A photovoltaic data reduction method based on principal component analysis (PCA), FA and data integration

Through principal component analysis and factor analysis technology, dimensionality reduction and timing synchronization processing, combined with spatiotemporal clustering algorithm, photovoltaic data integration is solved, and the problem of poor integration effect in complex environments of photovoltaic data is achieved efficient monitoring and optimization of photovoltaic system data.

CN118760934BActive Publication Date: 2024-12-06NORTH CHINA ELECTRIC POWER UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411238968.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-12-06
Estimated Expiration
2044-09-05

AI Technical Summary

Technical Problem

The uneven temporal and spatial distribution, dispersed layout, confusing data and multi-source nature of photovoltaic data lead to poor data integration results, and poor data regulation in complex environments.

Method used

Principal component analysis (PCA) is used to reduce dimensionality, and timing synchronization is performed through filtering, interpolation and mixed timing data processing, R-type and Q-type factor analysis is further carried out to obtain objective classification results, build a spatio-temporal clustering matrix and use data integration Functions perform multi-source heterogeneous data integration.

Benefits of technology

Effectively reduce the dimension of photovoltaic data, improve data processing and analysis efficiency, and realize monitoring, analysis and optimization of photovoltaic systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118760934B_ABST
    Figure CN118760934B_ABST
Patent Text Reader

Abstract

The present invention discloses a photovoltaic data reduction method based on principal component analysis PCA, FA and data integration, which relates to the field of electric power technology, including: obtaining an original photovoltaic data set, reducing the dimension of the original photovoltaic data set, and obtaining photovoltaic data after dimension reduction; collecting time series data by filtering, interpolating and hybrid time series data, and synchronously processing chaotic data with asynchronous time series to obtain photovoltaic data after synchronization; objectively classifying the photovoltaic data after synchronization to obtain objective classification results; constructing a spatiotemporal clustering matrix, constructing a data integration function based on the spatiotemporal clustering matrix, inputting the objective classification results into the data integration function, and judging whether the heterogeneous conditions are met. If so, multi-source heterogeneous time series data integration is obtained, and if not, the data integration function is re-input. The present invention realizes the monitoring, analysis and optimization of photovoltaic systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of electric power technology, and in particular relates to a photovoltaic data reduction method based on principal component analysis (PCA), FA and data integration. Background Art

[0002] Due to the current uneven spatiotemporal distribution, scattered layout, chaotic data, disorder and multi-source of photovoltaic data, the data integration effect is still poor under the above circumstances. The data is disordered and multi-source. In view of the poor performance of data integration in complex environments in current photovoltaic data reduction research, the present invention first reduces the dimension of the collected photovoltaic data, then performs time series synchronization processing, further classifies and analyzes it, and finally integrates the multi-source heterogeneous photovoltaic data. The above steps are important links in photovoltaic data reduction, and the goal of photovoltaic data reduction is to process these data, which is of great significance for achieving better monitoring, analysis and optimization of photovoltaic system performance. Summary of the invention

[0003] To solve the above technical problems, the present invention proposes a photovoltaic data reduction method based on principal component analysis (PCA), FA and data integration, which reduces the dimension of photovoltaic data, then performs data time series synchronization processing, further classifies and analyzes it, and finally integrates multi-source heterogeneous photovoltaic data, so as to more effectively process and analyze the large amount of data generated by the photovoltaic system, and ultimately realize the monitoring, analysis and optimization of the photovoltaic system.

[0004] To achieve the above object, the present invention provides a photovoltaic data reduction method based on principal component analysis (PCA), FA and data integration, comprising:

[0005] Acquire an original photovoltaic data set, and perform dimension reduction on the original photovoltaic data set to obtain photovoltaic data after dimension reduction;

[0006] The photovoltaic data after dimension reduction are collected in three ways: filtering, interpolation and hybrid time series data, and the chaotic data with asynchronous time series are synchronously processed to obtain the photovoltaic data after synchronization;

[0007] Objectively classifying the synchronously processed photovoltaic data to obtain objective classification results;

[0008] Construct a spatiotemporal clustering matrix, construct a data integration function based on the spatiotemporal clustering matrix, input the objective classification result into the data integration function, and determine whether the heterogeneous conditions are met. If so, obtain multi-source heterogeneous time series data integration; if not, re-input the data integration function.

