A purification system real-time monitoring and repair warning method and system
By clustering and optimizing distance calculations of air quality monitoring data, the problem of low accuracy of air quality monitoring in the existing technology is solved, more efficient outlier data detection is achieved, and the accuracy and reliability of air quality monitoring is improved. .
Patent Information
- Application Number
- CN202411463302.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-21
- Publication Date
- 2024-12-24
- Estimated Expiration
- 2044-10-21
AI Technical Summary
Under the requirements of high accuracy, the existing air quality monitoring system has increased the frequency and dimension of air monitoring data collection, resulting in large amount of calculations. The LOF algorithm has repeated redundant calculations when detecting outliers, and it is impossible to accurately detect air quality abnormalities, resulting in accurate monitoring. Lower sex.
By clustering the air quality monitoring data, determining the state dispersion and trend characteristic values of clusters, optimizing the distance calculation between clusters, improving the LOF algorithm for abnormal detection, and improving the accuracy of air quality monitoring.
It improves the detection accuracy and robustness of outlier data points, and enhances the accuracy and reliability of air quality monitoring.
Smart Images
Figure CN118981735B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of air purification technology, specifically to a method and system for real-time monitoring and early warning of air purification systems. Background Technology
[0002] Factory cleanroom engineering requires factories to more strictly consider the impact of the production environment in their overall planning and design. For example, cleanrooms in pharmaceutical factories, electronics factories, and food factories must meet clean production requirements to ensure product quality. Therefore, configuring air purification systems in factories to purify the production environment is a crucial step in ensuring stable product quality. During the operation of the air purification system, various air monitoring data need to be collected at any given time, and the LOF (Local Outlier Factor) algorithm is used to detect outliers in these air monitoring data to identify abnormal air quality conditions and provide early warnings and maintenance alerts for the air purification system.
[0003] When high monitoring accuracy is required, it is often necessary to increase the collection frequency and data dimensionality of air monitoring data. This results in a massive amount of air monitoring data. When using the LOF algorithm for outlier detection, the computational burden is high due to the need to traverse such a large amount of data and the presence of numerous redundant calculations. To reduce the computational burden, the air monitoring data can be clustered first, and then each cluster can be used as a baseline for outlier detection. However, due to the influence of the cluster data distribution characteristics, using the distance between the centroids of any two clusters as a distance metric often fails to represent the true distance between the clusters as a whole, thus failing to accurately detect outliers and leading to low accuracy in air quality monitoring. Summary of the Invention
[0004] The purpose of this invention is to provide a method and system for real-time monitoring and early warning of air purification systems, in order to solve the problem of low accuracy in existing air quality monitoring.
[0005] To address the aforementioned technical problems, this invention provides a method for real-time monitoring and early warning of a purification system, comprising the following steps:
[0006] Acquire air quality monitoring data at various set times, wherein the air quality monitoring data includes at least two types of air quality parameters;
[0007] Construct the data points corresponding to the air quality monitoring data, cluster all data points to obtain at least two clusters and their cluster centers, and determine the state dispersion of each cluster based on the number of data points in each cluster, the location distribution characteristics, and the adjacency of the data points at a set time.
[0008] Project the data points in any two clusters onto a straight line passing through the class center points of these two clusters to determine the projection points of the data points in any two clusters. Based on the positional distribution of the projection points, the positional relationship between the class center points of any two clusters and their projection points, and the distribution of the surrounding neighboring data points of each data point in each cluster, determine the directional feature value between any two clusters.
[0009] The Euclidean distance between any two clusters is determined. Based on the directional characteristic values between any two clusters and the state dispersion corresponding to any two clusters, the Euclidean distance is improved to determine the improved distance value between any two clusters. Based on the improved distance value, anomaly detection is performed on the air quality monitoring data to determine the air quality status.
[0010] Furthermore, the state discreteness corresponding to each cluster class is determined, including:
[0011] Determine the distance from each data point in each cluster to the cluster center and the distance between any two data points in each cluster;
[0012] Based on the set time corresponding to each data point in each cluster, determine the number of time-series adjacent data points corresponding to each data point in each cluster;
[0013] The state dispersion of each cluster is determined based on the number of data points in each cluster, the dispersion of the distance from each data point to the cluster center, the maximum value of the distance between any two data points, and the number of temporally adjacent data points corresponding to each data point.
