An intelligent photovoltaic power quality control system

Abstract

The present invention discloses an intelligent photovoltaic power quality control system, comprising a photovoltaic data acquisition module, a feature space trend graph construction module, an isolation correction module, a preliminary clustering module, a boundary disturbance allocation module, a clustering optimization module, and a photovoltaic power quality control module. The present invention belongs to the field of power quality control, and specifically refers to an intelligent photovoltaic power quality control system. The scheme adaptively captures hierarchical inflection points ranging from small fluctuations to severe interference, more flexibly identifies events of different intensities; eliminates trend offsets caused by abnormal disturbances; introduces density consistency constraints to calculate the merging index; thereby improving the accuracy of photovoltaic power quality control; reduces the probability of blind spots through initialization, and better adapts to the characteristics of photovoltaic data; introduces a dynamic balance mechanism to balance global search and local convergence; and avoids the long-term retention of inferior solutions based on a local search enhancement mechanism, ultimately improving the photovoltaic power quality control effect.

Description

Technical Field

[0001] The present invention relates to the field of power quality control, and in particular to an intelligent photovoltaic power quality control system. Background Art

[0002] A photovoltaic power quality control system is a comprehensive system used to monitor, analyze, and improve the power quality of photovoltaic power generation systems. However, typical photovoltaic power quality control systems are prone to missing weak interference or falsely reporting strong interference, are severely affected by anomalies, and are unable to adapt to the density differences of photovoltaic systems under various operating conditions, resulting in low accuracy in photovoltaic power quality control. These systems also struggle to adapt to complex operating environments and balance global exploration with local development, leading to poor photovoltaic power quality control. Summary of the Invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent photovoltaic power quality control system. In view of the fact that the general photovoltaic power quality control system is prone to miss detection of weak interference or falsely report strong interference, is seriously interfered by abnormal points, and cannot adapt to the density differences of the photovoltaic system under multiple working conditions, which leads to low accuracy of photovoltaic power quality control, this solution adaptively captures the hierarchical inflection points from small fluctuations to severe interference, and more flexibly identifies events of different intensities; through LOF Capture transient harmonic clusters, capture mutation points through isolation forests, and eliminate trend deviations caused by abnormal disturbances; introduce density consistency constraints to calculate the merging index to prevent incorrect merging between clusters of different densities and adapt to various working conditions of photovoltaic operation; thereby improving the accuracy of photovoltaic power quality control; in view of the fact that general photovoltaic power quality control systems are unable to adapt to complex operating environments and find it difficult to simultaneously take into account global exploration and local development, resulting in poor photovoltaic power quality control effects, this scheme reduces the blind spot probability through initialization, enhances initial diversity, and is more adaptable to the characteristics of wide fluctuation range and large distribution differences in photovoltaic data; introduces a dynamic balance mechanism to balance global search and local convergence, and adapts to the non-stationary nature of photovoltaic data over time; avoids long-term retention of inferior solutions based on the local search enhancement mechanism; and ultimately improves the photovoltaic power quality control effect.

[0004] The technical solution adopted by the present invention is as follows: the present invention provides an intelligent photovoltaic power quality control system, including a photovoltaic data acquisition module, a feature space trend graph construction module, an isolated correction module, a preliminary clustering module, a boundary disturbance allocation module, a clustering optimization module and a photovoltaic power quality control module;

[0005] The photovoltaic data acquisition module collects power quality data of the photovoltaic system to form a sample set;

[0006] The feature space trend graph construction module constructs a fully connected graph between samples and constructs a core set by calculating elbow point distance and adjacency relationship;

[0007] The isolated correction module detects isolated points and improves the core set through anomaly scores;

[0008] The preliminary clustering module forms preliminary clusters based on the density of core points and their neighborhoods, and introduces a density consistency factor to merge clusters;

[0009] The boundary disturbance allocation module calculates and allocates confidence of boundary points based on local similarity, and finally performs boundary point allocation;

[0010] The clustering optimization module automatically adjusts clustering parameters based on the particle swarm optimization algorithm and introduces a dynamic balance mechanism and a local search enhancement strategy;

[0011] The photovoltaic power quality control module implements photovoltaic power quality control on real-time power quality monitoring data based on the clustering results of historical power quality monitoring data.