[0009] Optionally, the method for obtaining the photovoltaic data after dimension reduction is:

[0010] ;

[0011] in, is the photovoltaic data after dimension reduction; is the corresponding eigenvector The coefficient of is the number of eigenvectors.

[0012] Optionally, in the process of collecting time series data through filtering, interpolation and hybrid time series data, filtering requires selecting the largest monitoring data interval as the time series. If time series data is missing, the time series data closest to the time node is selected to replace it; interpolation requires using the minimum monitoring data as the time series data, and filling the missing number by collecting the average value of multi-source heterogeneous time series data at the nearest point.

[0013] Optionally, the method of objectively classifying the synchronously processed photovoltaic data to obtain an objective classification result includes:

[0014] R-type factor analysis and Q-type factor analysis are performed on the photovoltaic data after synchronous processing, and the R-type and Q-type factor loading matrices are calculated respectively. The Q-type factor loadings F0 and F1 of the collected photovoltaic data are used to draw a plane graph, and the R-type factor loadings of the abnormal data are projected onto the plane graph to form a point cluster composed of abnormal data and normal data; the correlation statistics and analysis between abnormal data and abnormal data, abnormal data and normal data, and normal data and normal data are carried out to obtain objective classification results of photovoltaic data.

[0015] Optionally, the method of performing R-type factor analysis on the synchronously processed photovoltaic data includes:

[0016] Assuming there are n samples and each sample has m variables, the matrix is:

[0017] ;

[0018] Where: , and at least one data in each row and column is not 0;

[0019] according to

[0020] ;

[0021] Transform, and They are

[0022] ;

[0023] ;

[0024] According to the eigenvalues ​​of the variable covariance matrix , take the first P eigenvalues ​​whose cumulative eigenvalue percentage ≥ 80% , calculate the corresponding unit eigenvector , we get the R-type factor loading matrix:

[0025] .

[0026] Optionally, the method of performing Q-factor analysis on the synchronously processed photovoltaic data includes:

[0027] Depend on The Q-type factor loading matrix is ​​obtained as follows:

[0028] ;

[0029] The sample scatter plot is made and analyzed on the factor plane according to the Q-type factor loading matrix, which is called Q-type factor analysis, to study the correlation between photovoltaic data.

[0030] Optionally, the method for constructing a data integration function is:

[0031] ;

[0032] Where D is the data integration function; It is the minimum parameter for multi-source heterogeneous time series data; It is the maximum parameter of multi-source heterogeneous time series data; is the within-sample variable; is the number of samples; is the data value after R-type factor analysis.

[0033] Optionally, the heterogeneous condition is the expectation of multi-source heterogeneous time series data, which is calculated as follows:

[0034] ;

[0035] in, is a statistic; is the expected value of the statistic; It is the weight of multi-source heterogeneous time series data; is the number of samples; The data are processed by R-type factor analysis.

[0036] Technical effect of the invention: The present invention discloses a photovoltaic data reduction method based on principal component analysis PCA, FA and data integration, which reduces the dimension of photovoltaic data, then performs data time series synchronization processing, further classifies and analyzes it, and finally integrates multi-source heterogeneous photovoltaic data, so as to more effectively process and analyze the large amount of data generated by the photovoltaic system, and finally realize the monitoring, analysis and optimization of the photovoltaic system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0038] Figure 1 This is a schematic diagram of a data processing process with asynchronous timing according to an embodiment of the present invention;

[0039] Figure 2 Schematic diagram of the data integration process of an embodiment of the present invention. DETAILED DESCRIPTION

[0040] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0042] This embodiment provides a photovoltaic data reduction method based on principal component analysis PCA, FA and data integration, including:

[0043] Step 1: By using the dimensionality reduction technology of principal component analysis, the dimension of the original data set can be significantly reduced while retaining key information, thereby reducing the cost of data processing and storage and improving computing efficiency;

[0044] Step 2: Aggregate time series data by filtering, interpolating, and mixing time series data, and synchronize chaotic data with asynchronous time series;

[0045] Step 3: Calculate the R-type and Q-type factor loading matrices respectively to obtain objective classification results of photovoltaic data;

[0046] Step 4: Use spatiotemporal clustering algorithm to achieve data integration.