[0014] Furthermore, the state discreteness corresponding to each cluster is determined, and the corresponding calculation formula is as follows:
[0015] ;in, Indicates the first The state discreteness corresponding to each cluster class; Indicates the first The first in the cluster class From the data point to the The distance between the class centers of each cluster class; Indicates the first Each data point in each cluster class to the... The average distance between the class centers of each cluster class; Indicates the first The first in the cluster class The number of time-series adjacent data points corresponding to each data point; Indicates the first The maximum value among the distances between any two data points in each cluster class; Indicates the first The set of distances between any two data points in a cluster class; Indicates the first The number of data points in each cluster class; represents the normalization function; e represents the natural constant.
[0016] Furthermore, determining the directional eigenvalues between any two clusters includes:
[0017] The projection point between the projection points of the class center points of any two cluster classes is determined as the target projection point of any two cluster classes.
[0018] Determine the center position of the projection point of the class center point of any two clusters, and determine the distance from the center position to each target projection point corresponding to any two clusters, thereby obtaining the first distance of each target projection point corresponding to any two clusters.
[0019] Determine the distance from the center point of the cluster to the center position of the cluster to which the data point corresponding to each target projection point belongs for any two clusters, thereby obtaining the second distance for each target projection point corresponding to any two clusters;
[0020] Based on the distribution of neighboring data points of each pixel in each cluster, determine the local data point density of each target projection point corresponding to any two clusters;
[0021] Based on the first distance, second distance, and local data point density of each target projection point corresponding to any two clusters, and the proportion of the target projection points of any two clusters in all projection points, determine the directional feature value between any two clusters.
[0022] Furthermore, the directional eigenvalues between any two clusters are determined using the following formula:
[0023] ;in, Indicates the first The first cluster and the first The directional characteristic values among the clusters; Indicates the first The first cluster and the first The first cluster class corresponding to the th The first distance between each target projection point; Indicates the first The first cluster and the first The first cluster class corresponding to the th The second distance between each target projection point; Indicates the first The first cluster and the first The first cluster class corresponding to the th Local data point density of each target projection point; Indicates the first The first cluster and the first The total number of target projection points corresponding to each cluster class; Indicates the first The first cluster and the first The percentage of target projection points corresponding to each cluster class among all projection points; Indicates the first The first cluster and the first The number of all projection points corresponding to each cluster class; e represents the natural constant.
[0024] Furthermore, the improved distance value between any two clusters is determined, and the corresponding calculation formula is as follows:
[0025] ;in, Indicates the first The first cluster and the first Improved distance values between cluster classes; Indicates the first The first cluster and the first The Euclidean distance between the class centers of each cluster class; Indicates the first The positive correlation normalization result of the state discreteness corresponding to each cluster class; Indicates the first The positive correlation normalization result of the state discreteness corresponding to each cluster class; Indicates the first The first cluster and the first The negative correlation normalization result of the directional feature values among the clusters.
[0026] Furthermore, based on the improved distance value between any two clusters, the LOF algorithm is used to detect anomalies in air quality monitoring data.
[0027] Furthermore, constructing the data points corresponding to the air quality monitoring data includes:
[0028] Each type of air quality parameter contained in the air quality monitoring data is determined as a coordinate value on the corresponding coordinate axis in the sample space, and the sample points determined by each coordinate value of the air quality monitoring data in the sample space are determined as the data points corresponding to the air quality monitoring data.
[0029] Furthermore, based on the contour coefficient, it is determined The optimal number of clusters in the algorithm, and utilize The algorithm clusters all data points to obtain the optimal number of clusters.
[0030] To address the aforementioned technical problems, the present invention also provides a real-time monitoring and repair early warning system for a purification system, comprising a processor and a memory. The processor is used to process computer instructions stored in the memory to implement the steps of a real-time monitoring and repair early warning method for a purification system as described in any of the preceding claims.
