A method and system for identifying and cleaning abnormal wind power data

By constructing a comprehensive abnormal data identification result matrix and optimization algorithm, combining multiple features and a single abnormal data cleaning method, the problem of excessive deletion or data preservation in wind power abnormal data cleaning is solved, and the applicability and effectiveness of data cleaning is improved.

CN114996253BActive Publication Date: 2024-11-05XIANGTAN UNIV

Patent Information

Application Number
CN202210652602.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-06-07
Publication Date
2024-11-05
Estimated Expiration
2042-06-07

AI Technical Summary

Technical Problem

Existing methods for cleaning abnormal data of wind power are prone to excessively deleting normal data or excessively retaining abnormal data, and are weak in applicability and cannot effectively ensure the data cleaning effect.

Method used

By constructing a comprehensive abnormal data identification result matrix, combining multiple features and a single abnormal data cleaning method, the weight coefficient and data cleaning threshold are obtained using an optimization algorithm, and the data cleaning threshold are cleaned using a comprehensive scoring indicator to avoid deviations from a single evaluation indicator.

Benefits of technology

It improves the applicability and effectiveness of data cleaning, avoids excessive deletion of normal data and excessive preservation of abnormal data, and achieves more reasonable abnormal data identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114996253B_ABST
    Figure CN114996253B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for wind power abnormal data identification and cleaning, which is characterized by being performed in accordance with the following steps: step S1, step S2, step S3, step S4, step S5, step S6 and step S7. The present invention can comprehensively consider a variety of original data information by constructing a comprehensive abnormal data identification result matrix CA; by optimizing the weight coefficient ω i The data cleaning threshold F is optimized to avoid artificially setting weights and thresholds; a comprehensive scoring index SOCSI is proposed to achieve a comprehensive evaluation of the complex data cleaning effect. The wind power abnormal data identification and cleaning method and system proposed in the present invention can avoid problems such as excessive deletion of normal data and excessive retention of abnormal data, and effectively improve the applicability of the abnormal data cleaning method.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of wind power technology, and in particular to a method and system for identifying and cleaning abnormal wind power data. Background Art

[0002] High-quality data cleaning results are the guarantee for describing the actual output of wind turbines in each wind speed range, and are the basis for data analysis, mining and wind power forecasting.

[0003] Existing wind power data cleaning methods are mainly divided into two categories. The first category is to establish a power curve boundary model and regard the data outside the power curve boundary as abnormal data; the second category is to distinguish and clean abnormal data based on the different mathematical statistical characteristics of abnormal data and normal data, or to detect based on the location distribution of outliers.

[0004] Since the existing abnormal data cleaning methods are based on a single mathematical statistical characteristic, and the wind power abnormal data evaluation indicators are often evaluated from a unilateral perspective, setting weights and thresholds based on a single evaluation indicator will result in excessive deletion of normal data, excessive retention of abnormal data, or weak applicability.

[0005] In summary, how to ensure that normal data is not excessively deleted and abnormal data is not excessively retained, and how to improve the cleaning effect and applicability, is an urgent problem that needs to be solved in the field of wind power abnormal data cleaning. Summary of the invention

[0006] The purpose of the present invention is to solve the problems of excessive deletion of normal data and excessive retention of abnormal data or weak applicability, and to provide a method and system for identifying and cleaning abnormal data of wind power.

[0007] To achieve the above object, the present invention provides a method and system for identifying and cleaning abnormal wind power data, the method comprising the following steps:

[0008] Obtain the original SCADA system data, preliminarily eliminate the original SCADA system data, and establish the original data matrix H zk ;

[0009] According to the numerical values ​​of u features, the original data matrix H is zk Sort in ascending order to obtain the sorted data matrix H n zk, ,n=1,2,...,u;

[0010] Using m single abnormal data cleaning methods, the sorted data matrix H n zk Perform abnormal power data identification and obtain u×m types of abnormal data identification result matrix A izt ; i=1,2,3…u×m; t=1,2;