[0047] Step 1: Data dimensionality reduction specifically includes:

[0048] PCA transforms the original photovoltaic data into a set of linearly independent representations in each dimension through linear transformation, which can achieve dimensionality reduction of high-dimensional data. The main steps to achieve PCA dimensionality reduction are:

[0049] (1) For a given set of data sample points , each sample has p-dimensional features, that is Without loss of generality, the data set is centralized, that is, the mean of the sample data is calculated , the sample after centering is recorded as .

[0050] (2) The central idea of ​​principal component analysis is to reduce the dimension of a data set while retaining as much variance as possible. From a geometric perspective, principal component analysis is to rotate the original coordinate axes to obtain new orthogonal coordinate axes that can disperse the data points sufficiently. The implementation method is to first calculate the covariance matrix of the original data set, that is, , and then perform eigendecomposition on the covariance matrix to obtain the eigenvalues ​​of the covariance matrix The corresponding feature (Also called mode or base mode). The eigenvalues ​​are sorted from large to small, and the eigenvalues ​​correspond to the eigenvectors one by one. The size of the eigenvalue represents the number of features contained in the corresponding eigenvector. In PCA, the contribution rate is used as an indicator to calculate the degree to which each mode represents the amount of information in the original data:

[0051] ;

[0052] (3) According to the contribution rate, the eigenvectors corresponding to the first k eigenvalues ​​are taken as the main modes after dimensionality reduction. It is generally believed that when the cumulative contribution rate of the current k main modes reaches more than 90%, it has the ability to describe the main characteristics of the original data. After dimensionality reduction, the original centralized data set can be characterized as:

[0053] ;

[0054] in, is the photovoltaic data after dimension reduction; is the corresponding eigenvector The coefficient of is the number of eigenvectors.

[0055] Step 2: Photovoltaic data timing synchronization processing specifically includes:

[0056] To obtain objective classification of photovoltaic data after dimension reduction, the time series data will be aggregated through three methods: filtering, interpolation, and hybrid time series data, and the chaotic data with asynchronous time series will be synchronized, such as Figure 1 shown.

[0057] In the filtering process, the largest monitoring data interval is selected as the time series. If the time series data is missing, the time series data closest to the time node is selected to replace it. In contrast to the filtering process, the minimum monitoring data is used as the time series data in the interpolation process, and the missing data is filled by collecting the average value of the multi-source heterogeneous time series data at the nearest point.

[0058] The time for the collection of mixed time series data is determined by the content of the information set. Usually, one of the time series data in the information set is regarded as a new time or a new time series is redefined. The monitoring time series number is generated based on the new collection order between each time series data.

[0059] After completing the data collection, the multi-source heterogeneous time series data are processed collaboratively. After sorting the time series data sources, three basic characteristics are added to the fields of the time series data, including the time series data flag, start date, and time interval. After synchronization, the time series data is entered into the virtual photovoltaic database according to the synchronization mechanism.

[0060] Time series data synchronization is performed by analyzing the coordination degree of the data aggregation layer. The synchronization of time series data has a great influence on the coordination degree of data aggregation. When the aggregated data are all time series data, the time series data synchronization mechanism starts to run, starting from the data aggregation layer to collect all the time series data in the time series data, such as the name, date, size, sampling interval, destination of the time series data, etc. and its synchronization method, so as to achieve data synchronization.