[0031] This invention offers the following advantages: It constructs data points corresponding to air quality monitoring data at various set times and clusters all data points to obtain multiple clusters, each representing an air quality state. By analyzing the discreteness between data points within each cluster, the state dispersion of each cluster is determined. Simultaneously, by combining the distribution characteristics of data points from any two clusters, the directional characteristics of data points from any two clusters are quantified to measure the probability that the two clusters represent the same air quality state, thus obtaining the directional characteristic value between any two clusters. Based on the state dispersion of each cluster and the directional characteristic values between pairs of clusters, the distance between different clusters is optimized, thereby determining a more accurate distance between different clusters. This effectively improves the detection accuracy and robustness of outlier data points, thereby enhancing the accuracy and reliability of air quality monitoring. Attached Figure Description
[0032] 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.
[0033] Figure 1 This is a flowchart of a real-time monitoring and repair early warning method for a purification system according to an embodiment of the present invention. Detailed Implementation
[0034] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific implementation methods, structures, features, and effects of the technical solution proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. 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.
[0035] 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. Furthermore, all parameters or indices in the formulas discussed herein are normalized values to eliminate the influence of dimensions.
[0036] Example of a real-time monitoring and early warning method for purification systems:
[0037] To address the issue of low accuracy in existing air quality monitoring systems, this embodiment provides a method for real-time monitoring and early warning of air purification systems. The corresponding flowchart for this method is shown below. Figure 1 As shown, it includes the following steps:
[0038] Step S1: Obtain air quality monitoring data at each set time, wherein the air quality monitoring data includes at least two types of air quality parameters.
[0039] During the operation of the purification system, air quality monitoring data needs to be collected to effectively monitor the quality of the purified air. This data includes various gaseous components such as oxygen and carbon dioxide, as well as various air quality parameters such as temperature and humidity. When collecting this data, various air quality parameters are collected synchronously according to a set sampling frequency. Each sampling moment is designated as a set time point, allowing for real-time access to various types of air quality parameters for each set time point within the most recent historical period.
[0040] Step S2: Construct the data points corresponding to the air quality monitoring data, cluster all data points to obtain at least two clusters and their cluster centers, and determine the state dispersion of each cluster based on the number of data points in each cluster, the location distribution characteristics, and the adjacency of the data points at a set time.
[0041] For each set time of air quality monitoring data, that is, various types of air quality parameters, corresponding data points are constructed. That is, each type of air quality parameter is used as the coordinate value on the corresponding coordinate axis of the sample space, and the sample points determined by all coordinate values in the sample space are determined as the corresponding data points.
[0042] After obtaining all the data points, due to the large number and high dimensionality of the data points, directly using the LOF algorithm to detect outliers would result in excessive computation. However, considering that the same type of data points often corresponds to a certain air quality state, in order to reduce the computational load to a greater extent and to monitor air quality state more reasonably, we can first cluster all the data points and then perform outlier detection on each cluster representing a certain air quality state, thereby ultimately achieving the monitoring of the purified air quality.
[0043] Based on the above analysis, the Euclidean distance between any two data points in the sample space is determined and used as a distance metric between pairwise data points. This is then used to cluster all data points in the sample space, resulting in multiple clusters and the cluster centroids for each cluster. In this embodiment, clustering is performed using... The k-means clustering algorithm clusters all data points in the sample space. The specific process involves determining the silhouette coefficient. The optimal number of clusters in the algorithm is The specific steps for determining the optimal number of clusters are existing technology and will not be elaborated here. Based on the determined optimal number of clusters... Perform on all data points in the sample space Clustering, thus obtaining Individual categories and The cluster centers of each cluster contain multiple data points, and each cluster represents a specific air quality state. It should be understood that since outlier data points are subsequently detected based on clusters, an insufficient number of clusters may lead to inaccurate results. Therefore, the number of clusters is limited here. Minimum value limit is applied, if obtained through the profile coefficient. If the value is less than or equal to the set threshold for the number of clusters, then directly... The value is set to the threshold for the number of clusters. The specific value of the threshold for the number of clusters can be reasonably set according to needs. In this embodiment, the value of the threshold for the number of clusters is set to 20.
[0044] When each cluster is used as an object in the LOF algorithm, the conventional approach is to use the cluster centroid of each cluster as the reference, and calculate the distance between clusters based on the Euclidean distance between the cluster centroids, and then perform the LOF algorithm. However, in practical applications, using the Euclidean distance between the cluster centroids of any two clusters as the distance metric often fails to represent the actual differences between the clusters as a whole. For example, different air quality states correspond to different degrees of clustering of data points, and the air quality monitoring data representing that cluster may be relatively more disordered, meaning it does not approximate a specific air quality state as mentioned above. Instead, its dispersion forces most data points to be clustered together. For these clusters, adjustments are needed when calculating the distance metric between clusters to improve the detection accuracy and robustness of outlier clusters, thereby improving the accuracy of air quality monitoring.
