristics of the state representation points in the state cluster and the data discrete characteristics between the state representation points. The abnormal degree of the latest time is obtained according to the distance characteristics of the state representation point corresponding to the latest time and the nearest state cluster, the range characteristics of the nearest state cluster and the fault representation credibility; the running state of the concentrator is monitored according to the abnormal degree, and the monitoring accuracy of the concentrator terminal is improved.

Description

Technical Field

[0001] This invention relates to the field of fault monitoring technology, and specifically to a concentrator intelligent fusion terminal based on multi-sensor monitoring and maintenance. Background Technology

[0002] A concentrator intelligent fusion terminal is an intelligent terminal device that integrates functions such as data acquisition, processing, transmission, and control. It is mainly used for centralized data management and intelligent fusion control in industries such as power. Concentrators require frequent maintenance and repair during operation to ensure normal operation. Monitoring with multiple sensors can further enable intelligent and automated management and maintenance of the concentrator terminal.

[0003] Current technologies for monitoring concentrator smart terminals typically perform sensor monitoring independently, with each sensor determining anomalies based on its own set thresholds. This ignores the fact that equipment faults often manifest across multiple dimensions. For example, changes in monitoring data such as temperature rises detected by a temperature sensor or fan speed changes detected by a speed sensor do not definitively indicate the presence of an anomaly. However, simultaneously detecting both temperature rises and abnormal fan speeds provides a stronger indication that a fan malfunction is causing abnormal temperature changes at the concentrator terminal. Therefore, current technologies that rely solely on thresholds set for each sensor for concentrator terminal anomaly detection are prone to false alarms, reducing the accuracy of terminal anomaly monitoring. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a smart fusion terminal for concentrator monitoring and maintenance based on multi-sensor technology. The specific technical solution adopted is as follows:

[0005] The data acquisition module is used to acquire status data sequences of different dimensions of the monitoring concentrator's operating status;

[0006] The data analysis module is used to obtain the suspected fault time based on the discrete characteristics of the data distribution in the state data sequence of any dimension; and to obtain the target time based on the data difference characteristics between the suspected fault time and adjacent times in the arbitrary dimension.

[0007] The state cluster analysis module is used to construct state representation points in a multi-dimensional space based on data from all dimensions at the target time; cluster all state representation points to obtain different state clusters; and obtain the fault representation reliability of the state clusters based on the distribution characteristics of the state representation points in the state clusters and the data discreteness characteristics between the state representation points.

[0008] The fault monitoring module is used to obtain the nearest state cluster based on the state representation points corresponding to all dimensions of data at the latest time; to obtain the degree of anomaly at the latest time based on the distance characteristics between the state representation points corresponding to the latest time and the nearest state cluster, the range characteristics of the nearest state cluster, and the reliability of the fault representation; and to monitor the operating status of the concentrator based on the degree of anomaly.

[0009] Furthermore, the step of obtaining the suspected fault moment based on the discrete characteristics of the data distribution in the state data sequence of any dimension includes:

[0010] Calculate the absolute value of the difference between the data at any time in the state data sequence and the average value of the data at all times, and normalize it to obtain the data difference degree at any time; the time when the data difference degree exceeds a preset difference threshold is taken as the suspected fault time corresponding to the arbitrary dimension.

[0011] Furthermore, the step of obtaining the target time based on the data difference features between the suspected fault time and adjacent times in the arbitrary dimension includes:

[0012] In any given dimension, the sum of the absolute values ​​of the differences between the suspected fault time and its immediate neighbors is calculated to obtain the neighbor difference value; the reciprocal of the neighbor difference value is calculated and normalized to obtain the neighborhood similarity of the suspected fault time; the suspected fault time whose neighborhood similarity exceeds a preset similarity threshold is taken as the target time.

[0013] Furthermore, the step of clustering all state representation points to obtain different state clusters includes:

[0014] Clustering is performed using the K-means clustering algorithm based on the distance between state representation points to obtain different state clusters.

