Multi-user feature analysis and numerical control integrated modeling method based on comprehensive correlation

Abstract

The invention discloses a correlation-integrated multi-user feature analysis and numerical control integrated modeling method, particularly relates to the field of power system fault recovery, and comprises a control model correction link, a multi-user feature correlation analysis link and a data-driven model induction link. According to the method, the advantages of a physical model and a data-driven model are integrated, data are corrected through mechanism analysis, the requirements for the number and quality of data samples are reduced, then data-driven modeling is utilized, the model has higher adaptability and predictive capacity, multi-user characteristics can be reflected more accurately, the uncertainty of a comprehensive energy system is effectively dealt with, and the method is suitable for popularization and application. Compared with an existing single model, the energy management and decision are optimized, and the performance is remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of power systems, and more specifically, to a method for multi-user feature analysis and numerical control integrated modeling with comprehensive correlation. Background Art

[0002] There are a lot of uncertainties at the input and output ends of the integrated energy (IES), including the fluctuations of distributed energy such as wind power and photovoltaics, and the real-time changes in energy demand. And those uncertainties that are not included in the scope of advance planning are also the focus of the power grid, such as the sudden and substantial increase in new energy generation or the unplanned and unprepared increase and decrease in load. Therefore, the rational development and utilization of the integrated energy system (IES), the correlation analysis of energy networks, energy production, energy conversion, energy storage and energy consumption, and the establishment of accurate user characteristic models to cope with the large amount of uncertainty in the operation of IES can coordinate and optimize the coupling relationship and conversion use between various energy sources.

[0003] However, user feature modeling mainly relies on the scale and accuracy of data. When data is missing, broken, or noisy, it will be difficult to accurately and efficiently analyze and process the data. The existence of these bad data will affect the accurate analysis of user behavior. Existing identification methods can be roughly divided into bad data identification based on statistical methods and based on machine learning algorithms. However, these methods can only detect a certain type of bad data or take a long time to process.

[0004] Therefore, a multi-user feature analysis and numerical control integrated modeling method with comprehensive correlation was proposed to address the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present application provides a method for multi-user feature analysis and numerical control integrated modeling with comprehensive correlation to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above objectives, the present application provides the following technical solutions: a method for multi-user feature analysis and numerical control integrated modeling based on comprehensive correlation, the modeling method comprising the following steps:

[0007] Control model correction stage, processing raw data input by multiple users, identifying and correcting missing and abnormal data, and using control models to correct data;

[0008] In the multi-user feature correlation analysis phase, a correlation analysis of multiple users with multiple parameters is conducted according to the season, and the ones with strong correlation are selected as the input features of the data-driven model;

[0009] In the data-driven model induction phase, the R-Vine Copula method is used to generate and reduce high-dimensional random variable sets, retain correlation characteristics, and form multi-user output characteristic curves;

[0010] Multiple users include wind power generation equipment, photovoltaic power generation equipment, energy storage equipment, electrical load, thermal load and cooling load.

[0011] Furthermore, the control model correction link includes bad data identification and bad data correction processes.

[0012] Furthermore, in the bad data identification, the bad data is identified by comparing the change rate of multiple users at each moment relative to the previous moment, and the specific steps are as follows:

[0013] Step 1: Preprocess multi-user data to form an m×n dimensional data set:

[0014]

[0015] Step 2: Identify missing values ​​in multi-user data and directly mark them as bad data;

[0016] Step 3: Calculate the mean of each row of the multi-user dataset and standard deviation

[0017] Step 4: Calculate the rate of change of the data sample at each moment relative to the previous moment

[0018] Step 5: If the rate of change of the data at the i-th moment in the data set is μ i -3·σ i <e<μ i +3·σ i , then it is normal data; otherwise it is identified as bad data.