[0012] Furthermore, the photovoltaic data acquisition module collects historical power quality monitoring data according to a fixed sampling time sequence during the operation of the photovoltaic system, pre-processes the collected data, and forms a sample set. ,in, and are the 1st and nth samples respectively; each sample is an extracted feature vector; the operating status of each sample is labeled, and the operating status is used as a data label and is only selected and used as a cluster label; the operating status includes normal operation, mild abnormality, moderate abnormality and severe abnormality.

[0013] Furthermore, the feature space trend graph construction module is to construct a fully connected graph between samples, with the edge weight being the Euclidean distance, and to initialize all samples to be unvisited; for each sample u, the elbow distance between the sample and its k nearest neighbors is calculated. , connecting Euclidean distance Less than neighbors; take sample u as the leaf node and sample v as the parent node, jump to the nearest neighbor each time, and take the stopping point as the root node; repeat until all points are visited, and collect all root node sets as the core set R; where, is the Euclidean distance from u to v; k nearest neighbors is the set of k samples with the smallest Euclidean distance; elbow distance It is the distance corresponding to the point with the largest curvature on the curve after plotting the Euclidean distance of sample u to all other k-nearest neighbor samples and sorting them by size; v is the sample among u's k-nearest neighbors.

[0014] Furthermore, the isolated correction module uses the anomaly score to fuse the local outlier factor LOF to detect isolated points and calculate the anomaly score of each sample. , the formula used is: ; ; Calculate the global anomaly score , the formula used is: ;in, is the length of the isolated path; E[·] is the expectation; is the i-th sample the expected path length required to be isolated; is the sample normalization constant; is a harmonic number; is the anomaly weight; is the local outlier factor of the sample; is the maximum value of the local outlier factor among all samples; is the jth sample; N is the size of each tree sample subset when the sample participates in the construction of the isolation forest; according to the threshold, the samples with high global anomaly scores are marked as sparse areas SR, and the rest are dense areas DR; for each trend tree edge (x a ,x b ), if one end x a Belongs to SR, the other end x b If it belongs to DR, disconnect the trend tree edge and treat both ends as new root nodes and add R; x a and x b These are the two ends of the trend tree edge.

[0015] Furthermore, the preliminary clustering module is to cluster the i-th sample As the core point, the core point , let the core point k nearest neighbor set be NN(x i ); For each core point , and belongs to NN(x i ) to form a subgraph ; Each subgraph corresponds to a cluster; introduce density consistency factor , calculate the merger index, the formula used is: ; ; ; ;in, is the merge index; i and j are both sample indices, i≠j; The core point The obtained subgraph; and are the local densities of samples x and y, respectively; It is a subgraph and subgraphs The closest Euclidean distance between them; is the density normalization parameter; is the k-nearest neighbor set of the sample; iterative merging of the maximum Yes, until the number of remaining clusters is K, and the final core clustering is obtained; It is a subgraph samples; It is a subgraph A sample of ; e is a natural constant.

[0016] Furthermore, the boundary perturbation allocation module is used to treat samples that are not classified into any cluster as boundary points. , calculate its allocation confidence for the c-th cluster , the formula used is: ;in, It is a boundary point and The total Euclidean distance between points; the boundary points are assigned to the cluster with the maximum distribution confidence; for the clustering results of the historical power quality monitoring data after boundary disturbance allocation, the silhouette coefficient is evaluated and the silhouette coefficient threshold is set. If the average silhouette coefficient of all samples is lower than the silhouette coefficient threshold, clustering optimization is performed; otherwise, the clustering of the historical power quality monitoring data is completed.

