Artificial Intelligence-based Intelligent Monitoring Platform for Power Operation and Maintenance Substations

Abstract

The present invention relates to the technical field of power operation and maintenance data processing, and specifically relates to an intelligent monitoring platform for a power operation and maintenance substation based on artificial intelligence. The platform includes: a data acquisition unit for collecting current monitoring data of the substation; a sample space construction unit for obtaining data curves of each dimension and calculating the fluctuation state characteristic values of each dimension, and establishing a sample space based on the fluctuation state characteristic values; a data point suspected anomaly degree calculation unit for calculating the spatial distribution anomaly degree and multi-dimensional similarity correlation of the data points, and the sum of the spatial distribution anomaly degree and multi-dimensional similarity correlation of the data points is the suspected anomaly degree; an abnormal data point detection unit for using the suspected anomaly degree to correct the initial neighbor number parameter of the data points; and detecting abnormal data points in the current monitoring data in the sample space by using the corrected neighbor number parameter of the data points. The present invention can accurately monitor the substation and obtain abnormal data.

Description

Technical Field

[0001] The present invention relates to the technical field of power operation and maintenance data processing, and particularly to an intelligent monitoring platform for power operation and maintenance substations based on artificial intelligence. Background Art

[0002] The intelligent operation and maintenance service for power distribution takes the intelligent operation and maintenance platform as the core. By installing online monitoring devices on distribution rooms, high- and low-voltage switch cabinets, the operation status is transmitted to the intelligent operation and maintenance platform in real time for remote operation and maintenance duty, providing a power operation and maintenance solution in the big data era for customers, better ensuring the safety and reliability of power supply, and at the same time achieving the purposes of loss reduction, energy conservation, reduction of labor costs, and improvement of economic benefits.

[0003] In the prior art for the intelligent monitoring of distribution rooms, usually various types of data in the monitoring room can be collected, and the LOF outlier detection algorithm is used to obtain abnormal data during the monitoring process. However, when using the LOF outlier detection algorithm to detect abnormal data in the monitoring data of substations, since it is necessary to combine multi-dimensional data, it is not easy to determine the value of the neighbor number parameter. If the value of the neighbor number parameter k is too small, it may cause the algorithm to be too sensitive to local outliers during the detection process. If the value of the neighbor number parameter k is too large, it may cause the algorithm to be unable to effectively capture local anomalies during the detection process. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide an intelligent monitoring platform for power operation and maintenance substations based on artificial intelligence, and the specific technical solutions adopted are as follows:

[0005] An embodiment of the present invention provides an intelligent monitoring platform for power operation and maintenance substations based on artificial intelligence. The platform includes a data acquisition unit, a sample space construction unit, a data point suspected anomaly degree calculation unit, and an abnormal data point detection unit. Specifically:

[0006] The data acquisition unit is used to collect the current monitoring data of the substation by using sensors, and the monitoring data is multi-dimensional data;

[0007] The sample space construction unit is used to obtain the data curves of each dimension according to the data of each dimension in the historical monitoring data; calculate the fluctuation state characteristic values of each dimension by using the longitudinal maximum distance from each peak to its corresponding valley point in the data curve; establish a sample space based on the fluctuation state characteristic values;

[0008] The data point suspected anomaly degree calculation unit is used to calculate the spatial distribution anomaly degree of the data point according to the clustering result of the current monitoring data in the sample space; calculate the multi-dimensional similarity correlation of the data point by using the values of the data point in each dimension in the sample space; the sum of the spatial distribution anomaly degree and the multi-dimensional similarity correlation of the data point is the suspected anomaly degree;

[0009] An abnormal data point detection unit is used to correct the initial neighbor number parameter of a data point by using the suspected abnormality degree of the data point; and detect abnormal data points in the current monitored data in the sample space by using the corrected neighbor number parameter of the data point.

[0010] Preferably, the fluctuation state eigenvalue of each dimension is calculated by using the longitudinal maximum distance from each peak to its corresponding valley point in the data curve, including:

[0011] Obtain the variance of the longitudinal maximum distance from each peak to its corresponding valley point in the data curve of one dimension, denoted as the longitudinal variance; obtain the sum of the distances from the abscissa of each peak to the abscissas of the corresponding valley points on the left and right sides, denoted as the horizontal span of the peak; obtain the variance of the horizontal spans of each peak, denoted as the horizontal variance; add the longitudinal variance and the horizontal variance, and compare and sum the longitudinal maximum distance from each peak to its corresponding valley point with the added result to obtain the fluctuation state eigenvalue of this dimension.