[0015] Furthermore, the step of obtaining the fault representation reliability of the state cluster based on the distribution characteristics of the state representation points in the state cluster and the data discrete characteristics between the state representation points includes:

[0016] Calculate the average Euclidean distance between all state representation points in the state cluster and their nearest other state representation points to obtain the average nearest distance; calculate the reciprocal of the average nearest distance and normalize it to obtain the intra-cluster density; calculate the reciprocal of the Euclidean distance between the two farthest state representation points in the state cluster and normalize it to obtain the range concentration; calculate the sum of the standard deviations of all dimensions corresponding to all state representation points in the state cluster to obtain the comprehensive dispersion; calculate the reciprocal of the comprehensive dispersion and normalize it to obtain the distribution uniformity; calculate the sum of the intra-cluster density and the range concentration to obtain the cluster feature value; calculate the product of the cluster feature value and the distribution uniformity and positively correlate them to obtain the fault representation reliability of the state cluster.

[0017] Furthermore, the step of obtaining the most recent state cluster based on the state representation points corresponding to all dimensions of data at the latest time includes:

[0018] The state cluster closest to the state representation point corresponding to the latest time is taken as the nearest state cluster.

[0019] Further, the step of obtaining the degree of anomaly at the latest moment based on the distance feature between the state representation point corresponding to the latest moment and the nearest state cluster, the range feature of the nearest state cluster, and the reliability of the fault representation includes:

[0020] Calculate the Euclidean distance between the state representation point corresponding to the latest time and the cluster center of the nearest state cluster to obtain the interval distance; calculate the maximum Euclidean distance between the state representation point and the cluster center in the nearest state cluster to obtain the maximum distance within the cluster; calculate the ratio of the maximum distance within the cluster to the interval distance to obtain the proximity; calculate the product of the proximity and the reliability of the fault representation of the nearest state cluster to obtain the anomaly degree at the latest time.

[0021] Furthermore, the step of monitoring the operating status of the concentrator based on the degree of anomaly includes:

[0022] When the degree of abnormality exceeds the preset abnormality threshold, the concentrator's operating status becomes abnormal.

[0023] The present invention has the following beneficial effects:

[0024] In this invention, acquiring suspected fault times allows for the determination of potential concentrator failure times based on the discrete characteristics of historical data distribution. Since not all suspected fault times are caused by actual data anomalies, acquiring target times can eliminate times that do not reflect historical data anomalies, improving the accuracy of subsequent fault monitoring. Acquiring state representation points reflects the data distribution across various dimensions at the target time, and acquiring different state clusters distinguishes state representation points representing different multidimensional data characteristics. Simultaneously, acquiring state clusters facilitates final fault monitoring of the concentrator at the latest time. The distribution of state representation points within a state cluster reflects the similarity of data across different dimensions and the correlation of multidimensional data, thereby determining whether the state cluster accurately reflects the concentrator's fault. Therefore, acquiring fault representation credibility reflects the credibility of the state cluster in representing the concentrator's fault, further improving the accuracy of final fault monitoring. Acquiring the most recent state cluster determines the state clusters that the state representation point at the latest time is close to, thereby determining whether the state representation point represents a concentrator fault. Obtaining the degree of anomaly can accurately reflect whether the concentrator has malfunctioned at the latest moment; finally, monitoring is carried out based on the degree of anomaly. Compared with the method of judging anomalies based solely on the threshold set by each sensor, this reduces the false alarm rate and improves the accuracy of anomaly monitoring of the concentrator terminal. Attached Figure Description

[0025] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a block diagram of a concentrator intelligent fusion terminal based on multi-sensor monitoring and maintenance, provided as an embodiment of the present invention. Detailed Implementation

[0027] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a concentrator intelligent fusion terminal based on multi-sensor monitoring and maintenance proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0029] The following description, in conjunction with the accompanying drawings, details a specific solution for a concentrator intelligent fusion terminal based on multi-sensor monitoring and maintenance provided by the present invention.