[0017] Furthermore, the cluster optimization module specifically includes the following contents:

[0018] Initialize individual positions; construct parameter optimization space based on target cluster number, density normalization parameter, k and silhouette coefficient threshold; quasi-initialization of the qth dimension of each particle is expressed as: ; Generate individual positions and obtain the final initialized particles. The formula used is: ; The average silhouette coefficient of all samples clustered based on individual positions is used as the individual fitness value; the lower the individual fitness value, the worse the solution corresponding to the individual; where, is the quasi-initial position of the mth particle in the qth dimension; and are the upper and lower limits of the qth dimension respectively; is uniformly distributed Random number; is the final initial position of the mth particle in the qth dimension; rnd(·,·) is a function that takes uniform random values in a given interval;

[0019] Particle position update; introduce the position dynamic balance mechanism and define the time function. The formula used is: ; ; The formula used to update speed and position is: ; ;in, is the dynamic cognitive learning factor; g is the current iteration number; is the maximum number of iterations; is the inertia weight; is the search mode switching threshold; and are the q-th dimension velocities of the m-th particle at the g+1-th iteration and the g-th iteration, respectively; 、 and They are all random numbers between 0 and 1, independent of each other; and are the positions of the qth dimension of the mth particle at the g+1th iteration and the gth iteration respectively; It is a social learning factor; is the position of the optimal particle in the qth dimension; F is a random number that obeys the uniform distribution U(0,1), and Independent of each other; is the historical optimal position of the mth particle in the qth dimension;

[0020] Local search enhancement: every k iterations, the 10% individuals with the lowest fitness values are perturbated and adjusted. The formula used is: ;in, is the position after local search enhancement; The degrees of freedom are t-distributed random variable;

[0021] Optimization judgment; when there is an individual fitness value higher than the silhouette coefficient threshold, the clustering result of the historical power quality monitoring data is obtained based on the individual position; if the maximum number of iterations is reached, the individual position is returned to the initialization; otherwise, it returns to the particle position update.

[0022] Furthermore, the photovoltaic power quality control module selects the label with the largest number in the cluster as the cluster label based on the clustering results of historical power quality monitoring data; collects real-time power quality monitoring data and performs cluster allocation based on the minimum Euclidean distance principle; uses the cluster label corresponding to the real-time power quality monitoring data as the power quality monitoring result; if the monitoring result is moderately abnormal, increases the sampling frequency of the monitoring data; if the monitoring result is seriously abnormal, performs early warning processing, thereby achieving power quality control.

[0023] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0024] (1) In view of the fact that general photovoltaic power quality control systems are prone to missing weak interference or falsely reporting strong interference, are seriously disturbed by abnormal points, and cannot adapt to the density differences of photovoltaic systems under multiple working conditions, which leads to low accuracy of photovoltaic power quality control, this scheme adaptively captures the hierarchical inflection points from small fluctuations to severe interference, and more flexibly identifies events of different intensities; captures transient harmonic clusters through LOF, captures mutation points through isolation forest, and eliminates trend deviations caused by abnormal disturbances; introduces density consistency constraints to calculate the merging index to prevent incorrect merging between clusters of different densities, and adapts to various working conditions of photovoltaic operation; thereby improving the accuracy of photovoltaic power quality control.

[0025] (2) In view of the fact that general photovoltaic power quality control systems are unable to adapt to complex operating environments and find it difficult to simultaneously take into account global exploration and local development, resulting in poor photovoltaic power quality control effects, this scheme reduces the probability of blind spots through initialization, enhances initial diversity, and is more adaptable to the characteristics of photovoltaic data with wide fluctuation range and large distribution differences; introduces a dynamic balance mechanism to balance global search and local convergence, and adapts to the non-stationary nature of photovoltaic data over time; avoids the long-term retention of inferior solutions based on the local search enhancement mechanism; and ultimately improves the photovoltaic power quality control effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of the process of an intelligent photovoltaic power quality control system provided by the present invention;

[0027] Figure 2 Schematic diagram of the clustering optimization module.

[0028] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0031] Example 1, see Figure 1 The present invention provides an intelligent photovoltaic power quality control system, which includes a photovoltaic data acquisition module, a feature space trend graph construction module, an isolated correction module, a preliminary clustering module, a boundary disturbance allocation module, a clustering optimization module and a photovoltaic power quality control module;

[0032] The photovoltaic data acquisition module collects power quality data of the photovoltaic system to form a sample set; and sends the data to the feature space trend graph construction module;

[0033] The feature space trend graph construction module constructs a fully connected graph between samples, constructs a core set by calculating elbow point distance and adjacency relationship; and sends the data to the isolation correction module;

[0034] The isolated correction module detects isolated points and improves the core set by anomaly scores; and sends the data to the preliminary clustering module;