[0019] Furthermore, the bad data correction process includes the following steps:

[0020] S1. Identify user categories and take into account various conditions and factors that may cause bad data, including but not limited to temperature, humidity, time and other physical factors;

[0021] S2. Substitute the specific conditions and factors for the occurrence of bad data into the physical model to simulate the change pattern of normal load data under different combinations of conditions;

[0022] S3. Use the calculated normal data to replace the original bad data, so as to achieve accurate correction of the bad data.

[0023] Furthermore, correlation analysis is performed on the corrected multi-user data, including correlation analysis between wind-solar, wind-storage, wind-electricity, wind-heat, wind-gas and solar-electricity, solar-storage, solar-heat, solar-gas and electricity-heat, electricity-storage, electricity-gas and gas-storage, gas-heat and heat-storage.

[0024] Furthermore, in the multi-user feature correlation analysis link, in order to avoid the one-sided and limited results of a single correlation analysis method when seeking correlation among many users, a comprehensive correlation evaluation index I is used. c , including Spearman's rank correlation coefficient ρ, Pearson's correlation coefficient R and maximum information coefficient I.

[0025] Furthermore, the multi-coefficient comprehensive correlation evaluation index is constructed based on principal component analysis, and the specific steps are as follows:

[0026] A1. Collect the Spearman rank correlation coefficient, Pearson correlation coefficient and maximum information coefficient data between variables and construct the sample matrix X.

[0027]

[0028] n is the sample size, r is ρi 、r Ri 、r Ii are the Spearman rank correlation coefficient, Pearson correlation coefficient and maximum information coefficient of the i-th sample respectively;

[0029] A2. To eliminate the influence of coefficient dimension and magnitude differences, the matrix X is standardized using the z-score method. After standardization, the matrix X * element With s j is the mean and standard deviation of the jth column;

[0030] A3. Calculate the normalized matrix X * Covariance matrix Cov(X * ),

[0031] The formula is: The elements of this covariance matrix reflect the closeness of the relationship between the standardized coefficients. Its diagonal elements are the variances of each coefficient, which measure the fluctuation of the coefficient value; the off-diagonal elements reflect the degree of coordinated change between the coefficients;

[0032] A4. Solve the covariance matrix Cov(X * ) Eigenvalues ​​λ1, λ2, λ3 and corresponding eigenvectors e1, e2, e3. The size of the eigenvalue determines the variance contribution of the corresponding principal component, that is, the degree of data dispersion in the direction of the principal component. The principal component with a large eigenvalue retains more original coefficient information.

[0033] A5. Select the principal component based on the eigenvalue and cumulative contribution rate. The principal component was determined when CR ≥ 85%;

[0034] A6. After determining the principal component, construct comprehensive related evaluation index I c , suppose K principal components are selected, and their score matrix is:

[0035] Principal component weight vector Normalized by the corresponding eigenvalue but

[0036] Furthermore, the data-driven model is an R-Vine-Copula correlation analysis model, and the relationship quantity with strong correlation after correction is selected as the input data of the data-driven model;

[0037] The specific sampling steps for scene generation based on R-Vine Copula are as follows:

[0038] C1. The corrected wind power sequence, photovoltaic sequence, energy storage output sequence, electric load sequence, thermal load sequence and cooling load sequence are subjected to non-parametric kernel density estimation to obtain the probability density function f(x) and empirical distribution F(x), and the joint distribution C is obtained. k (F X (X));

[0039] C2. Calculate the Kendall rank correlation coefficient for the opposite side e=x(e) As the weight of the edge in the R-Vine structure, according to the weight of each edge Optimize the topological structure of the vine to achieve the best representation effect on the correlation between high-dimensional random variables;

[0040] C3, select the function family for each binary Copula function of each edge in the R-Vine-Copula tree, and estimate the parameters of the Copula function using maximum likelihood estimation to determine the high-dimensional joint distribution function;

[0041] C4, generate a high-dimensional multi-user output scenario set by sampling the high-dimensional joint distribution;

[0042] C5. Use ISDATA to perform scene reduction to determine the multi-user characteristics in each typical scene.