[0030] Please see Figure 1 The diagram illustrates a block diagram of a smart fusion terminal for concentrators based on multi-sensor monitoring and maintenance, according to an embodiment of the present invention. The terminal includes the following modules:

[0031] The data acquisition module S1 is used to acquire status data sequences of different dimensions of the monitoring concentrator's operating status.

[0032] Multiple sensors are installed on the intelligent fusion terminal of the concentrator for data monitoring. In this embodiment, temperature, humidity, voltage and current, fan speed, and smoke sensors are used to monitor the terminal. For data fusion analysis, all sensors collect data at the same frequency, once per second in this embodiment. The implementer can determine the data type and frequency based on the implementation scenario. The data sequences collected by each sensor are normalized. In this embodiment, linear normalization is used to obtain state data sequences of different dimensions for monitoring the concentrator's operating status. Each dimension's state data sequence reflects the concentrator's operating status characteristics. In this embodiment, the state data sequence covers six months of historical data prior to the latest time. The state data sequence is updated with the latest update, and the implementer can determine the appropriate parameters based on the implementation scenario.

[0033] The data analysis module S2 is used to obtain the suspected fault time based on the discrete characteristics of the data distribution in the state data sequence of any dimension; and to obtain the target time based on the data difference characteristics between the suspected fault time and the adjacent time in any dimension.

[0034] Because the time range of the status data sequence is relatively wide, it will include data from various abnormal states during the operation of the concentrator. Therefore, the first step is to determine the moments in the status data sequence that may represent concentrator malfunctions. During monitoring, since the duration of a fault only accounts for a very small portion of the concentrator's operation, the concentrator operates normally for most of the time, and the status data corresponding to normal periods is relatively stable. Therefore, the average level of the status data sequence will be close to the data level during normal operation. Thus, suspected fault moments can be obtained based on the discrete characteristics of the data distribution in the status data sequence of any dimension. Preferably, in this embodiment of the invention, the step of obtaining suspected fault moments includes: calculating the absolute value of the difference between the data at any moment in the status data sequence and the average data value at all moments, and normalizing it to obtain the data difference degree at that arbitrary moment. The more normal the operation at that arbitrary moment, the closer its data is to the average data value at all moments, and the smaller the data difference degree; the more abnormal the operation at that arbitrary moment, the more its abnormal data deviates from the average data value at all moments, and the greater the data difference degree. The moment when the data difference exceeds a preset difference threshold is taken as the suspected fault moment corresponding to that arbitrary dimension. The data difference is obtained by linear normalization. For data at normal times, the data difference will tend to 0 after linear normalization. However, for data at obvious fault or abnormal moments, the data difference will tend to 1 after linear normalization. The two tend to be at opposite ends of the value range. Therefore, in this embodiment of the invention, the preset difference threshold is 0.5. Exceeding the preset difference threshold means that the arbitrary moment has obvious data abnormality characteristics, and it is more likely to characterize the concentrator fault or abnormality under that arbitrary dimension.