[0012] Preferably, establishing a sample space based on the fluctuation state eigenvalue includes:

[0013] Obtain a preset number of dimensions with the largest fluctuation state eigenvalues as the main dimensions to establish a sample space.

[0014] Preferably, calculating the spatial distribution abnormality degree of a data point according to the clustering result of the current monitored data in the sample space includes:

[0015] Obtain the volume of the minimum enclosing sphere of each point cluster in the clustering result; obtain the volume of the minimum enclosing sphere of the point cluster where the data point is located in the current monitored data after clustering; obtain the connection line between the centroid of the point cluster where the data point is located and the centroids of other point clusters, obtain the number of data points within two unit lengths of the perpendicular distance connection line, and exclude the data points belonging to the two connected point clusters, and the remaining number of data points is the number of free data points corresponding to the connection line; obtain the spatial distribution abnormality degree of the data point according to the volume of the minimum enclosing sphere of the point cluster where the data point is located and the number of free data points corresponding to the connection line between the centroid of the point cluster where the data point is located and the centroids of other point clusters.

[0016] Preferably, the specific calculation formula of the spatial distribution abnormality degree is:

[0017] ,

[0018] Wherein, represents the spatial distribution abnormality degree of the p-th data point; represents the volume of the minimum circumscribed sphere of the point cluster where the p-th data point is located; represents the number of all point clusters; It represents the number of outlier data points corresponding to the line connecting the centroid of the cluster where the p-th data point is located and the median line of the i-th cluster outside the cluster where the p-th data point is located.

[0019] Preferably, the calculation formula for the multi-dimensional similarity correlation of data points is specifically:

[0020] ,

[0021] where, represents the multi-dimensional similarity correlation of the p-th data point; represents the value of the p-th data point in the t-th dimension; represents the mean value of the values of data points in the t-th dimension; and represent the maximum and minimum values of the values of data points in the t-th dimension respectively; σ[ ] represents the variance calculation function; t represents the t-th dimension.

[0022] Preferably, the initial neighbor number parameter of a data point is corrected using the suspected anomaly degree of the data point, including:

[0023] The ratio of the initial neighbor number parameter of a data point to the suspected anomaly degree of this data point is the corrected neighbor number parameter of this data point.

[0024] Preferably, the corrected neighbor number parameter of a data point is used to detect outlier data points in the current monitored data in the sample space, including:

[0025] In the sample space, based on the corrected neighbor number parameter of each data point, the LOF outlier detection algorithm is used to calculate the data points in the sample space to obtain the LOF value of each data point; calculate the absolute value of the difference between the LOF value of each data point and the mean value of the LOF values of all data points and normalize it to obtain the monitored anomaly degree of each data point; set an anomaly threshold, and when the monitored anomaly degree of a data point is greater than or equal to the anomaly threshold, this data point is an outlier data point.

[0026] The embodiments of the present invention have at least the following beneficial effects: The present invention analyzes historical monitoring data to obtain the fluctuation state characteristic values of each dimension and obtains the main dimensions, and then constructs a new sample space, avoiding the problem that the LOF algorithm performs poorly on high-dimensional data sets, making subsequent anomaly detection more accurate; further, calculating the spatial distribution anomaly degree and multi-dimensional similarity correlation of data points, and comprehensively obtaining the suspected anomaly degree of each data point, and using this to correct the initial neighbor number parameter of each data point. Compared with the problem that directly using the fixed empirical neighbor number parameter k for LOF outlier detection in the prior art will lead to inaccurate detection results of outlier abnormal data, in the present invention, the corrected neighbor number parameter is used for LOF outlier detection. For data with higher anomaly risk, correcting its neighbor number parameter can make it more easily detected as an outlier, and vice versa for data with lower anomaly risk, correcting its neighbor number parameter can make it not easily detected as outlier abnormal data, thereby obtaining a more accurate anomaly detection result for the power distribution room monitoring data and improving the accuracy of anomaly detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0028] Figure 1 It is a platform block diagram of a power operation and maintenance distribution station intelligent monitoring platform based on artificial intelligence provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, in conjunction with the drawings and preferred embodiments, detail the specific implementation manners, structures, features, and effects of a power operation and maintenance distribution station intelligent monitoring platform based on artificial intelligence proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0031] The following will specifically describe the specific solution of a power operation and maintenance distribution station intelligent monitoring platform based on artificial intelligence provided by the present invention with reference to the drawings.