[0043] Technical effects and advantages of this application:

[0044] Compared with the existing technology, the control model correction link of the present application can effectively identify and correct missing and abnormal values ​​in multi-user data. By accurately calculating the mean, standard deviation and rate of change, etc., and comprehensively considering the data characteristics, it can not only accurately mark bad data, but also make corrections in combination with the physical model laws, significantly improving data accuracy, laying a solid foundation for subsequent analysis and modeling, and overcoming the problems of incomplete and inaccurate processing in the existing technology.

[0045] This application adopts a comprehensive correlation evaluation index, which integrates the Spearman rank correlation coefficient, Pearson correlation coefficient and maximum information coefficient based on principal component analysis to comprehensively measure the complex correlations between multiple users. It can avoid the one-sidedness of a single method, scientifically select the principal components according to the eigenvalues ​​and cumulative contribution rates to determine the weights, and construct a comprehensive indicator that more accurately reflects the data relationship. Compared with the existing technology, it is more scientific and comprehensive in correlation analysis, and provides strong support for in-depth understanding of multi-user characteristics.

[0046] This application uses the R-Vine Copula method for data-driven modeling. Through kernel density estimation, rank correlation coefficient calculation, function family selection and parameter estimation, it generates and reasonably reduces the set of high-dimensional random variables, reduces the data dimension while retaining the correlation characteristics, and improves modeling efficiency and accuracy. The resulting multi-user output characteristic curve can better simulate the system response and overcome the shortcomings of traditional modeling methods in processing high-dimensional data and correlation.

[0047] This application proposes a CNC integrated modeling method that combines the advantages of physical models and data-driven models. It first corrects the data through mechanism analysis to reduce the requirements for the number and quality of data samples, and then uses data-driven modeling to make the model more adaptable and predictive. It can more accurately reflect the characteristics of multiple users, effectively respond to the uncertainty of the integrated energy system, and optimize energy management and decision-making. Compared with the existing single model, it has a significant improvement in performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the structure of the multi-user feature analysis and numerical control integration model;

[0049] Figure 2 Schematic diagram for control model data correction. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0051] Example

[0052] As attached Figure 1 and Figure 2 A method for multi-user feature analysis and integrated numerical control modeling with comprehensive correlation is shown, including a control model correction link, a multi-user feature correlation analysis link and a data-driven model induction link.

[0053] Control model correction stage, processing raw data input by multiple users, identifying and correcting missing and abnormal data, and using control models to correct data;

[0054] In the multi-user feature correlation analysis phase, a correlation analysis of multiple users with multiple parameters is conducted according to the season, and the ones with strong correlation are selected as the input features of the data-driven model;

[0055] In the data-driven model induction phase, the R-Vine Copula method is used to generate and reduce high-dimensional random variable sets, retain correlation characteristics, and form multi-user output characteristic curves;

[0056] Multiple users include wind power generation equipment, photovoltaic power generation equipment, energy storage equipment, electrical load, thermal load and cooling load.

[0057] The control model correction process includes bad data identification and bad data correction process.

[0058] In bad data identification, bad data is identified by comparing the change rate of multiple users at each moment relative to the previous moment. The specific steps are as follows:

[0059] Step 1: Preprocess multi-user data to form an m×n dimensional data set:

[0060]

[0061] Step 2: Identify missing values ​​in multi-user data and directly mark them as bad data;

[0062] Step 3: Calculate the mean of each row of the multi-user dataset and standard deviation

[0063] Step 4: Calculate the rate of change of the data sample at each moment relative to the previous moment

[0064] Step 5: If the rate of change of the data at the i-th moment in the data set is μ i -3·σ i <e<μ i +3·σ i , then it is normal data; otherwise it is identified as bad data.