[0035] Furthermore, since the sensor may generate occasional noise data during data acquisition, the values ​​of which may differ significantly from the average data level and be considered as suspected fault moments, it is necessary to identify the true abnormal moments among the suspected fault moments in order to improve the accuracy of subsequent concentrator anomaly monitoring. Because noise data exhibits random and irregular mutations, while concentrator malfunctions typically last for a period of time, the true abnormal data moments exhibit a continuous and concentrated characteristic. Therefore, the target moment can be obtained based on the data difference characteristics between the suspected fault moment and adjacent moments in any dimension. Preferably, in this embodiment of the invention, the step of obtaining the target moment includes: calculating the sum of the absolute values ​​of the differences between the suspected fault moment and the adjacent moments in any dimension to obtain the adjacent difference value; the larger the sum of the absolute values ​​of the differences between the suspected fault moment and the adjacent moments, the larger the adjacent difference value, meaning that the discrete mutation characteristics of the suspected fault moment are more obvious, and the more likely the suspected fault moment is caused by noise data. The reciprocal of the adjacent difference values ​​is calculated and normalized to obtain the neighborhood similarity of the suspected fault time. The larger the neighborhood similarity, the more similar the data between adjacent times is, and the more likely the suspected fault time represents the real data anomaly time. The suspected fault time with a neighborhood similarity exceeding a preset similarity threshold is taken as the target time. Since the neighborhood similarity is obtained by linear normalization, the neighborhood similarity of the suspected fault time corresponding to noise tends to 0, while the neighborhood similarity of the suspected fault time corresponding to real data anomaly tends to 1, and the two tend to be at opposite ends of the value range. Therefore, in this embodiment of the invention, the preset similarity threshold is 0.5. The target time represents the time when the concentrator has real data anomalies in different dimensions in history.

[0036] The state cluster analysis module S3 is used to construct state representation points in a multidimensional space based on data from all dimensions at the target time; cluster all state representation points to obtain different state clusters; and obtain the fault representation credibility of the state clusters based on the distribution characteristics of state representation points in the state clusters and the data discrete characteristics between state representation points.

[0037] After obtaining the target time in all dimensions, the multidimensional correlation features of the fault can be analyzed. First, state representation points in the multidimensional space are constructed based on the data of all dimensions of the target time. For example, in the state data sequence corresponding to the temperature dimension, if a certain time is the target time, then the data of all dimensions corresponding to that target time are taken to construct a state representation point in the multidimensional space. Since the state representation points corresponding to all historical target times are distributed in the multidimensional space, all state representation points can be clustered to obtain different state clusters. Preferably, in this embodiment of the invention, clustering is performed using the K-means clustering algorithm based on the distance between state representation points to obtain different state clusters. It should be noted that the K-means clustering algorithm is existing technology, and the number of clusters is obtained using the elbow method; the specific clustering steps will not be elaborated further. Clustering can group state representation points with similar multidimensional data correlation features into a cluster. For example, when a concentrator experiences a circuit abnormality due to high humidity, the monitoring dimensions of humidity, voltage, and current will show anomalies. Therefore, in the multidimensional space, state representation points with high anomalies in these three dimensions and low anomalies in other dimensions will form state clusters corresponding to abnormal fault behaviors. After clustering is completed, each state cluster in the clustering results can reflect a type of concentrator failure.

[0038] Furthermore, the denser and more concentrated the distribution of state representation points within a state cluster, the higher the similarity of the data across different dimensions of the state representation points within that cluster. This indicates that the state cluster can accurately reflect the fault type under multi-dimensional data association features, and the reliability of representing the fault type through this state cluster is high. Conversely, when the distribution of state representation points within a state cluster exhibits a non-uniform characteristic of being partially discrete and partially dense, it indicates poor density of the state representation points, low similarity of data in each dimension, and weak similarity of multi-dimensional data association features. This means that the state cluster is unlikely to accurately represent a certain fault type of the concentrator, and the reliability of its fault representation is low. Therefore, the reliability of the fault representation of a state cluster can be obtained based on the distribution characteristics of state representation points within the cluster and the discreteness of the data between state representation points.