[0035] The preliminary clustering module forms preliminary clusters based on the core points and their neighborhood densities, introduces density consistency factors to merge clusters, and sends the data to the boundary disturbance allocation module;

[0036] The boundary perturbation allocation module calculates the allocation confidence of the boundary points that are not classified into any cluster based on the local similarity, and finally allocates the boundary points; and sends the data to the cluster optimization module;

[0037] The clustering optimization module automatically adjusts clustering parameters based on the particle swarm optimization algorithm, introduces a dynamic balance mechanism and a local search enhancement strategy, and sends data to the photovoltaic power quality control module;

[0038] The photovoltaic power quality control module implements photovoltaic power quality control on real-time power quality monitoring data based on the clustering results of historical power quality monitoring data.

[0039] Example 2, see Figure 1 This embodiment is based on the above embodiment. The photovoltaic data acquisition module collects historical power quality monitoring data according to a fixed sampling time sequence during the operation of the photovoltaic system, pre-processes the collected data, and forms a sample set. ,in, and are the 1st and nth samples respectively; each sample is an extracted feature vector; the operating status of each sample is labeled, and the operating status is used as a data label and is only selected and used as a cluster label; the feature vector includes voltage characteristics, frequency characteristics, harmonic characteristics, power characteristics and current characteristics; the operating status includes normal operation, mild abnormality, moderate abnormality and severe abnormality.

[0040] Example 3, see Figure 1 This embodiment is based on the above embodiment. The feature space trend graph construction module is to construct a fully connected graph between samples. The edge weight is the Euclidean distance. All samples are initialized to an unvisited state. For each sample u, the elbow distance between the sample and its k nearest neighbors is calculated. , connecting Euclidean distance Less than neighbors; take sample u as the leaf node and sample v as the parent node, jump to the nearest neighbor each time, and take the stopping point as the root node; repeat until all points are visited, and collect all root node sets as the core set R; where, is the Euclidean distance from u to v; k nearest neighbors is the set of k samples with the smallest Euclidean distance; elbow distance It is the distance corresponding to the point with the largest curvature on the curve after plotting the Euclidean distances from sample u to all other k-nearest neighbors and sorting them by size; v is the sample among u's k-nearest neighbors; it is used to capture multi-level density peaks, corresponding to the hierarchical inflection points from small fluctuations to strong events.

[0041] Example 4, see Figure 1 This embodiment is based on the above embodiment. The isolated correction module uses the anomaly score to fuse the local outlier factor LOF to detect isolated points and calculate the anomaly score of each sample. , the formula used is: ; ; Calculate the global anomaly score , the formula used is: ;in, is the length of the isolated path; E[·] is the expectation; It is a sample the expected path length required to be isolated; is the sample normalization constant; is a harmonic number; is the anomaly weight; is the local outlier factor of the sample; is the maximum value of the local outlier factor among all samples; is the jth sample; N is the size of each tree sample subset when the sample participates in the construction of the isolation forest; according to the threshold, the samples with high global anomaly scores are marked as sparse areas SR, and the rest are dense areas DR; for each trend tree edge (x a ,x b ), if one end x a Belongs to SR, the other end x b If it belongs to DR, disconnect the trend tree edge and treat both ends as new root nodes and add R; x a and x bThey are the two ends of the trend tree edge; LOF is used to capture instantaneous harmonic clusters, and anomaly scores are used to capture isolated mutations; the deviation caused by abnormal disturbances in the trend structure is cleaned up, and the core set's recognition accuracy of abnormal events is improved.

[0042] Example 5, see Figure 1 This embodiment is based on the above embodiment. The preliminary clustering module is to cluster the i-th sample As the core point, the core point , let the core point k nearest neighbor set be NN(x i ); For each core point , and belongs to NN(x i ) to form a subgraph ; Each subgraph corresponds to a cluster; introduce density consistency factor , resist the merger of density differences that are too large, calculate the merger index, and the formula used is: ; ; ; ;in, is the merge index; i and j are both sample indices, i≠j; The core point The obtained subgraph; and are the local densities of samples x and y, respectively; It is a subgraph and subgraphs The closest Euclidean distance between them; is the density normalization parameter; is the k-nearest neighbor set of the sample; iterative merging of the maximum Yes, until the number of remaining clusters is K, and the final core clustering is obtained; It is a subgraph samples; It is a subgraph samples; the normal fluctuation mode is subdivided into K typical operating states to support quality control under different working conditions; e is a natural constant.