[0065] The bad data correction process includes the following steps:

[0066] S1. Identify user categories and take into account various conditions and factors that may cause bad data, including but not limited to temperature, humidity, time and other physical factors;

[0067] S2. Substitute the specific conditions and factors for the occurrence of bad data into the physical model to simulate the change pattern of normal load data under different combinations of conditions;

[0068] S3. Use the calculated normal data to replace the original bad data, so as to achieve accurate correction of the bad data.

[0069] It can effectively identify and correct missing and outliers in multi-user data. By accurately calculating the mean, standard deviation, and rate of change, and comprehensively considering data characteristics, it can not only accurately mark bad data, but also correct it in combination with physical model laws, significantly improving data accuracy, laying a solid foundation for subsequent analysis and modeling, and overcoming the problems of incomplete and inaccurate processing of existing technologies.

[0070] Correlation analysis is performed on the corrected multi-user data, including correlation analysis between wind-solar, wind-storage, wind-electricity, wind-heat, wind-gas and solar-electricity, solar-storage, solar-heat, solar-gas and electricity-heat, electricity-storage, electricity-gas and gas-storage, gas-heat and heat-storage.

[0071] In the multi-user feature correlation analysis phase, in order to avoid the one-sided and limited results of a single correlation analysis method when seeking correlation among many users, a comprehensive correlation evaluation index I is used. c , including Spearman's rank correlation coefficient ρ, Pearson's correlation coefficient R and maximum information coefficient I.

[0072] Based on principal component analysis, a multi-coefficient comprehensive correlation evaluation index is constructed. The specific steps are as follows:

[0073] A1. Collect the Spearman rank correlation coefficient, Pearson correlation coefficient and maximum information coefficient data between variables and construct the sample matrix X.

[0074]

[0075] n is the sample size, r is ρi 、r Ri 、r Ii are the Spearman rank correlation coefficient, Pearson correlation coefficient and maximum information coefficient of the i-th sample respectively;

[0076] A2. To eliminate the influence of coefficient dimension and magnitude differences, the matrix X is standardized using the z-score method. After standardization, the matrix X * element With s j is the mean and standard deviation of the jth column;

[0077] A3. Calculate the normalized matrix X * Covariance matrix Cov(X * ),

[0078] The formula is: The elements of this covariance matrix reflect the closeness of the relationship between the standardized coefficients. Its diagonal elements are the variances of each coefficient, which measure the fluctuation of the coefficient value; the off-diagonal elements reflect the degree of coordinated change between the coefficients;

[0079] A4. Solve the covariance matrix Cov(X * ) Eigenvalues ​​λ1, λ2, λ3 and corresponding eigenvectors e1, e2, e3. The size of the eigenvalue determines the variance contribution of the corresponding principal component, that is, the degree of data dispersion in the direction of the principal component. The principal component with a large eigenvalue retains more original coefficient information.

[0080] A5. Select the principal component based on the eigenvalue and cumulative contribution rate. The principal component was determined when CR ≥ 85%;

[0081] A6. After determining the principal component, construct comprehensive related evaluation index I c , suppose K principal components are selected, and their score matrix is:

[0082] Principal component weight vector Normalized by the corresponding eigenvalue but

[0083] Based on principal component analysis, the Spearman rank correlation coefficient, Pearson correlation coefficient and maximum information coefficient are integrated to comprehensively measure the complex correlation between multiple users. It can avoid the one-sidedness of a single method, scientifically select the principal components to determine the weights based on the eigenvalues ​​and cumulative contribution rates, and construct a comprehensive index that more accurately reflects the data relationship. Compared with existing technologies, it is more scientific and comprehensive in correlation analysis, providing strong support for in-depth understanding of multi-user characteristics.