[0061] Step 3: Objective classification of PV data includes:

[0062] The data after dimensionality reduction are further subjected to factor analysis, and the R-type and Q-type factor loading matrices are calculated respectively. The Q-type factor loadings F0 and F1 of the collected photovoltaic data are used to draw a plane graph, and then the R-type factor loadings of the abnormal data are projected onto the graph, and finally a point cluster consisting of abnormal data and normal data is formed. Based on this, the correlation statistics and analysis between abnormal data and abnormal data, abnormal data and normal data, and normal data and normal data are carried out to obtain the objective classification results of photovoltaic data.

[0063] (1) R-type factor analysis

[0064] Assuming there are n samples and each sample has m variables, the matrix is:

[0065] ;

[0066] Where: , and at least one data in each row and column is not 0.

[0067] ;

[0068] Transform formula (4) and They are

[0069] ;

[0070] ;

[0071] According to the eigenvalues ​​of the variable covariance matrix , take the first P eigenvalues ​​whose cumulative eigenvalue percentage ≥ 80% , calculate the corresponding unit eigenvector , we get the R-type factor loading matrix:

[0072] .

[0073] (2) Q-factor analysis

[0074] Depend on The Q-type factor loading matrix is ​​obtained as follows:

[0075] ;

[0076] The sample scatter plot is made and analyzed on the factor plane according to the Q-type factor loading matrix, which is called Q-type factor analysis, to study the correlation between photovoltaic data.

[0077] Step 4: Data integration includes: Figure 2 As shown:

[0078] (1) Spatiotemporal clustering matrix

[0079] When the results of the spatiotemporal set calibration are known, a spatiotemporal clustering matrix is ​​established. In the spatiotemporal clustering matrix, multi-source heterogeneous time series data are divided into two forms: horizontal and vertical. In the process of determining the horizontal multi-source heterogeneous time series data, the nearest neighbor analysis algorithm in the spatiotemporal clustering algorithm is used, and its expression is:

[0080] ;

[0081] Where R represents the nearest neighbor analysis result; dobs represents the average distance between the node and the nearest neighbor during the observation process; dexp represents the expected value of the node under different states.

[0082] Since the nearest neighbor algorithm only considers the point-to-point situation in the horizontal data, it is necessary to introduce the Moran index in the spatiotemporal clustering algorithm when analyzing the vertical multi-source heterogeneous time series data. , the determination of the longitudinal multi-source heterogeneous time series is realized by calculating the expected value I of the Moran index:

[0083] ;

[0084] Where n is the number of vertical arrangements. The spatiotemporal clustering matrix can be used as an accurate information discrimination condition. The spatiotemporal clustering matrix is ​​used to describe the storage capacity of the spatiotemporal database. When the amount of data in the spatiotemporal database reaches the target value, the spatiotemporal clustering matrix can be considered to be saturated.

[0085] (2) Data integration function

[0086] The data integration function is a query restriction condition for multi-source heterogeneous time series data based on the spatiotemporal clustering matrix. Since the spatiotemporal clustering algorithm is highly applicable, as the amount of data information increases, the coverage area of ​​the limit value of the integration function will continue to increase until it can meet the transmission requirements of multi-source heterogeneous time series data. The data integration function D can be expressed as:

[0087] ;

[0088] Where D is the data integration function; is the within-sample variable; is the number of samples; is the data value after R-type factor analysis; Represents the minimum parameter of multi-source heterogeneous time series data, Represents the maximum parameter of multi-source heterogeneous time series data, and takes into account the transmission of chaotic information parameters. and The larger the difference, the better the performance of the data integration function.

[0089] (3) Integration of multi-source heterogeneous time series data

[0090] Multi-source heterogeneous time series data integration is to transform all data using the spatiotemporal clustering algorithm. Data integration is mainly divided into two processes, namely data separation and data integration. The integration method is selected through the hierarchical structure of data integration. Each multi-source heterogeneous time series data information is regarded as a class, and larger and larger classes are integrated according to the rules of the spatiotemporal clustering algorithm until the preset conditions are met. The statistic G in the spatiotemporal clustering algorithm is used to judge the expectation of multi-source heterogeneous time series data. The expression is:

[0091] ;

[0092] in, is a statistic; is the expected value of the statistic; It is the weight of multi-source heterogeneous time series data; is the number of samples; The data are processed by R-type factor analysis.