[0045] Based on the above analysis, the degree of dispersion of data points in each cluster is analyzed to characterize the clustering degree of the current cluster. Furthermore, since each cluster should correspond to a certain control quality state, and states typically last for a period of time, the state dispersion of data points in each cluster is determined to obtain the state dispersion of each cluster. The corresponding calculation formula is as follows:
[0046] ;in, Indicates the first The state discreteness corresponding to each cluster class; Indicates the first The first in the cluster class From the data point to the The distance between the class centers of each cluster class; Indicates the first Each data point in each cluster class to the... The average distance between the class centers of each cluster class; Indicates the first The first in the cluster class The number of time-series adjacent data points corresponding to each data point; Indicates the first The maximum value among the distances between any two data points in each cluster class; Indicates the first The set of distances between any two data points in a cluster class; Indicates the first The number of data points in each cluster class; represents the normalization function; e represents the natural constant.
[0047] In the above-mentioned section In the formula for calculating the state discreteness corresponding to the cluster class, the calculation of the th cluster class is performed by... Each data point in each cluster to the... The difference between the distance to the class centroid of each cluster and the mean distance can characterize the value of the first cluster. The degree of overall difference in distance between data points within a cluster is considered. A higher degree of overall difference in distance indicates a more discrete distribution of data points within that cluster, and consequently, a greater degree of state dispersion for that cluster. This is determined by... The number of temporally adjacent data points corresponding to each data point in each cluster class, where the number of temporally adjacent data points refers to the number of data points in the first cluster. The total number of consecutive adjacent data points at a given time corresponding to each data point in each cluster class is used, and an exponential function is used to perform a negative correlation mapping on the number of adjacent data points in this time series. Since the position of a data point in the sample space and the distance between data points are essentially determined by the magnitude of its own multi-dimensional air quality parameters, therefore, in the first... The more temporally adjacent data points a cluster contains, the longer the duration of relatively small changes in the data point values, and the closer the corresponding cluster is to a certain air quality state. Conversely, for data points with a small number of temporally adjacent data points in a cluster, we tend to consider them as discrete outliers. Therefore, when determining whether a cluster primarily represents a certain air quality state, we prefer to give higher weight to the degree of dispersion represented by the distance differences determined by these discrete outliers. By determining the first... The farthest distance between any two data points within a cluster is considered. A greater distance indicates a more dispersed distribution of data points within that cluster, meaning the data points constituting that cluster have a larger distribution range. Under the same cluster conditions, the... The wider the distribution range of a cluster of organisms, the worse their overall aggregation; that is, the less they resemble clusters characterized by a specific air quality state. Ultimately, when the first... The state discreteness corresponding to each cluster class The larger the value of , the worse the degree to which the multiple data points constituting the cluster exhibit a similar air quality state, and the greater the dispersion between the data points within it. Therefore, when the cluster participates in the LOF algorithm calculation, it is considered that the distance between the cluster and other clusters is often greater than the actual distance.
[0048] Step S3: Project the data points in any two clusters onto the straight line passing through the class center points of these two clusters to determine the projection points of the data points in any two clusters. Based on the positional distribution of the projection points, the positional relationship between the class center points of any two clusters and their projection points, and the distribution of the surrounding neighboring data points of each data point in each cluster, determine the directional feature value between any two clusters.
[0049] In the LOF algorithm, although each cluster can be treated as a whole for calculation, due to the surrounding discrete points, when clustering to consider the silhouette coefficient, the surrounding outliers usually cause a dense area to be divided into multiple clusters. However, for these clusters, quantifying the distance based solely on the center represented by the mean of the data points within the cluster will result in distance values that are either too large or too small, thus failing to accurately represent the true distance between clusters.