[0084] The data-driven model is the R-Vine-Copula correlation analysis model, and the relationship quantities with strong correlation after correction are selected as the input data of the data-driven model;

[0085] The specific sampling steps for scene generation based on R-Vine Copula are as follows:

[0086] C1. The corrected wind power sequence, photovoltaic sequence, energy storage output sequence, electric load sequence, thermal load sequence and cooling load sequence are subjected to non-parametric kernel density estimation to obtain the probability density function f(x) and empirical distribution F(x), and the joint distribution C is obtained. k (F X (X));

[0087] C2. Calculate the Kendall rank correlation coefficient for the opposite side e=x(e) As the weight of the edge in the R-Vine structure, according to the weight of each edge Optimize the topological structure of the vine to achieve the best representation effect on the correlation between high-dimensional random variables;

[0088] C3, select the function family for each binary Copula function of each edge in the R-Vine-Copula tree, and estimate the parameters of the Copula function using maximum likelihood estimation to determine the high-dimensional joint distribution function;

[0089] C4, by sampling the high-dimensional joint distribution, a high-dimensional multi-user output scenario set is generated;

[0090] C5. Use ISDATA to perform scene reduction to determine the multi-user characteristics in each typical scene.

[0091] The R-Vine Copula method is used for data-driven modeling. Through kernel density estimation, rank correlation coefficient calculation, function family selection and parameter estimation, a high-dimensional random variable set is generated and reasonably reduced. The data dimension is reduced while retaining the correlation characteristics, and the modeling efficiency and accuracy are improved. The obtained multi-user output characteristic curve can better simulate the system response and overcome the shortcomings of traditional modeling methods in dealing with high-dimensional data and correlation.

[0092] The implementation principle of the embodiment of the present application is: first, the original data input by multiple users is processed through the control model correction link, missing and abnormal data are identified and corrected, and the control model is used to correct the data; then, in the multi-user feature correlation analysis link, a multi-parameter comprehensive correlation analysis is performed on multiple users according to the season, and the ones with strong correlation are selected as the input features of the data-driven model; then in the data-driven model induction link, the R-Vine Copula method is used to build a model, generate and reduce a high-dimensional random variable set, retain the correlation characteristics, and form a multi-user output characteristic curve. The above is the working principle of this comprehensive correlation multi-user feature analysis and CNC integrated modeling method.

Claims

1. A method for multi-user feature analysis and numerical control integrated modeling based on comprehensive correlation, characterized in that: The modeling method It includes the following steps: Control model correction link, Process raw data input from multiple users, identify and correct missing and abnormal data, and use control models to correct data; Multi-user feature correlation analysis phase, Conduct a correlation analysis of multiple parameters for multiple users according to the season, and select those with strong correlation as input features for the data-driven model; Data-driven model induction phase, The R-Vine Copula method is used to generate and reduce high-dimensional random variable sets, retain correlation characteristics, and form multi-user output characteristic curves; Multiple users include wind power generation equipment, photovoltaic power generation equipment, energy storage equipment, electrical load, thermal load and cooling load.

2. The method for multi-user feature analysis and integrated numerical control modeling based on comprehensive correlation according to claim 1, characterized in that: The control model correction link includes bad data identification and bad data correction processes.

3. The method for multi-user feature analysis and integrated numerical control modeling based on comprehensive correlation according to claim 2, characterized in that: In the bad data identification, bad data is identified by comparing the change rate of multiple users at each moment relative to the previous moment. The specific steps are as follows: Step 1: Preprocess multi-user data to form an m×n dimensional data set: Step 2: Identify missing values ​​in multi-user data and directly mark them as bad data; Step 3: Calculate the mean of each row of the multi-user dataset and standard deviation Step 4: Calculate the rate of change of the data sample at each moment relative to the previous moment Step 5: If the rate of change of the data at the i-th moment in the data set is μ i -3·σ i <e<μ i +3·σ i , then it is normal data; otherwise it is identified as bad data.

4. The method for multi-user feature analysis and integrated numerical control modeling based on comprehensive correlation according to claim 3 is characterized in that: The bad data correction process includes the following steps: S1. Identify user categories and take into account various conditions and factors that may cause bad data, including but not limited to temperature, humidity, time and other physical factors; S2. Substitute the specific conditions and factors for the occurrence of bad data into the physical model to simulate the change pattern of normal load data under different combinations of conditions; S3. Use the calculated normal data to replace the original bad data, so as to achieve accurate correction of the bad data.