[0011] The u×m types of abnormal data recognition result matrix A are respectively i zt The second column of is sorted in ascending order to obtain the sorted u×m abnormal data recognition result matrix B i zt ;i=1,2,3...u×m; t=1,2;

[0012] Through the optimization algorithm Opt, with the goal of maximizing the comprehensive score index SOCSI, the u×m types of abnormal data recognition result matrix B is obtained by searching for the best result. i zt The weight coefficient ω i and data cleaning threshold F; i=1,2,3…u×m;

[0013] The sorted u×m abnormal data recognition result matrix B i zt The first column in is calculated as follows to perform weighted combination to construct a comprehensive abnormal data recognition result matrix CA; the dimension of the matrix CA is z rows and 1 column;

[0014]

[0015] Among them, CA is the comprehensive abnormal data recognition result matrix, z is the B i zt The total number of sample data; ω i is the weight coefficient of the matrix corresponding to the i-th abnormal data recognition result;

[0016] According to the data cleaning threshold F, the comprehensive abnormal data identification result matrix CA is cleaned to obtain an abnormal data set AD and a normal data set ND.

[0017] Optionally, the original data matrix H zk The number of columns is k and the number of rows is z; the original data matrix H zk A row is defined as a single sample; the original data matrix H zk The first column is the original data matrix H zk The sample number, the original data matrix H zk The remaining columns are the original data matrix H zk The characteristics of the original data matrix H zkThe characteristics include: power characteristics, wind speed characteristics, wind direction characteristics, temperature characteristics, air pressure characteristics, humidity characteristics, turbulence characteristics, wind shear characteristics, air density characteristics, yaw error characteristics, pitch angle characteristics, rotation speed characteristics, and tip speed ratio characteristics.

[0018] Optionally, the u features are the original data matrix H zk Any u characteristics except power characteristics, optionally, the u characteristics are wind speed characteristics, rotation speed characteristics, and tip speed ratio characteristics, and u=3.

[0019] Optionally, the m types of single abnormal data cleaning methods include the Raida criterion method, the quartile method, the sliding standard deviation method, and the density-based clustering method with noise, m=4; the recognition result matrix A i zt The first column is the sorted data matrix H n zk The judgment result of abnormal data is composed of 0 and 1, where 0 represents abnormality and 1 represents normality; the recognition result matrix A i zt The second column is the sorted data matrix H n zk The sample serial number.

[0020] Optionally, the optimization algorithm Opt may be any one of a genetic algorithm, a particle swarm algorithm, a hybrid algorithm of mathematical morphology clustering and a fruit fly optimization algorithm, an artificial bee colony algorithm, or a chaotic cuckoo search algorithm; the comprehensive scoring index SOCSI is calculated according to the following formula:

[0021]

[0022] Among them, SOCSI is the comprehensive scoring index, SI i N SI The score values ​​corresponding to the single evaluation indicators are: and are the subjective weight coefficient and objective weight coefficient of the single evaluation index respectively; α is a subjective experience factor, which is optional and takes α=0.5; the N SI The single evaluation indicators include: data retention rate, Pearson coefficient, mean standard deviation and coefficient of variation, N SI =4; the SI i The calculation method is: sort the original index values ​​of the data retention rate and the Pearson coefficient in ascending order, and the resulting sequence number is SI i ; Sort the original index values ​​of the average standard deviation and the coefficient of variation in descending order, and the resulting serial number is SI i .

[0023] Optionally, the data cleaning is specifically as follows: if the sample point in the i-th row of the matrix CA is lower than a threshold F, then in the original data matrix H zk In the above, all the feature data in the i-th row are added to the abnormal data set AD; if the sample point in the i-th row of the matrix CA is higher than the threshold F, then in the original data matrix H zk In , all feature data in the i-th row are added to the normal data set ND.