[0050] Therefore, for any two clusters, draw a straight line through the class center points of both clusters, and draw a perpendicular line from each data point within the cluster to this straight line. The foot of the perpendicular on this line is the projection point, thus obtaining the projection point of each data point in the two clusters. Then, the projection point between the projection points of the class centers of the two clusters is determined as the target projection point for the two clusters. Determine the center position of the projection points of the class centers of the two clusters, and determine the distance from each target projection point to this center position, thus obtaining the first distance for each target projection point of the two clusters. Determine the distance from the class center point of the cluster to which the data point corresponding to each target projection point belongs to the target projection point of the two clusters to this center position, thus obtaining the second distance for each target projection point of the cluster. Using the data point corresponding to each target projection point of the two clusters as the center point, define a circular local neighborhood. The ratio of the number of data points in this local neighborhood to the area of the local neighborhood is used as the local data point density, thus obtaining the local data point density for each target projection point of the two clusters. The radius of this local neighborhood can be reasonably set according to actual needs. In this embodiment, the radius of the local neighborhood is set to 1 / 20 of the maximum value of the distance between any two data points in the cluster of the data point corresponding to the target projection point. For example, when the cluster of the data point corresponding to the target projection point is the first... When there are multiple clusters, the radius of the local neighborhood is as described above. 1 / 20. Based on this, according to the first distance, second distance, and local data point density of each target projection point corresponding to any two clusters, and the proportion of the number of target projection points corresponding to any two clusters in all projection points, the directional characteristic value between any two clusters is determined, and the corresponding calculation formula is:
[0051] ;in, Indicates the first The first cluster and the first The directional characteristic values among the clusters; Indicates the first The first cluster and the first The first cluster class corresponding to the th The first distance between each target projection point; Indicates the first The first cluster and the first The first cluster class corresponding to the th The second distance between each target projection point; Indicates the first The first cluster and the first The first cluster class corresponding to the th Local data point density of each target projection point; Indicates the first The first cluster and the first The total number of target projection points corresponding to each cluster class; Indicates the first The first cluster and the first The percentage of target projection points corresponding to each cluster class among all projection points; Indicates the first The first cluster and the first The number of all projection points corresponding to each cluster class; e represents the natural constant.
[0052] In the above-mentioned section The first cluster and the first In the formula for calculating the directional eigenvalues between clusters, the ratio of the first distance and the second distance of each target projection point is calculated. The smaller this ratio, the closer the target projection point is to the center of the projection point of the corresponding cluster center, indicating a higher directional eigenvalue in the two clusters and a more similar air quality state. Simultaneously, the local data point density of each target projection point is calculated and used as the weight of the aforementioned ratio. A higher local data point density results in a higher weight for the distance between the target projection point and the center, and vice versa. Considering that low local data point density may correspond to outlier data points, an exponential function is used to map this density. Since the exponential function changes rapidly when the independent variable is small, it can both ensure that lower local data point densities correspond to lower weights and significantly reduce the weight of lower local data point densities. By calculating the proportion of the target projection points corresponding to two clusters among all projection points, the higher the proportion, the stronger the tendency between the two clusters, and the more likely the same air quality state is to be divided into multiple parts due to outliers.
[0053] Step S4: Determine the Euclidean distance between any two clusters. Based on the directional characteristic values between any two clusters and the state dispersion corresponding to any two clusters, improve the Euclidean distance to determine the improved distance value between any two clusters. Based on the improved distance value, perform anomaly detection on the air quality monitoring data to determine the air quality status.
[0054] Based on the state dispersion of each cluster determined in the above steps, and the directional feature values of any two clusters, the Euclidean distance between the two clusters can be adjusted to obtain a more accurate distance feature representation between them. When the state dispersion of a cluster is higher, we believe the distance between that cluster and the others should be greater than the Euclidean distance obtained from the cluster center. Furthermore, when the directional feature value between two clusters is high, we believe that even if the previously obtained state dispersion is high, there is still a certain directional feature between the two clusters. The clusters are ultimately divided into multiple clusters only due to the limited number of clusters and some outliers. Therefore, the Euclidean distance between these two clusters needs to be optimized.
[0055] To optimize the Euclidean distance between any two clusters, the state discreteness corresponding to each cluster is positively correlated and normalized, such that the value range of the positively correlated normalized state discreteness is [value range missing]. At this point, the greater the state dispersion, the closer the positive correlation normalization result of that state dispersion will be to 1. Simultaneously, negative correlation normalization is performed on the directional eigenvalues between any two clusters, ensuring that the range of the negative correlation normalization result for the directional eigenvalues is [range missing]. At this point, the larger the trend eigenvalue, the closer the negative correlation normalization result of the trend eigenvalue will be to -1.