[0032] Embodiment:

[0033] The main application scenarios of the present invention are:

[0034] When using the LOF outlier detection algorithm to perform anomaly detection on the monitoring data of various equipment sensors in the power distribution room, since it is necessary to combine multi-dimensional data, it is not easy to determine the value of the neighbor number parameter for each data point. Using fixed empirical parameters easily leads to inaccurate anomaly detection results and inaccurate power operation and maintenance feedback information in the power distribution room. To address the above problems, this solution obtains the suspected anomaly degree of data points through the distribution characteristics in the data space composed of various monitoring data, weights and corrects the neighbor number parameter k value of the data points to obtain a more appropriate value of the neighbor number parameter, and then uses the LOF outlier detection algorithm to perform anomaly detection.

[0035] Please refer to Figure 1 , which shows the platform block diagram of the intelligent monitoring platform for power operation and maintenance distribution stations based on artificial intelligence provided by the embodiments of the present invention. The platform includes a data acquisition unit, a sample space construction unit, a data point suspected anomaly degree calculation unit, and an abnormal data point detection unit.

[0036] The data acquisition unit is used to collect the current monitoring data of the distribution station by using sensors, and the monitoring data is multi-dimensional data.

[0037] During the operation of the distribution station, it is necessary to collect various data for the operation monitoring of the distribution station. In the present invention, six different types of data are collected, namely voltage, current, temperature, humidity, noise, and dust content. When performing specific collection, the corresponding sensor devices are used to collect the multi-dimensional relevant monitoring data of the distribution station. The collection frequencies of various sensors are set uniformly at once per second, and the collection period is 24 hours, that is, detection is performed in a cycle of 24 hours. It should be noted that the data obtained by each sensor is stored separately in the corresponding sequence, and the newly collected data is added to the end of the corresponding sequence. Thus, the monitoring data of the distribution station can be obtained, where the monitoring data is multi-dimensional data, and then the current monitoring data can be obtained.

[0038] The sample space construction unit is used to obtain the data curves of each dimension according to the data of each dimension in the historical monitoring data; calculate the fluctuation state characteristic values of each dimension by using the longitudinal maximum distance from each peak to its corresponding valley point in the data curve; and establish a sample space based on the fluctuation state characteristic values.

[0039] The monitoring data of multiple dimensions collected at each moment can form a multi-dimensional data point. The data points of multiple moments are placed in a multi-dimensional space to form a discrete multi-dimensional point set. The LOF outlier detection algorithm can be used in the multi-dimensional point set to obtain outlier points. However, since LOF performs poorly on high-dimensional data sets, in a high-dimensional space, the distances between sample points tend to be the same, and it is difficult to distinguish abnormal points by local density, making outlier detection in high-dimensional data unreliable. Therefore, it is necessary to reduce the dimension to construct a sample space for LOF outlier detection, which serves as the basis for subsequent LOF outlier detection and improves the accuracy of subsequent detection.

[0040] The present invention obtains the main dimensions by evaluating the fluctuation state of dimension data. The more frequent the data fluctuation in a dimension and the greater the fluctuation degree, the more sensitive it is to abnormalities and the more timely it can reflect abnormalities. On the other hand, the higher the similarity of normal fluctuations and the higher the repetition frequency, the more prominent the abnormal fluctuations can be highlighted, and the stronger the capture effect on abnormal data fluctuations.

[0041] Further, historical monitoring data is obtained. Specifically, relative to the current cycle, if there are no abnormal data points in the previous cycle, the monitoring data in the previous acquisition cycle is used as the historical monitoring data. If the monitoring data in the previous cycle shows abnormalities, the monitoring data in the cycle before the previous cycle is used as the historical monitoring data, and so on, to obtain the historical monitoring data. The fluctuation state evaluation is performed on the data of each dimension in the historical monitoring data to obtain the fluctuation state characteristic value of the dimension.