[0093] The time series clustering results are grouped according to the measurement of similarity, and the similarity of the time series in the database is compared. The data with high similarity in the same cluster are grouped together, while the information data in different clusters are quite different. The integration of multi-source heterogeneous time series is to mine data information in the same attribute time series database in different time domains. Whether it is data separation or data integration, users can design the separation and integration requirements of multi-source heterogeneous time series data according to their own requirements, so as to better realize the integration of multi-source heterogeneous time series data.

[0094] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A photovoltaic data reduction method based on principal component analysis (PCA), FA and data integration. It is characterized in that include: Acquire an original photovoltaic data set, and perform dimension reduction on the original photovoltaic data set to obtain photovoltaic data after dimension reduction; The photovoltaic data after dimension reduction are collected in three ways: filtering, interpolation and hybrid time series data, and the chaotic data with asynchronous time series are synchronously processed to obtain the photovoltaic data after synchronization; Objectively classifying the synchronously processed photovoltaic data to obtain objective classification results; Constructing a spatiotemporal clustering matrix, constructing a data integration function based on the spatiotemporal clustering matrix, inputting the objective classification result into the data integration function, judging whether the heterogeneous condition is met, and obtaining multi-source heterogeneous time series data integration if it is met, and re-inputting the data integration function if it is not met; The method to obtain the photovoltaic data after dimension reduction is: ; in, is the photovoltaic data after dimension reduction; is the corresponding eigenvector The coefficient of is the number of eigenvectors; In the process of collecting time series data through filtering, interpolation and mixed time series data, filtering needs to select the largest monitoring data interval as the time series. If the time series data is missing, the time series data closest to the time node is selected to replace it; interpolation needs to use the minimum monitoring data as the time series data, and fill the missing data by collecting the average value of multi-source heterogeneous time series data at the nearest point; The method of objectively classifying the synchronously processed photovoltaic data to obtain an objective classification result includes: Perform R-type factor analysis and Q-type factor analysis on the photovoltaic data after synchronous processing, calculate the R-type and Q-type factor loading matrices respectively, draw a plane diagram using the Q-type factor loadings F0 and F1 of the collected photovoltaic data, project the R-type factor loadings of the abnormal data onto the plane diagram, and form a point cluster composed of abnormal data and normal data; carry out statistics and analysis on the correlation between abnormal data and abnormal data, abnormal data and normal data, and normal data and normal data, and obtain objective classification results of photovoltaic data; The methods for performing R-type factor analysis on synchronously processed photovoltaic data include: Assuming there are n samples and each sample has m variables, the matrix is: ; Where: , and at least one data in each row and column is not 0; according to ; Transform, and They are ; ; According to the eigenvalues ​​of the variable covariance matrix , take the first P eigenvalues ​​whose cumulative eigenvalue percentage ≥ 80% , calculate the corresponding unit eigenvector , we get the R-type factor loading matrix: ; The method of performing Q-factor analysis on synchronously processed photovoltaic data includes: Depend on The Q-type factor loading matrix is ​​obtained as follows: ; According to the Q-type factor loading matrix, sample scatter plots are made on the factor plane and analyzed, which is called Q-type factor analysis to study the correlation between photovoltaic data; The method to construct the data integration function is: ; Where D is the data integration function; It is the minimum parameter for multi-source heterogeneous time series data; It is the maximum parameter of multi-source heterogeneous time series data; is the within-sample variable; is the number of samples; is the data value after R-type factor analysis; The heterogeneous condition is the expectation of multi-source heterogeneous time series data, which is calculated as follows: ; in, is a statistic; is the expected value of the statistic; It is the weight of multi-source heterogeneous time series data; is the number of samples; The data are processed by R-type factor analysis.