[0039] Preferably, in this embodiment of the invention, the step of obtaining the reliability of the fault representation includes: calculating the average Euclidean distance between all state representation points in the state cluster and the nearest other state representation points to obtain the average nearest distance; the average nearest distance represents the proximity between state representation points within the cluster. The smaller the average nearest distance, the denser the state representation points within the cluster. Calculating and normalizing the reciprocal of the average nearest distance to obtain the cluster density; the larger the cluster density, the higher the similarity of the data in each dimension, the more obvious the multidimensional data correlation features, and the higher the reliability of the state cluster in reflecting the concentrator fault. Calculating and normalizing the reciprocal of the Euclidean distance between the two farthest state representation points in the state cluster to obtain the range concentration; the larger the range concentration, the smaller the range of the state cluster, the more concentrated the state representation points, and the higher the reliability of the state cluster in reflecting the concentrator fault. The sum of the standard deviations of all dimensions corresponding to all state representation points in a state cluster is calculated to obtain the comprehensive dispersion. The comprehensive dispersion reflects the degree of data dispersion in each dimension of the state representation points within the cluster. A smaller comprehensive dispersion indicates a more uniform distribution of values ​​across all dimensions, and thus a more uniform and consistent distribution of state representation points within the cluster. The reciprocal of the comprehensive dispersion is calculated and normalized to obtain the distribution uniformity. A larger distribution uniformity indicates a higher reliability of the state cluster in representing concentrator faults. The sum of the cluster density and the range concentration is calculated to obtain the cluster class characteristic value. A larger cluster class characteristic value indicates a smaller range for the state cluster and closer distances between adjacent state representation points. The product of cluster feature values ​​and distribution uniformity is calculated and positively correlated to obtain the fault characterization reliability of the state cluster. When multiple dense clusters of state characterization points exist within a state cluster, and there is a certain distance between each cluster, the values ​​of intra-cluster density and range concentration are still relatively large. However, because the state characterization points within the cluster exhibit multiple cluster distributions, the intra-cluster density feature is weak, leading to errors in the fault characterization reliability. Therefore, by calculating the product of cluster feature values ​​and distribution uniformity, the accuracy of the fault characterization reliability of the state cluster is further improved. A higher fault characterization reliability means that the state cluster is more able to reflect the concentrator's fault anomaly. Obtaining the fault characterization reliability can improve the accuracy of subsequent judgments on whether the latest time is abnormal. The formula for obtaining the fault characterization reliability includes:

[0040]

[0041] In the formula, R represents the reliability of the fault characterization. This indicates normalization, and L represents the average nearest distance. H represents the cluster density, and H represents the Euclidean distance between the two farthest state representation points in the state cluster. Indicates the concentration of the range. S represents the cluster feature value, and S represents the overall dispersion. This indicates the uniformity of distribution. The constant 2 in the denominator is used to perform a positive correlation mapping on the numerator. The numerator's range is 0 to 2, and after the positive correlation mapping, it becomes 0 to 1, which facilitates subsequent analysis.

[0042] The fault monitoring module S4 is used to obtain the most recent state cluster based on the state representation points corresponding to all dimensions of data at the latest time; to obtain the degree of anomaly at the latest time based on the distance characteristics between the state representation points corresponding to the latest time and the most recent state cluster, the range characteristics of the most recent state cluster, and the reliability of the fault representation; and to monitor the operating status of the concentrator based on the degree of anomaly.

[0043] After obtaining the fault representation reliability of different state clusters, it is possible to determine whether the concentrator has experienced a fault or anomaly based on the state representation points corresponding to the latest time. First, the nearest state cluster is obtained based on the state representation points corresponding to all dimensions of data at the latest time. Specifically, the state cluster closest to the state representation point corresponding to the latest time is taken as the nearest state cluster. The closer the state representation point at the latest time is to this nearest state cluster, the closer the state characteristics represented by the state representation point at the latest time are to the state characteristics represented by the nearest state cluster. Therefore, the degree of anomaly at the latest time is obtained based on the distance characteristics between the state representation point corresponding to the latest time and the nearest state cluster, the range characteristics of the nearest state cluster, and the fault representation reliability.