[0042] Specifically, the data of each dimension in the historical monitoring data is projected onto a two-dimensional plane coordinate system, with the horizontal axis being the time series and the vertical axis being the dimension value, to obtain the data curve of each dimension. The AMPD peak search algorithm is used on the data curve to obtain the peaks on the data curve of each dimension, where each peak has two corresponding valley points on the left and right sides.

[0043] Obtain the variance of the longitudinal maximum distance from each peak in the data curve to its corresponding valley point, denoted as the longitudinal variance; obtain the sum of the distances from the abscissa of each peak to the abscissas of the corresponding valley points on the left and right sides, denoted as the horizontal span of the peak; obtain the variance of the horizontal spans of each peak, denoted as the horizontal variance; add the longitudinal variance and the horizontal variance, and compare and sum the longitudinal maximum distance from each peak to its corresponding valley point with the added result to obtain the fluctuation state characteristic value of the data curve, that is, the fluctuation state characteristic value of the dimension corresponding to the data curve. The specific calculation formula is:

[0044] ,

[0045] where, represents the fluctuation state characteristic value of the q-th dimension in the historical monitoring data; represents the number of peaks in the data curve corresponding to the q-th dimension; represents the maximum vertical distance between the i-th peak and its corresponding trough point. That is, a peak has two corresponding trough points. The vertical distances between the peak and its two corresponding trough points are obtained respectively in the vertical direction, and the larger of the two distances is the maximum vertical distance; represents the variance of the maximum vertical distances from each peak in the data curve to its corresponding trough point, that is, the vertical variance; represents the variance of the sum of the distances from the abscissa of each peak to the abscissas of the corresponding trough points on the left and right sides, that is, the horizontal variance;

[0046] In the above formula, represents the cumulative sum of the number of peaks weighted by the maximum vertical distance of each peak. The larger this weighted cumulative sum is, the more frequent the data fluctuations in the q-th dimension, the greater the degree of fluctuation, the larger the fluctuation state eigenvalue of the q-th dimension, and the more suitable it is as the main dimension; and the smaller it is, the higher the similarity degree between fluctuations, the higher the repetition frequency. When abnormal data fluctuations occur, the detection effect of anomalies is more obvious. Therefore the larger it is, the larger the fluctuation state eigenvalue of the q-th dimension, and the more suitable it is as the main dimension.

[0047] Thus, the fluctuation state eigenvalues of each dimension are obtained, and the preset number of dimensions with the largest fluctuation state eigenvalues is selected as the main dimensions to construct the sample space for LOF outlier detection. Preferably, in the embodiment of the present invention, the value of the preset number is 3, and the three coordinate axes of the space represent the numerical values of three dimensions. The data after dimensionality reduction improves the accuracy of LOF outlier detection and is the basis for subsequent analysis of data distribution characteristics.

[0048] The data point suspected anomaly degree calculation unit is used to calculate the spatial distribution anomaly degree of the data point according to the clustering result of the current monitoring data in the sample space; calculate the multi-dimensional similarity correlation of the data point using the values of the data point on each dimension in the sample space; the sum of the spatial distribution anomaly degree and the multi-dimensional similarity correlation of the data point is the suspected anomaly degree.

[0049] In the sample space, analyze the neighborhood space distribution characteristics of data points. In the power distribution room, common anomalies are caused by unstable power grids or local load fluctuations. The corresponding changes in voltage and current are sudden numerical jumps, and at the same time, the changes in other sensor monitoring data such as temperature and noise caused by the changes in load equipment will also lead to sudden temperature changes, sudden sound changes, etc. Therefore, outside the point set where normal data aggregates in the sample space, a small point set is likely to be formed. In LOF anomaly detection, this cluster will be identified as a local cluster, resulting in inaccurate anomaly detection results. The volume of this type of cluster is relatively small, and because it is an aggregation of mutation data, the number of discrete data points between it and the normal data point set is relatively small. Therefore, when a data point is located within this type of point cluster, the possibility that this data point is a suspected anomaly point is relatively high.