[0024] The present invention also provides a wind power abnormal data identification and cleaning system, the system comprising:

[0025] Module M1: Obtain the original SCADA system data, preliminarily eliminate the original SCADA system data, and establish the original data matrix H zk ;

[0026] Module M2: The original data matrix H is processed according to the numerical values ​​of u features. zk Sort in ascending order to obtain the sorted data matrix H n zk ,n=1,2,...,u;

[0027] Module M3: Use m single abnormal data cleaning methods to clean the sorted data matrix H n zk Perform abnormal power data identification and obtain u×m types of abnormal data identification result matrix A i zt ; i=1,2,3…u×m; t=1,2;

[0028] Module M4: respectively identify the result matrix A of the u×m types of abnormal data i zt The second column of is sorted in ascending order to obtain the sorted u×m abnormal data recognition result matrix B i zt ; i=1,2,3…u×m; t=1,2;

[0029] Module M5: Through the optimization algorithm Opt, with the goal of maximizing the comprehensive scoring index SOCSI, the weight coefficient ω of the u×m types of abnormal data recognition result matrix is ​​obtained. i and data cleaning threshold F; i = 1, 2, 3…u×m;

[0030] Module M6: The sorted u×m abnormal data recognition result matrix B i ztThe first column in is calculated by the following formula for weighted combination to construct a comprehensive abnormal data recognition result matrix CA; the dimension of the matrix CA is z rows and 1 column;

[0031]

[0032] Where: CA is the comprehensive abnormal data recognition result matrix, z is the B i zt The total number of sample data; ω i is the weight coefficient of the matrix corresponding to the i-th abnormal data recognition result;

[0033] Module M7: According to the data cleaning threshold F, the comprehensive abnormal data identification result matrix CA is cleaned to obtain an abnormal data set AD and a normal data set ND.

[0034] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0035] 1. The present invention constructs a comprehensive abnormal data identification result matrix CA through u features of the original data matrix and m single abnormal data cleaning methods, which can comprehensively consider multiple original data information to avoid using a single data statistical feature, thereby making abnormal data identification more reasonable and improving the applicability of the data cleaning method.

[0036] 2. The optimization algorithm Opt proposed in this invention has a weight coefficient ω for the abnormal data recognition result matrix i Optimizing the data cleaning threshold F can avoid the adverse effects of artificially setting weights and thresholds on data cleaning.

[0037] 3. The comprehensive scoring index SOCSI proposed in the present invention quantitatively evaluates the data cleaning effect, which can avoid the data cleaning threshold being set too low or too high due to a single evaluation index, and can achieve a comprehensive evaluation of the complex data cleaning effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0039] Figure 1 The present invention is a flow chart of an embodiment of a method and system for identifying and cleaning abnormal wind power data. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] The purpose of the present invention is to provide a method and system for identifying and cleaning abnormal data of wind power generation, so as to solve the problems of excessive deletion of normal data and excessive retention of abnormal data or weak applicability.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] An embodiment of a method and system for identifying and cleaning abnormal wind power data of the present invention comprises the following steps:

[0044] Step S1, obtaining the original SCADA system data, preliminarily eliminating the original SCADA system data, and establishing the original data matrix H zk ;

[0045] Optionally, the original data matrix H zk The number of columns is k and the number of rows is z; the original data matrix H zk A row is defined as a single sample; the original data matrix H zk The first column is the original data matrix H zk The sample number, the original data matrix H zk The remaining columns are the original data matrix H zk The characteristics of the original data matrix H zk The characteristics include: power characteristics, wind speed characteristics, wind direction characteristics, temperature characteristics, air pressure characteristics, humidity characteristics, turbulence characteristics, wind shear characteristics, air density characteristics, yaw error characteristics, pitch angle characteristics, rotation speed characteristics, and tip speed ratio characteristics.

[0046] Step S2: the original data matrix H is sorted according to the numerical values ​​of u features. zk Sort in ascending order to obtain the sorted data matrix H n zk ,,n=1,2,...,u;

[0047] Optionally, the u features are the original data matrix H zk Any u characteristics except power characteristics, optionally, the u characteristics are wind speed characteristics, rotation speed characteristics, and tip speed ratio characteristics, and u=3.