[0043] Example 6, see Figure 1 This embodiment is based on the above embodiment. The boundary perturbation allocation module is to treat samples that are not classified into any cluster as boundary points. , corresponding to the uncertainty points of slight flicker and harmonics, calculate the distribution confidence of the cth cluster , the formula used is: ;in, It is a boundary point and The total Euclidean distance between points; assign boundary points to the cluster with the maximum distribution confidence; ensure that boundary points are smoothly assigned according to local similarity to avoid hard threshold misclassification; for the clustering results of historical power quality monitoring data after boundary disturbance allocation, use silhouette coefficient evaluation and set silhouette coefficient threshold. If the average silhouette coefficient of all samples is lower than the silhouette coefficient threshold, clustering optimization is performed; otherwise, the clustering of historical power quality monitoring data is completed.

[0044] By performing the above operations, this solution adaptively captures the layered inflection points from small fluctuations to severe interference, allowing for more flexible identification of events of varying intensities. This solution addresses the issues with general photovoltaic power quality control systems, which are prone to missing weak interference or falsely reporting strong interference, are severely affected by anomalies, and are unable to adapt to the density differences of the photovoltaic system under multiple operating conditions, leading to low accuracy in photovoltaic power quality control. It also captures transient harmonic clusters through LOF and mutation points through isolation forest, eliminating trend deviations caused by abnormal disturbances. It also introduces density consistency constraints to calculate the merging index, preventing the mismerging of clusters of different densities and adapting to various photovoltaic operating conditions, thereby improving the accuracy of photovoltaic power quality control.

[0045] Example 7, see Figure 1 and Figure 2 This embodiment is based on the above embodiment, and the clustering optimization module specifically includes the following contents:

[0046] Initialize individual positions; construct parameter optimization space based on target cluster number, density normalization parameter, k and silhouette coefficient threshold; quasi-initialization of the qth dimension of each particle is expressed as: ; Generate individual positions and obtain the final initialized particles. The formula used is: ; The average silhouette coefficient of all samples clustered based on individual positions is used as the individual fitness value; the lower the individual fitness value, the worse the solution corresponding to the individual; where, is the quasi-initial position of the mth particle in the qth dimension; and are the upper and lower limits of the qth dimension respectively; is uniformly distributed Random number; is the final initial position of the mth particle in the qth dimension; rnd(·,·) is a function that takes uniform random values in a given interval;

[0047] Particle position update; introduce the position dynamic balance mechanism and define the time function. The formula used is: ; ; The formula used to update speed and position is: ; ;in, is the dynamic cognitive learning factor; g is the current iteration number; is the maximum number of iterations; is the inertia weight; is the search mode switching threshold; and are the q-th dimension velocities of the m-th particle at the g+1-th iteration and the g-th iteration, respectively; 、 and They are all random numbers between 0 and 1, independent of each other; and are the positions of the qth dimension of the mth particle at the g+1th iteration and the gth iteration respectively; It is a social learning factor; is the position of the optimal particle in the qth dimension; F is a random number that obeys the uniform distribution U(0,1), and Independent of each other; is the historical optimal position of the mth particle in the qth dimension;

[0048] Local search enhancement: every k iterations, the 10% individuals with the lowest fitness values are perturbated and adjusted. The formula used is: ;in, is the position after local search enhancement; The degrees of freedom are t-distributed random variable;

[0049] Optimization judgment; when there is an individual fitness value higher than the silhouette coefficient threshold, the clustering result of the historical power quality monitoring data is obtained based on the individual position; if the maximum number of iterations is reached, the individual position is returned to the initialization; otherwise, it returns to the particle position update.

[0050] By performing the above operations, in order to address the problem that general photovoltaic power quality control systems are unable to adapt to complex operating environments and find it difficult to simultaneously take into account global exploration and local development, resulting in poor photovoltaic power quality control effects, this scheme reduces the probability of blind spots through initialization, enhances initial diversity, and is more adaptable to the wide fluctuation range and large distribution differences in photovoltaic data; introduces a dynamic balance mechanism to balance global search and local convergence, and adapts to the non-stationary nature of photovoltaic data over time; and avoids the long-term retention of inferior solutions based on the local search enhancement mechanism. Ultimately, the photovoltaic power quality control effect is improved.