5. The method for multi-user feature analysis and integrated numerical control modeling based on comprehensive correlation according to claim 1, characterized in that: Correlation analysis is performed on the corrected multi-user data, including correlation analysis between wind-solar, wind-storage, wind-electricity, wind-heat, wind-gas and solar-electricity, solar-storage, solar-heat, solar-gas and electricity-heat, electricity-storage, electricity-gas and gas-storage, gas-heat and heat-storage.

6. The method for multi-user feature analysis and integrated numerical control modeling based on comprehensive correlation according to claim 1, characterized in that: In the multi-user feature correlation analysis link, in order to avoid the one-sided and limited results of a single correlation analysis method when seeking correlation among many users, a comprehensive correlation evaluation index I is used. c , including Spearman's rank correlation coefficient ρ, Pearson's correlation coefficient R and maximum information coefficient I.

7. The method for multi-user feature analysis and integrated numerical control modeling based on comprehensive correlation according to claim 6, characterized in that: The specific steps of constructing a multi-coefficient comprehensive correlation evaluation index based on principal component analysis are as follows: A1. Collect the Spearman rank correlation coefficient, Pearson correlation coefficient and maximum information coefficient data between variables and construct the sample matrix X. n is the sample size, r is ρi 、r Ri 、r Ii are the Spearman rank correlation coefficient, Pearson correlation coefficient and maximum information coefficient of the i-th sample respectively; A2. To eliminate the influence of coefficient dimension and magnitude differences, the matrix X is standardized using the z-score method. After standardization, the matrix X * element With s j is the mean and standard deviation of the jth column; A3. Calculate the normalized matrix X * Covariance matrix Cov(X * ), The formula is: The elements of this covariance matrix reflect the closeness of the relationship between the standardized coefficients. Its diagonal elements are the variances of each coefficient, which measure the fluctuation of the coefficient value; the off-diagonal elements reflect the degree of coordinated change between the coefficients; A4. Solve the covariance matrix Cov(X * ) Eigenvalues ​​λ1, λ2, λ3 and corresponding eigenvectors e1, e2, e3. The size of the eigenvalue determines the variance contribution of the corresponding principal component, that is, the degree of data dispersion in the direction of the principal component. The principal component with a large eigenvalue retains more original coefficient information. A5. Select the principal component based on the eigenvalue and cumulative contribution rate. The principal component was determined when CR ≥ 85%; A6. After determining the principal component, construct comprehensive related evaluation index I c , suppose K principal components are selected, and their score matrix is: Principal component weight vector Normalized by the corresponding eigenvalue but 8. The method for multi-user feature analysis and integrated numerical control modeling based on comprehensive correlation according to claim 1, characterized in that: The data-driven model is an R-Vine-Copula correlation analysis model, and the relationship quantity with strong correlation after correction is selected as the input data of the data-driven model; The specific sampling steps for scene generation based on R-Vine Copula are as follows: C1. The corrected wind power sequence, photovoltaic sequence, energy storage output sequence, electric load sequence, thermal load sequence and cooling load sequence are subjected to non-parametric kernel density estimation to obtain the probability density function f(x) and empirical distribution F(x), and the joint distribution C is obtained. k (F X (X)); C2. Calculate the Kendall rank correlation coefficient for the opposite side e=x(e) As the weight of the edge in the R-Vine structure, according to the weight of each edge Optimize the topological structure of the vine to achieve the best representation effect on the correlation between high-dimensional random variables; C3, select the function family for each binary Copula function of each edge in the R-Vine-Copula tree, and estimate the parameters of the Copula function using maximum likelihood estimation to determine the high-dimensional joint distribution function; C4, by sampling the high-dimensional joint distribution, a high-dimensional multi-user output scenario set is generated; C5. Use ISDATA to perform scene reduction to determine the multi-user characteristics in each typical scene.

Images