[0044] Preferably, in this embodiment of the invention, the step of obtaining the degree of anomaly includes: calculating the Euclidean distance between the state representation point corresponding to the latest time and the cluster center of the nearest state cluster to obtain the interval distance; the smaller the interval distance, the more likely the state representation point belongs to the nearest state cluster. Calculating the maximum Euclidean distance between the state representation point in the nearest state cluster and the cluster center to obtain the maximum distance within the cluster; the maximum distance within the cluster reflects the size of the state cluster. Calculating the ratio of the maximum distance within the cluster to the interval distance to obtain the proximity; the larger the interval distance is than the maximum distance within the cluster, the smaller the proximity, meaning the state representation point is farther from the nearest state cluster, and the lower the probability of the concentrator failing at the latest time; the smaller the interval distance is than the maximum distance within the cluster, the larger the proximity, meaning the state representation point is within the nearest state cluster and close to the cluster center, and the more likely the concentrator is to experience an anomaly at the latest time. The anomaly level at the latest time is obtained by multiplying the proximity degree by the reliability of the fault representation of the most recent state cluster. A higher reliability of the fault representation of the most recent state cluster means a higher reliability of the cluster representing a concentrator fault. Furthermore, the closer the state representation point is to the most recent state cluster, the greater the probability of an anomaly occurring in the concentrator at that latest time, and the higher the anomaly level. Therefore, a higher anomaly level indicates that the state representation point at the latest time is closer to the most recent state cluster, and the most recent state cluster has a higher reliability representing a concentrator fault. The formula for obtaining the anomaly level includes:

[0045]

[0046] In the formula, F represents the anomaly level at the latest time, D represents the maximum distance within the cluster, and d represents the interval distance. Indicates proximity. This indicates the reliability of the fault representation of the most recent state cluster.

[0047] Ultimately, the operating status of the concentrator can be monitored based on the degree of anomaly. Specifically, when the degree of anomaly exceeds a preset anomaly threshold, the concentrator's operating status is abnormal. In this embodiment of the invention, the preset anomaly threshold is 0.6, which can be determined by the implementer according to the implementation scenario. When the concentrator malfunctions, a timely warning is issued and maintenance personnel are notified to perform inspection and maintenance. This enables the concentrator's intelligent terminal to automatically monitor faults and anomalies through multiple sensors, reducing the frequency of routine maintenance. Thus, by analyzing the multi-dimensional data correlation features of multiple sensors to determine whether the concentrator's intelligent terminal is abnormal, compared to methods that rely solely on each sensor's own set threshold for anomaly judgment, the false alarm rate is reduced, and the accuracy of concentrator terminal anomaly monitoring is improved.

[0048] In summary, this invention provides an intelligent fusion terminal for concentrator monitoring and maintenance based on multi-sensor monitoring. It obtains suspected fault times based on the discrete characteristics of data distribution in a state data sequence of any dimension; obtains target times based on the data difference characteristics between suspected fault times and adjacent times; constructs state representation points in a multi-dimensional space based on data from all dimensions of the target time and performs clustering; obtains the fault representation reliability of state clusters based on the distribution characteristics of state representation points within the state clusters and the discrete characteristics of data between state representation points; and obtains the nearest state cluster based on the state representation point corresponding to the latest time. This invention obtains the anomaly degree of the latest time based on the distance characteristics between the state representation point corresponding to the latest time and the nearest state cluster, the range characteristics of the nearest state cluster, and the fault representation reliability; and monitors the operating status of the concentrator based on the anomaly degree, thereby improving the monitoring accuracy of the concentrator terminal.