[0048] Step S3: Using m single abnormal data cleaning methods, the sorted data matrix H n zk Perform abnormal power data identification and obtain u×m types of abnormal data identification result matrix A i zt ; i=1,2,3…u×m; t=1,2;

[0049] Optionally, the m types of single abnormal data cleaning methods include the Raida criterion method, the quartile method, the sliding standard deviation method, and the density-based clustering method with noise, m=4; the recognition result matrix A i zt The first column is the sorted data matrix H n zk The judgment result of abnormal data is composed of 0 and 1, where 0 represents abnormality and 1 represents normality; the recognition result matrix A i zt The second column is the sorted data matrix H n zk The sample serial number.

[0050] Step S4: respectively identify the u×m types of abnormal data result matrix A i zt The second column of is sorted in ascending order to obtain the sorted u×m abnormal data recognition result matrix B i zt ; i=1,2,3…u×m; t=1,2;

[0051] Step S5: By optimizing the algorithm Opt, with the goal of maximizing the comprehensive score index SOCSI, the u×m types of abnormal data recognition result matrix B is obtained by searching for the optimal solution. i zt The weight coefficient ω i and data cleaning threshold F; i = 1, 2, 3…u×m;

[0052] Optionally, the optimization algorithm Opt may be any one of a genetic algorithm, a particle swarm algorithm, a hybrid algorithm of mathematical morphology clustering and a fruit fly optimization algorithm, an artificial bee colony algorithm, or a chaotic cuckoo search algorithm; the comprehensive scoring index SOCSI is calculated according to the following formula:

[0053]

[0054] In the formula, SOCSI is the comprehensive scoring index, SI i N SI The score values ​​corresponding to the single evaluation indicators are: and are the subjective weight coefficient and objective weight coefficient of the single evaluation index respectively; α is a subjective experience factor, which is optional and takes α=0.5; the N SI The single evaluation indicators include: data retention rate, Pearson coefficient, mean standard deviation and coefficient of variation, N SI =4; the SI i The calculation method is: sort the original index values ​​of the data retention rate and the Pearson coefficient in ascending order, and the resulting sequence number is SI i ; Sort the original index values ​​of the average standard deviation and the coefficient of variation in descending order, and the resulting serial number is SI i .

[0055] The sorted u×m abnormal data recognition result matrix B i zt The first column in is calculated by the following formula for weighted combination to construct a comprehensive abnormal data recognition result matrix CA; the dimension of the matrix CA is z rows and 1 column;

[0056]

[0057] Among them, CA is the comprehensive abnormal data recognition result matrix, z is the B i zt The total number of sample data; ω i is the weight coefficient of the matrix corresponding to the i-th abnormal data recognition result;

[0058] Step S7: According to the data cleaning threshold F, the comprehensive abnormal data identification result matrix CA is cleaned to obtain an abnormal data set AD and a normal data set ND.

[0059] Optionally, the data cleaning is specifically as follows: if the sample point in the i-th row of the matrix CA is lower than a threshold F, then in the original data matrix H zk In the above, all the feature data in the i-th row are added to the abnormal data set AD; if the sample point in the i-th row of the matrix CA is higher than the threshold F, then in the original data matrix H zk In , all feature data in the i-th row are added to the normal data set ND.

[0060] Based on the same inventive concept, an embodiment of the present invention also provides a wind power abnormal data identification and cleaning system. Since the principles of solving the problems by these devices are similar to a wind power abnormal data identification and cleaning method, the implementation of these devices can refer to the implementation of the method, and the repeated parts will not be repeated.