[0050] In the sample space, use kmeans clustering for the current monitoring data. The value of k is obtained by the elbow method to obtain the clustering result. The data in the sample space is divided into multiple point clusters. Use the random incremental algorithm to obtain the minimum enclosing sphere of each point cluster. This enclosing sphere represents the volume range of the point cluster, and obtain the volume of the minimum enclosing sphere of the point cluster where the data point is located in the current monitoring data after clustering; obtain the connection line between the centroid of the point cluster where the data point is located and the centroids of other point clusters, obtain the number of data points within two unit lengths of the vertical distance connection line, and exclude the data points belonging to the two connected point clusters. The remaining number of data points is the number of free data points corresponding to the connection line;

[0051] Obtain the spatial distribution anomaly degree of the data point according to the volume of the minimum enclosing sphere of the point cluster where the data point is located and the number of free data points corresponding to the connection line between the centroid of the point cluster where the data point is located and the centroids of other point clusters. The specific calculation formula for the spatial distribution anomaly degree is:

[0052] ,

[0053] Among them, represents the spatial distribution anomaly degree of the p-th data point; represents the volume of the minimum circumscribed sphere of the point cluster where the p-th data point is located; represents the number of other point clusters except the point cluster where the p-th data point is located, represents the number of all point clusters; represents the number of free data points corresponding to the connection line between the centroid of the point cluster where the p-th data point is located and the median line of the i-th point cluster outside the point cluster where the p-th data point is located;

[0054] The volume of the minimum enclosing sphere of the point cluster where the p-th data point is located. The smaller this volume is, the more the p-th cluster conforms to the characteristics of a small point set and may be a point cluster composed of mutant data, and the more abnormal the cluster where the p-th data point is located. The smaller the number of free data points, the fewer discrete data points there are between the point clusters. The cumulative sum of the number of data points around the connection line between the point cluster where the p-th data point is located and other point clusters. The smaller this cumulative sum is, the more isolated the point cluster where the p-th data point is located, the fewer discrete data there are from the normal point clusters, and the more it conforms to the characteristics of the point cluster formed by the aggregation of mutant data point characteristics. The greater the degree of spatial distribution abnormality corresponding to the p-th data point.

[0055] Thus, the degree of spatial distribution abnormality of all data points is obtained. However, relying solely on spatial distribution characteristics has limitations. Here, the multi-dimensional data correlation characteristics are further combined to improve the accuracy of obtaining suspected abnormal data points.

[0056] The environment inside the substation is relatively enclosed. Therefore, when an abnormality occurs inside, various monitoring data will have interrelated changes. The abnormal changes of various monitoring data are caused by the same abnormality and will be interfered with to the same extent. Therefore, the data points in the three-dimensional space are disassembled into one-dimensional data. When the data points have similar degrees of outliers in each dimension, it indicates that the data point is more likely to be an outlier data point caused by an abnormality inside the substation. If the degree of outliers varies greatly in each dimension, it indicates that the data point may be a normal change in the current dimension data, such as the change of current and voltage under the power grid call. Here, by obtaining the degree of outliers of the data points in each dimension, the similarity correlation between the data points in multiple dimensions is obtained.

[0057] Thus, the values of the data points in each dimension of the sample space are obtained, and then the multi-dimensional similarity correlation of the data points is calculated. The specific calculation formula for the multi-dimensional similarity correlation of the data points is:

[0058] ,

[0059] Among them, represents the multi-dimensional similarity correlation of the p-th data point; represents the value of the p-th data point in the t-th dimension; represents the mean value of the values of the data points in the t-th dimension; and represent the maximum and minimum values of the values of the data points in the t-th dimension respectively; σ[ ] represents the variance function;

[0060] is the difference between the value of the p-th data point in the t-th dimension and the mean value of the values of the data points in the t-th dimension, representing the degree of outliers of the p-th data point in the t-th dimension. The greater this difference is, the more outlier the p-th data point is in the t dimension. It represents the normalization of the degree of outlier to unify the magnitude of the degree of outlier in different dimensions. It represents the variance of the degree of outlier of the p-th data point in three dimensions. The smaller the variance, the stronger the similarity correlation between the corresponding multi-dimensions. Therefore, an inverse form is adopted.