[0049] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0050] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A concentrator intelligent fusion terminal based on multi-sensor monitoring and maintenance, characterized in that, The terminal includes the following modules: The data acquisition module is used to acquire status data sequences of different dimensions of the monitoring concentrator's operating status; The data analysis module is used to obtain the suspected fault time based on the discrete characteristics of the data distribution in the state data sequence of any dimension; and to obtain the target time based on the data difference characteristics between the suspected fault time and adjacent times in the arbitrary dimension. The state cluster analysis module is used to construct state representation points in a multi-dimensional space based on data from all dimensions at the target time; cluster all state representation points to obtain different state clusters; and obtain the fault representation reliability of the state clusters based on the distribution characteristics of the state representation points in the state clusters and the data discreteness characteristics between the state representation points. The fault monitoring module is used to obtain the most recent state cluster based on the state representation points corresponding to all dimensions of data at the latest time. The degree of anomaly at the latest moment is obtained based on the distance characteristics between the state representation point corresponding to the latest moment and the nearest state cluster, the range characteristics of the nearest state cluster, and the reliability of the fault representation. The operating status of the concentrator is monitored based on the degree of anomaly. The step of obtaining the fault representation reliability of the state cluster based on the distribution characteristics of the state representation points in the state cluster and the data discrete characteristics between the state representation points includes: Calculate the average Euclidean distance between all state representation points in the state cluster and their nearest other state representation points to obtain the average nearest distance; calculate the reciprocal of the average nearest distance and normalize it to obtain the intra-cluster density; calculate the reciprocal of the Euclidean distance between the two farthest state representation points in the state cluster and normalize it to obtain the range concentration; calculate the sum of the standard deviations of all dimensions corresponding to all state representation points in the state cluster to obtain the comprehensive dispersion; calculate the reciprocal of the comprehensive dispersion and normalize it to obtain the distribution uniformity; calculate the sum of the intra-cluster density and the range concentration to obtain the cluster feature value; calculate the product of the cluster feature value and the distribution uniformity and positively correlate them to obtain the fault representation reliability of the state cluster. The step of obtaining the anomaly level of the latest time based on the distance feature between the state representation point corresponding to the latest time and the nearest state cluster, the range feature of the nearest state cluster, and the reliability of the fault representation includes: Calculate the Euclidean distance between the state representation point corresponding to the latest time and the cluster center of the nearest state cluster to obtain the interval distance; calculate the maximum Euclidean distance between the state representation point and the cluster center in the nearest state cluster to obtain the maximum distance within the cluster; calculate the ratio of the maximum distance within the cluster to the interval distance to obtain the proximity; calculate the product of the proximity and the reliability of the fault representation of the nearest state cluster to obtain the anomaly degree at the latest time.

2. The intelligent fusion terminal for concentrator monitoring and maintenance based on multi-sensor monitoring according to claim 1, characterized in that, The step of obtaining the suspected fault moment based on the discrete characteristics of the data distribution in the state data sequence of any dimension includes: Calculate the absolute value of the difference between the data at any time in the state data sequence and the average value of the data at all times, and normalize it to obtain the data difference degree at any time; the time when the data difference degree exceeds a preset difference threshold is taken as the suspected fault time corresponding to the arbitrary dimension.

3. The intelligent fusion terminal for concentrator monitoring and maintenance based on multi-sensor monitoring according to claim 1, characterized in that, The step of obtaining the target time based on the data difference features between the suspected fault time and adjacent times in any dimension includes: In any given dimension, the sum of the absolute values ​​of the differences between the suspected fault time and its immediate neighbors is calculated to obtain the neighbor difference value; the reciprocal of the neighbor difference value is calculated and normalized to obtain the neighborhood similarity of the suspected fault time; the suspected fault time whose neighborhood similarity exceeds a preset similarity threshold is taken as the target time.

4. The intelligent fusion terminal for concentrator monitoring and maintenance based on multi-sensor monitoring according to claim 1, characterized in that, The step of clustering all state representation points to obtain different state clusters includes: Clustering is performed using the K-means clustering algorithm based on the distance between state representation points to obtain different state clusters.

5. The intelligent fusion terminal for concentrator monitoring and maintenance based on multi-sensor monitoring according to claim 1, characterized in that, The step of obtaining the most recent state cluster based on the state representation points corresponding to all dimensions of data at the latest time includes: The state cluster closest to the state representation point corresponding to the latest time is taken as the nearest state cluster.

6. The intelligent fusion terminal for concentrator monitoring and maintenance based on multi-sensor monitoring according to claim 1, characterized in that, The step of monitoring the operating status of the concentrator based on the degree of anomaly includes: When the degree of abnormality exceeds the preset abnormality threshold, the concentrator's operating status becomes abnormal.

Images