[0061] The contents described in the embodiments of this specification are merely an enumeration of implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms described in the embodiments. The protection scope of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for identifying and cleaning abnormal wind power data, characterized in that: The following steps are involved: Step S1: Obtain the original SCADA system data, preliminarily eliminate the original SCADA system data, and establish the original data matrix H zk ; Step S2: The original data matrix H is sorted according to the numerical values ​​of u features. zk Sort in ascending order to obtain the sorted data matrix H n zk ,n=1,2,...,u; Step S3: Using m single abnormal data cleaning methods, the sorted data matrix H n zk Perform abnormal power data identification and obtain u×m types of abnormal data identification result matrix A i zt ;i=1,2,3...u×m; t=1,2; Step S4: Respectively identify the u×m types of abnormal data result matrix A i zt The second column of is sorted in ascending order to obtain the sorted u×m abnormal data recognition result matrix B i zt ;i=1,2,3...u×m; t=1,2; Step S5: By optimizing the algorithm Opt, with the goal of maximizing the comprehensive score index SOCSI, the u×m types of abnormal data recognition result matrix B is obtained by searching for the optimal solution. i zt The weight coefficient ω i and data cleaning threshold F; i = 1, 2, 3 ... u × m; the optimization algorithm Opt is any one of a genetic algorithm, a particle swarm algorithm, a hybrid algorithm of mathematical morphology clustering and fruit fly optimization algorithm, an artificial bee colony algorithm or a chaotic cuckoo search algorithm; the comprehensive scoring index SOCSI is calculated according to formula (2): In the formula, SOCSI is the comprehensive scoring index, SI i N SI The score values ​​corresponding to the single evaluation indicators are: and are respectively the subjective weight coefficient and the objective weight coefficient of the single evaluation index; α is the subjective experience factor, taking α = 0.5; the N SI The single evaluation indicators include: data retention rate, Pearson coefficient, mean standard deviation and coefficient of variation, N SI =4; the SI i The calculation method is: sort the original index values ​​of the data retention rate and the Pearson coefficient in ascending order, and the resulting sequence number is SI i ; or the original index values ​​of the average standard deviation and the coefficient of variation are sorted in descending order, and the resulting serial number is SI i ; Step S6: The sorted u×m abnormal data recognition result matrix B i zt The first column of is calculated according to formula (1) for weighted combination to construct a comprehensive abnormal data recognition result matrix CA; the dimension of the matrix CA is z rows and 1 column; Where: CA is the comprehensive abnormal data recognition result matrix, z is the B i zt The total number of sample data; ω i is the weight coefficient of the matrix corresponding to the i-th abnormal data recognition result; Step S7: According to the data cleaning threshold F, the comprehensive abnormal data identification result matrix CA is cleaned to obtain an abnormal data set AD and a normal data set ND; the data cleaning is specifically as follows: if the sample point in the i-th row of the matrix CA is lower than the threshold F, then in the original data matrix H zk In the above, all the feature data in the i-th row are added to the abnormal data set AD; if the sample point in the i-th row of the matrix CA is higher than the threshold F, then in the original data matrix H zk In , all feature data in the i-th row are added to the normal data set ND.

2. A method for identifying and cleaning abnormal wind power data according to claim 1, characterized in that: The original data matrix H of step S1 zk The number of columns is k and the number of rows is z; the original data matrix H zk A row is defined as a single sample; the original data matrix H zk The first column is the original data matrix H zk The sample number, the original data matrix H zk The remaining columns are the original data matrix H zk Features; The original data matrix H zk The characteristics include: power characteristics, wind speed characteristics, wind direction characteristics, temperature characteristics, air pressure characteristics, humidity characteristics, turbulence characteristics, wind shear characteristics, air density characteristics, yaw error characteristics, pitch angle characteristics, rotation speed characteristics, and tip speed ratio characteristics.

3. A method for identifying and cleaning abnormal wind power data according to claim 1, characterized in that: The u features in step S2 are the original data matrix H zk Any u characteristics except power characteristics.