[0051] Example 8, see Figure 1This embodiment is based on the above embodiment. The photovoltaic power quality control module selects the label with the largest number in the cluster as the cluster label based on the clustering result of historical power quality monitoring data; collects real-time power quality monitoring data and performs cluster allocation based on the minimum Euclidean distance principle; uses the cluster label corresponding to the real-time power quality monitoring data as the power quality monitoring result; if the monitoring result is moderately abnormal, increases the sampling frequency of the monitoring data; if the monitoring result is seriously abnormal, performs early warning processing; thereby realizing power quality control; the power quality monitoring data after cluster allocation is regarded as historical power quality monitoring data, and the clustering result of the historical power quality monitoring data is periodically returned and updated.

[0052] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0053] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0054] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. An intelligent photovoltaic power quality control system, characterized by: The system includes a photovoltaic data acquisition module, a feature space trend graph construction module, an isolated correction module, a preliminary clustering module, a boundary disturbance allocation module, a clustering optimization module and a photovoltaic power quality control module; The photovoltaic data acquisition module collects power quality data of the photovoltaic system to form a sample set; The feature space trend graph construction module constructs a fully connected graph between samples and constructs a core set by calculating elbow point distance and adjacency relationship; The isolated correction module detects isolated points and improves the core set through anomaly scores; The preliminary clustering module forms preliminary clusters based on the density of core points and their neighborhoods, and introduces a density consistency factor to merge clusters; The boundary disturbance allocation module calculates and allocates confidence of boundary points based on local similarity, and finally performs boundary point allocation; The clustering optimization module automatically adjusts clustering parameters based on the particle swarm optimization algorithm and introduces a dynamic balance mechanism and a local search enhancement strategy; The photovoltaic power quality control module implements photovoltaic power quality control on real-time power quality monitoring data based on the clustering results of historical power quality monitoring data; The preliminary clustering module is to cluster the i-th sample As the core point, the core point , let the core point k nearest neighbor set be NN(x i ); For each core point , and belongs to NN(x i ) to form a subgraph ; Each subgraph corresponds to a cluster; introduce density consistency factor , calculate the merger index, the formula used is: ; ; ; ;in, is the merger index; i and j are both sample indices, i≠j; The core point The obtained subgraph; and are the local densities of samples x and y, respectively; It is a subgraph and subgraphs The closest Euclidean distance between them; is the density normalization parameter; is the k-nearest neighbor set of the sample; iterative merging of the maximum Yes, until the number of remaining clusters is K, and the final core clustering is obtained; It is a subgraph samples; It is a subgraph A sample of ; e is a natural constant.

2. The intelligent photovoltaic power quality control system according to claim 1, characterized in that: The feature space trend graph construction module is to construct a fully connected graph between samples, with edge weights as Euclidean distances, and initialize all samples to be unvisited; for each sample u, the elbow distance between the sample and its k nearest neighbors is calculated. , connecting Euclidean distance Less than neighbors; take sample u as the leaf node and sample v as the parent node, jump to the nearest neighbor each time, and take the stopping point as the root node; repeat until all points are visited, and collect all root node sets as the core set R; where, is the Euclidean distance from u to v; k nearest neighbors is the set of k samples with the smallest Euclidean distance; elbow distance It is the distance corresponding to the point with the largest curvature on the curve after plotting the Euclidean distance from sample u to all other k-nearest neighbors and sorting them by size; v is the sample among u's k-nearest neighbors.

3. The intelligent photovoltaic power quality control system according to claim 2, characterized in that: The isolated correction module uses the anomaly score to fuse the local outlier factor LOF to detect isolated points and calculate the anomaly score of each sample. , the formula used is: ; ; Calculating the global anomaly score , the formula used is: ;in, is the length of the isolated path; E[·] is the expectation; It is a sample the expected path length required to be isolated; is the sample normalization constant; is a harmonic number; is the anomaly weight; is the local outlier factor of the sample; is the maximum value of the local outlier factor among all samples; is the jth sample; N is the size of each tree sample subset when the sample participates in the construction of the isolation forest; according to the threshold, the samples with high global anomaly scores are marked as sparse areas SR, and the rest are dense areas DR; for each trend tree edge (x a ,x b ), if one end x a Belongs to SR, the other end x b If it belongs to DR, disconnect the trend tree edge and treat both ends as new root nodes and add R; x a and x b These are the two ends of the trend tree edge.