[0061] Here, the suspected outlier degree of the data point is obtained by combining the spatial distribution outlier degree and the multi-dimensional similarity correlation:

[0062] ,

[0063] is the suspected outlier degree of the p-th data point in the sample space; represents the multi-dimensional similarity correlation of the p-th data point in the sample space; represents the spatial distribution outlier degree of the p-th data point in the sample space. The stronger the spatial distribution outlier degree of the p-th data point, the stronger the multi-dimensional similarity correlation, and the stronger the suspected outlier degree corresponding to this point.

[0064] Thus, it can be obtained that the suspected outlier degree of each data point is stronger.

[0065] The outlier data point detection unit is used to correct the initial neighbor number parameter of the data point by using the suspected outlier degree of the data point; and detect the outlier data points in the current monitored data in the sample space by using the corrected neighbor number parameter of the data point.

[0066] After obtaining the suspected outlier degree in the sample space, it is necessary to correct the initial neighbor number parameter of each data point by using the suspected outlier degree to obtain the corrected neighbor number parameter; the ratio of the initial neighbor number parameter of the data point to the suspected outlier degree of the data point is the corrected neighbor number parameter of the data point. The specific calculation formula is:

[0067] ,

[0068] where, represents the corrected neighbor number parameter of the p-th data point in the sample space; represents the initial neighbor number parameter of the p-th data point in the sample space; represents the suspected outlier degree of the p-th data point in the sample space; It indicates that the stronger the suspected abnormality degree of the p-th point, the smaller the corresponding number of neighbors k. Thus, the sensitivity to outlier data points can be improved, and local anomalies can be effectively captured. The weaker the suspected abnormality degree of the p-th point, the more the data tends to be normal, the larger the number of neighbors k, and it is not easy to identify normal outlier points as anomalies. In addition, it should be noted that after optimizing the initial number of neighbor parameter, the obtained corrected number of neighbor parameter may not be an integer. Therefore, it is necessary to verify it. In the present invention, the method of rounding is used for rounding.

[0069] Further, in the sample space, based on the corrected number of neighbor parameter of each data point, the LOF outlier detection algorithm is used to calculate the data points in the sample space, and the LOF value of each data point is obtained. If the LOF value is larger, it indicates that it is more abnormal. On the contrary, if it is smaller, it indicates that it is more normal. Further, calculate the absolute value of the difference between the LOF value of each data point and the mean value of the LOF values of all data points and normalize it to obtain the monitoring abnormality degree of each data point. The specific calculation formula is:

[0070] ,

[0071] where, represents the monitoring abnormality degree of the p-th data point; represents the LOF value of the p-th data point; represents the mean value of the LOF values of all data points in the current monitoring data; norm represents the normalization operation; represents the difference between the LOF of the p-th data point and the mean LOF value. Because the base number of the LOF value of the smaller value is larger, the greater this difference, the stronger the abnormality degree of the p-th data point. Here, norm is used for linear normalization.

[0072] Finally, set the anomaly threshold Y. Preferably, in the embodiment of the present invention, the value of the anomaly threshold is 0.6. The implementer can adjust the anomaly threshold Y according to the actual scenario or obtain it based on statistical analysis. When the monitoring abnormality degree of the data point is greater than or equal to the anomaly threshold, the data point is an abnormal data point. Record these abnormal data points and transmit a warning to the monitoring platform to remind the staff.

[0073] It should be noted that: the above sequence of embodiments of the present invention is only for description and does not represent the advantages or disadvantages of the embodiments. And the above describes specific embodiments of this specification. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0074] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized.