4. A method for identifying and cleaning abnormal wind power data according to claim 1, characterized in that: The m types of single abnormal data cleaning methods in step S3 are Raida criterion method, quartile method, sliding standard deviation method, and density-based clustering method with noise, where m=4; The recognition result matrix A i zt The first column is the sorted data matrix H n zk The judgment result of abnormal data is composed of 0 and 1, where 0 represents abnormality and 1 represents normality; the recognition result matrix A i zt The second column is the sorted data matrix H n zk The sample serial number.

5. A wind power abnormal data identification and cleaning system, characterized in that: Includes the following modules: Module M1: Obtain the original SCADA system data, preliminarily eliminate the original SCADA system data, and establish the original data matrix H zk ; Module M2: The original data matrix H is processed according to the numerical values ​​of u features. zk Sort in ascending order to obtain the sorted data matrix H n zk ,n=1,2,...,u; Module M3: Use m single abnormal data cleaning methods to clean the sorted data matrix H n zk Perform abnormal power data identification and obtain u×m types of abnormal data identification result matrix A i zt ;i=1,2,3...u×m; t=1,2; Module M4: respectively identify the result matrix A of the u×m types of abnormal data i zt The second column of is sorted in ascending order to obtain the sorted u×m abnormal data recognition result matrix B i zt ;i=1,2,3...u×m; t=1,2; Module M5: Through the optimization algorithm Opt, with the goal of maximizing the comprehensive score index SOCSI, the u×m types of abnormal data recognition result matrix B is obtained by optimization. i zt The weight coefficient ω i and data cleaning threshold F; i = 1, 2, 3 ... u × m; the optimization algorithm Opt is any one of a genetic algorithm, a particle swarm algorithm, a hybrid algorithm of mathematical morphology clustering and fruit fly optimization algorithm, an artificial bee colony algorithm or a chaotic cuckoo search algorithm; the comprehensive scoring index SOCSI is calculated according to formula (2): In the formula, SOCSI is the comprehensive scoring index, SI i N SI The score values ​​corresponding to the single evaluation indicators are: and are respectively the subjective weight coefficient and the objective weight coefficient of the single evaluation index; α is the subjective experience factor, taking α = 0.5; the N SI The single evaluation indicators include: data retention rate, Pearson coefficient, mean standard deviation and coefficient of variation, N SI =4; the SI i The calculation method is: sort the original index values ​​of the data retention rate and the Pearson coefficient in ascending order, and the resulting sequence number is SI i ; or the original index values ​​of the average standard deviation and the coefficient of variation are sorted in descending order, and the resulting serial number is SI i ; Module M6: The sorted u×m abnormal data recognition result matrix B i zt The first column of is weightedly combined according to formula (3) to construct a comprehensive abnormal data recognition result matrix CA; the dimension of the matrix CA is z rows and 1 column; Where: CA is the comprehensive abnormal data recognition result matrix, z is the B i zt The total number of sample data; ω i is the weight coefficient of the matrix corresponding to the i-th abnormal data recognition result; Module M7: According to the data cleaning threshold F, the comprehensive abnormal data identification result matrix CA is cleaned to obtain an abnormal data set AD and a normal data set ND; the data cleaning is specifically as follows: if the sample point in the i-th row of the matrix CA is lower than the threshold F, then in the original data matrix H zk In the above, all the feature data in the i-th row are added to the abnormal data set AD; if the sample point in the i-th row of the matrix CA is higher than the threshold F, then in the original data matrix H zk In , all feature data in the i-th row are added to the normal data set ND.

Citation Information

Patent Citations

  • Cleaning method for wind power data of wind generating set

    CN108412710A

  • Wind turbine generator power characteristic evaluation method based on cloud model and optimal combination empowerment

    CN111461564A

  • Method and system for adaptively calculating IT intelligent operation and maintenance health index

    CN113360358A

  • Yaw calibration method and system for wind turbine

    US20210148331A1

Cited By

  • Abnormal data identification method and device of generator set, and electronic equipment

    CN116361705A

  • Generator set abnormal data identification method and device, and electronic equipment

    CN116361705B