[0056] Based on the negative correlation normalization result of the directional feature values between any two clusters and the positive correlation normalization result of the state dispersion corresponding to any two clusters, the Euclidean distance between the class centers of any two clusters is improved, thus obtaining a more accurate improved distance value between any two clusters. The corresponding calculation formula is as follows:
[0057] ;in, Indicates the first The first cluster and the first Improved distance values between cluster classes; Indicates the first The first cluster and the first The Euclidean distance between the class centers of each cluster class; Indicates the first The positive correlation normalization result of the state discreteness corresponding to each cluster class; Indicates the first The positive correlation normalization result of the state discreteness corresponding to each cluster class; Indicates the first The first cluster and the first The negative correlation normalization result of the directional feature values among the clusters.
[0058] In the above-mentioned section The first cluster and the first In the formula for calculating the improved distance value between clusters, the first cluster is used... The first cluster and the first The average of the positive correlation normalized results of the state dispersion corresponding to each cluster class is added to the negative correlation normalized results of the directional eigenvalues corresponding to these two cluster classes to obtain the hyperparameters. This hyperparameter The range of values is And utilize this hyperparameter The Euclidean distance between the class centroids of the two clusters is optimized, and the final improved distance value between the two clusters ranges from [value missing]. .
[0059] Based on the improved distance value obtained after optimization between any two clusters determined through the above method, the LOF algorithm is used to detect anomalies in the air quality monitoring data. Specifically, based on the improved distance value obtained after optimization between any two clusters, the LOF algorithm is used to detect outlier clusters. When no outlier clusters are detected, the air quality is determined to be normal, indicating that the purification system is working normally. However, when outlier clusters are detected, the air quality monitoring data corresponding to the data points contained in that outlier cluster is considered abnormal monitoring data, and the air quality is determined to be abnormal. This indicates that the purification system is highly likely to have malfunctioned, and a maintenance warning is issued to the purification system.
[0060] Example of a real-time monitoring and early warning system for purification systems:
[0061] To address the issue of low accuracy in existing air quality monitoring systems, this embodiment also provides a real-time monitoring and early warning system for a purification system. This system includes a processor and a memory. The processor processes computer instructions stored in the memory to implement the steps of a real-time monitoring and early warning method for a purification system. Since this system is essentially a software system, its core function is to implement a real-time monitoring and early warning method for a purification system. Because this method has already been described in detail in the above embodiment of a real-time monitoring and early warning method for a purification system, it will not be repeated here.
[0062] This invention constructs data points corresponding to air quality monitoring data at various set times, clusters these data points to obtain various clusters, and then takes each cluster as an object. Based on the discrete distribution characteristics of data points within each cluster, it quantifies the distribution trend characteristics of data points between clusters and optimizes the distance between different clusters accordingly. This improves the detection accuracy and robustness of outlier data, thereby enhancing the accuracy and reliability of air quality monitoring.