[0075] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An artificial intelligence-based intelligent monitoring platform for power distribution station operation and maintenance, characterized in that: The platform includes: A data acquisition unit, used to collect current monitoring data of the power distribution station using sensors, wherein the monitoring data is multi-dimensional data; A sample space construction unit is used to obtain a data curve of each dimension based on the data of each dimension in the historical monitoring data; calculate the fluctuation state characteristic value of each dimension using the maximum longitudinal distance from each peak to its corresponding trough point in the data curve; and establish a sample space based on the fluctuation state characteristic value; The suspected abnormality degree calculation unit of the data point is used to calculate the abnormality degree of the spatial distribution of the data point according to the clustering result of the current monitoring data in the sample space; the multidimensional similarity correlation of the data point is calculated using the value of the data point in each dimension in the sample space; the sum of the abnormality degree of the spatial distribution of the data point and the multidimensional similarity correlation is the suspected abnormality degree; An abnormal data point detection unit is used to modify the initial neighbor quantity parameter of the data point using the suspected abnormality degree of the data point; and detect abnormal data points in the current monitoring data in the sample space using the modified neighbor quantity parameter of the data point; The method of calculating the fluctuation state characteristic value of each dimension by using the maximum longitudinal distance from each peak to its corresponding trough point in the data curve includes: Obtain the variance of the maximum longitudinal distance from each peak to its corresponding trough point in a data curve of one dimension, recorded as the longitudinal variance; obtain the sum of the distances from the horizontal coordinates of each peak to the horizontal coordinates of the corresponding trough points on the left and right sides, recorded as the horizontal span of the peak; obtain the variance of the horizontal span of each peak, recorded as the horizontal variance; add the longitudinal variance and the horizontal variance, compare the maximum longitudinal distance from each peak to its corresponding trough point with the summed result and sum them up to obtain the characteristic value of the fluctuation state of this dimension; The establishing of a sample space based on the fluctuation state characteristic value comprises: The dimensions with the largest preset number of eigenvalues ​​of the fluctuation state are used to establish the sample space as the main dimensions; The method of modifying the initial neighbor quantity parameter of the data point by using the suspected abnormality degree of the data point includes: The ratio of the initial number of neighbors of a data point to the suspected abnormality degree of the data point is the corrected number of neighbors of the data point.

2. According to claim 1, an artificial intelligence-based power operation and maintenance distribution station intelligent monitoring platform is characterized in that: The calculation of the degree of abnormality of spatial distribution of data points according to the clustering result of the current monitoring data in the sample space includes: Obtain the volume of the minimum bounding sphere of each point cluster in the clustering result; obtain the volume of the minimum bounding sphere of the point cluster where the data point is located in the current monitoring data after clustering; obtain the connecting line between the centroid of the point cluster where the data point is located and the centroid of other point clusters, obtain the number of data points within two unit lengths of the vertical distance connecting line, and remove the data points of the point clusters belonging to the two connecting lines. The number of remaining data points is the number of free data points corresponding to the connecting line; the degree of abnormality of the spatial distribution of the data point is obtained according to the volume of the minimum bounding sphere of the point cluster where the data point is located and the number of free data points corresponding to the connecting line between the centroid of the point cluster where the data point is located and the centroid of other point clusters.

3. According to claim 2, an artificial intelligence-based intelligent monitoring platform for power distribution station operation and maintenance, characterized in that: The calculation formula of the spatial distribution anomaly degree is specifically: ， in, Indicates the degree of abnormality of the spatial distribution of the pth data point; Represents the volume of the minimum circumscribed sphere of the point cluster where the pth data point is located; Represents the number of all point clusters; It represents the number of free data points corresponding to the line connecting the centroid of the point cluster where the p-th data point is located and the midline of the i-th point cluster outside the point cluster where the p-th data point is located.

4. According to the artificial intelligence-based intelligent monitoring platform for power distribution station operation and maintenance according to claim 1, it is characterized in that: The calculation formula of the multi-dimensional similarity correlation of the data points is specifically: ， in, Represents the multidimensional similarity correlation of the p-th data point; Represents the value of the p-th data point in the t-th dimension; Represents the mean value of the data points in the tth dimension; and They represent the maximum and minimum values ​​of the data points in the t-th dimension respectively; σ[ ] represents the variance function; t represents the t-th dimension.

5. According to the artificial intelligence-based intelligent monitoring platform for power distribution station operation and maintenance according to claim 1, it is characterized in that: The method of detecting abnormal data points in the current monitoring data in the sample space by using the neighbor quantity parameter after the data point correction includes: In the sample space, the LOF outlier detection algorithm is used to calculate the data points in the sample space based on the corrected neighbor number parameter of each data point to obtain the LOF value of each data point; the absolute value of the difference between the LOF value of each data point and the mean LOF value of all data points is calculated and normalized to obtain the monitoring abnormality degree of each data point; an abnormal threshold is set, and when the monitoring abnormality degree of a data point is greater than or equal to the abnormal threshold, the data point is an abnormal data point.

Images