4. The intelligent photovoltaic power quality control system according to claim 3, characterized in that: The boundary perturbation allocation module is used to treat samples that are not classified into any cluster as boundary points. , calculate its allocation confidence for the c-th cluster , the formula used is: ;in, It is a boundary point and The total Euclidean distance between points; the boundary points are assigned to the cluster with the maximum distribution confidence; for the clustering results of the historical power quality monitoring data after boundary disturbance allocation, the silhouette coefficient is evaluated and the silhouette coefficient threshold is set. If the average silhouette coefficient of all samples is lower than the silhouette coefficient threshold, clustering optimization is performed; otherwise, the clustering of the historical power quality monitoring data is completed.

5. The intelligent photovoltaic power quality control system according to claim 4, characterized in that: The cluster optimization module specifically includes the following contents: Initialize individual positions; construct parameter optimization space based on target cluster number, density normalization parameter, k and silhouette coefficient threshold; quasi-initialization of the qth dimension of each particle is expressed as: ; Generate individual positions and obtain the final initialized particles. The formula used is: The average silhouette coefficient of all samples clustered based on individual positions is used as the individual fitness value; The lower the individual fitness value, the worse the solution corresponding to the individual; is the quasi-initial position of the mth particle in the qth dimension; and are the upper and lower limits of the qth dimension respectively; is uniformly distributed Random number; is the final initial position of the mth particle in the qth dimension; rnd(·,·) is a function that takes uniform random values in a given interval; Particle position update; introduce the position dynamic balance mechanism and define the time function. The formula used is: ; ; The formula used to update speed and position is: ; ;in, is the dynamic cognitive learning factor; g is the current iteration number; is the maximum number of iterations; is the inertia weight; is the search mode switching threshold; and are the q-th dimension velocities of the m-th particle at the g+1-th iteration and the g-th iteration, respectively; 、 and They are all random numbers between 0 and 1, independent of each other; and are the positions of the qth dimension of the mth particle at the g+1th iteration and the gth iteration respectively; It is a social learning factor; is the position of the optimal particle in the qth dimension; F is a random number that obeys the uniform distribution U(0,1), and Independent of each other; is the historical optimal position of the mth particle in the qth dimension; Local search enhancement: every k iterations, the 10% individuals with the lowest fitness values are perturbated and adjusted. The formula used is: ;in, is the position after local search enhancement; The degrees of freedom are t-distributed random variable; Optimization judgment; when there is an individual fitness value higher than the silhouette coefficient threshold, the clustering result of the historical power quality monitoring data is obtained based on the individual position; if the maximum number of iterations is reached, the individual position is returned to the initialization; otherwise, it returns to the particle position update.

6. The intelligent photovoltaic power quality control system according to claim 5, characterized in that: The photovoltaic data acquisition module collects historical power quality monitoring data according to a fixed sampling time sequence during the operation of the photovoltaic system, pre-processes the collected data, and forms a sample set. ,in, and are the 1st and nth samples respectively; each sample is an extracted feature vector; the operating status of each sample is labeled, and the operating status is used as a data label and is only selected and used as a cluster label; the operating status includes normal operation, mild abnormality, moderate abnormality and severe abnormality.

7. The intelligent photovoltaic power quality control system according to claim 6, characterized in that: The photovoltaic power quality control module selects the cluster with the largest number of tags as the cluster label based on the clustering results of historical power quality monitoring data; collects real-time power quality monitoring data and performs cluster assignment based on the minimum Euclidean distance principle; corresponds the cluster label of the real-time power quality monitoring data as the power quality monitoring result; increases the sampling frequency of the monitoring data if the monitoring result is moderately abnormal; and performs early warning processing if the monitoring result is seriously abnormal, thereby achieving power quality control.

Images