[0063] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A purification system real-time monitoring and repair warning method, characterized in that: The following steps are involved: Acquire air quality monitoring data at each set time, wherein the air quality monitoring data includes at least two types of air quality parameters; Constructing data points corresponding to the air quality monitoring data, clustering all data points to obtain at least two clusters and their cluster centers, and determining the state discreteness corresponding to each cluster according to the number of data points in each cluster, location distribution characteristics, and adjacent conditions at a set time; Project the data points in any two clusters onto a straight line passing through the class center points of the two clusters to determine the projection points of the data points in any two clusters, and determine the trend characteristic value between any two clusters according to the position distribution of the projection points, the position relationship between the class center points of any two clusters and their projection points, and the distribution of the surrounding neighboring data points of each data point in each cluster; Determine the Euclidean distance between any two clusters, improve the Euclidean distance according to the trend characteristic value between any two clusters and the state discreteness corresponding to any two clusters, determine the improved distance value between any two clusters, and perform anomaly detection on the air quality monitoring data based on the improved distance value to determine the air quality state; Determine the trend characteristic values between any two clusters, including: The projection points corresponding to any two clusters and located between the projection points of the class center points of the two clusters are determined as the target projection points corresponding to any two clusters; Determine the center position of the projection point of the class center points of any two clusters, and determine the distance from each target projection point corresponding to any two clusters to the center position, so as to obtain a first distance of each target projection point corresponding to any two clusters; Determine the distance from the class center point of the cluster to which the data point corresponding to each target projection point corresponding to any two clusters belongs to the center position, thereby obtaining a second distance of each target projection point corresponding to any two clusters; According to the distribution of neighboring data points around each pixel in each cluster, the local data point density of each target projection point corresponding to any two clusters is determined; Determine the trend characteristic value between any two clusters according to the first distance, the second distance and the local data point density of each target projection point corresponding to any two clusters, and the proportion of the number of target projection points corresponding to any two clusters in all projection points; Determine the trend characteristic value between any two clusters, and the corresponding calculation formula is: ;in, Indicates Clusters and The trend characteristic values between clusters; Indicates Clusters and The cluster class corresponds to The first distance of the target projection point; Indicates Clusters and The cluster class corresponds to The second distance of the target projection point; Indicates Clusters and The cluster class corresponds to The local data point density of each target projection point; Indicates Clusters and The total number of target projection points corresponding to each cluster; Indicates Clusters and The proportion of target projection points corresponding to each cluster in all projection points; Indicates Clusters and The number of all projection points corresponding to the clusters; e represents a natural constant.
2. A purification system real-time monitoring and repair warning method according to claim 1, characterized in that: Determine the state discreteness corresponding to each cluster, including: Determine the distance between each data point in each cluster and the class center point of the cluster and the distance between any two data points in each cluster; According to the set time corresponding to each data point in each cluster, determine the number of time-series adjacent data points corresponding to each data point in each cluster; The state discreteness corresponding to each cluster is determined based on the number of data points in each cluster, the discreteness of the distance from each data point to the class center point of the cluster, the maximum value of the distance between any two data points, and the number of time-series adjacent data points corresponding to each data point.
3. A purification system real-time monitoring and repair warning method according to claim 2, characterized in that: Determine the state discreteness corresponding to each cluster class, and the corresponding calculation formula is: ;in, Indicates The state dispersion corresponding to each cluster; Indicates The first Data points to The distance between the center points of the clusters; Indicates Each data point in the cluster to the The mean distance between the cluster centers; Indicates The first The number of temporally adjacent data points corresponding to a data point; Indicates The maximum value of the distance between any two data points in the clusters; Indicates The set consisting of the distances between any two data points in a cluster; Indicates The number of data points in each cluster; represents a normalized function; e represents a natural constant.
4. A purification system real-time monitoring and repair warning method according to claim 1, characterized in that: Determine the improved distance value between any two clusters. The corresponding calculation formula is: ;in, Indicates Clusters and Improved distance values between clusters; Indicates Clusters and The Euclidean distance between the cluster centers; Indicates The positive correlation normalization result of the state discreteness corresponding to each cluster; Indicates The positive correlation normalization result of the state discreteness corresponding to each cluster; Indicates Clusters and Normalized result of negative correlation of eigenvalues between clusters.
5. A purification system real-time monitoring and repair warning method according to claim 1, characterized in that: Based on the improved distance value between any two clusters, the LOF algorithm is used to detect anomalies in air quality monitoring data.
6. A purification system real-time monitoring and repair warning method according to claim 1, characterized in that: Constructing data points corresponding to the air quality monitoring data, including: Each type of air quality parameter contained in the air quality monitoring data is determined as a coordinate value on a coordinate axis corresponding to a sample space, and a sample point determined in the sample space by each coordinate value corresponding to the air quality monitoring data is determined as a data point corresponding to the air quality monitoring data.
7. A purification system real-time monitoring and repair warning method according to claim 1, characterized in that: Determined based on the silhouette coefficient The optimal number of clusters in the algorithm, and using The algorithm clusters all data points to obtain the optimal number of clusters.
8. A purification system real-time monitoring and repair warning system, characterized in that: It comprises a processor and a memory, wherein the processor is used to process computer instructions stored in the memory to implement the steps of a purification system real-time monitoring and repair warning method as described in any one of claims 1 to 7.