3.1

Electric distribution network data

Electric allocation and new energy data not only have different data types, but also have many data description dimensions. For equipment description, there are operation status timing data, basic file information, operation and maintenance records, work order working conditions, etc. Through the research of power allocation data resources taking into account new energy.

According to the above research results, the number of power allocation data resources is not only huge, but also has many business attributes and data description dimensions. Therefore, the data with a large number of dimensions must be dimension-ally reduced to describe the data resources with accurate data characteristics, laying a data foundation for subsequent data fusion and analysis.

The core of multi-source data fusion is to collect, filter, match and comprehensively apply data from different data sources. Analyzing the application scenario of multi-source data fusion in distribution network planning is the premise to improve the planning level by using multi-source data. On the basis of data demand analysis of distribution network planning algorithms, the following algorithm links are improved by multi-source data fusion.

As is shown in table 1, VARCHAR2 (n) represents the string with the maximum character length of n; NUMBER (P, S) means the maximum number of digits is P, and the exact S is after the decimal point; DATE is the default date type data; IMESTAMP is the default date time data..

Table 1.

Information of electric cable.

3.2

Electric distribution network data fusion method considering new energy

Principal component analysis is a data feature extraction method commonly used to extract data features. First, it is an analysis method to find the best feature vector from redundant data, and then convert the data from the original dimension space to a new feature space through a certain dimension transformation. For example, the original dimension space is three-dimensional (x, y, z), and x, y, z are the three bases of the original space, Use a new coordinate system (a, b, c) to represent the original data, then a, b, c are the new bases, forming a new feature space. In the new feature space, it is possible that the projection of all data on c is close to 0, that is, it can be ignored, and the data can be directly represented by (a, b). In this way, the data is reduced from three-dimensional (x, y, z) to two-dimensional (a, b). Therefore, using the principal component analysis method to extract and reduce the dimensions of power distribution data features will improve the data description efficiency. The feature extraction of power distribution data mainly includes the following steps:

①Zero average the original data. For the operation status description of a transformer equipment, including voltage, current, frequency, capacity, etc., the units and mathematical quantity levels of each parameter are inconsistent. The first step of feature extraction is to standardize the data of different units and quantity levels, integrate all kinds of data near the central store through centralization, and then further standardize the processing, so that all kinds of data have the same value range in each dimension. The original data is grouped into n rows and m columns of matrix X by column, and each row of X and the data representing an attribute field are averaged to zero, that is, the average value of this row is subtracted. For the method of centralization, see Formula 1.

In the above formula, x is the original sample data, μ Is the sample expectation. Through the above transformation, we get a new sample data set with the expectation of 0. For the standardized method, see Formula 2.

Divide by the standard deviation of the original sample data on the basis of centralization to obtain a new sample data set with an expectation of 0 and a standard deviation of 1. Compared with the original data set, the original data set has a large degree of dispersion. The data sets are not merged with the origin as the center. After centralization, the data set basically forms a cluster with the origin as the center. After standardization, The distribution range of the data set on the x and y dimensions basically falls within the [-2,2] interval.

4.1

Electric data resource classification

By sorting out the data classification rules of power distribution and distributed energy data in business themes, data indicators involved in each business are sorted out to form business theme data classification rules, and data resources are mapped to different business theme types according to business theme data classification rules.

Table 2.

Electric data resource classification based on electric service logic.

In addition to data classification based on the electric business, it is also necessary to classify data according to the basic characteristics of the data at the data level. In addition to considering the basic characteristics of the data, data classification at the data level can also be combined with manual data labels, by learning the manually labeled data labels, training in sample data to use depth learning to find the corresponding data characteristics, and then building a data classifier, Achieve intelligent classification of power distribution and distributed energy data at the data level. Data level classification is mainly based on manually labeled data category labels, through artificial intelligence machine learning algorithms, to learn the characteristics of sample data, and then solidify it into a data classifier, through which to achieve intelligent classification of data level.

As is shown in figure 2, the classification of electric distribution data considering new energy mainly includes category characteristics, deep machine learning, data sample, manual annotation, the classification results will be pushed to data classifier.

Figure 2.

Classification of electric distribution data including new energy

4.2

Intelligent matching of electric data resource

There is a strong correlation between data resources based on electric business. However, due to the certain independence between the data of each business system for power distribution and the difference in field definition and description, traditional key field matching methods are difficult to achieve automatic data matching. Therefore, it is proposed to carry out research on accurate identification methods for the business meaning of each business system data. Based on the data resources with known business meaning, a comparative sample data set is established. Through the similarity analysis between the unknown business meaning data and the sample data set, the business meaning of the unknown business meaning data is identified according to the analysis results, so as to achieve intelligent matching of data resources between the same business system and different data types. Similarity calculation is to calculate the similarity of two columns or two different data, and judge the similarity of the two data according to the similarity. This project uses the similarity calculation method of calculating Pearson correlation coefficient to achieve similarity judgment. First, standardize the data to be compared, unify the comparison data to the same dimension, and then calculate the Pearson correlation coefficient of unknown business identification data and known identification data, According to the calculation results, judge whether the Pearson correlation coefficient is within the specified confidence interval. If it is, it is considered that the similarity between the two is very high, and they can be considered as the same type of data, and the target data can be marked with the same data identifier, otherwise it is not. By means of similarity calculation, the service identification of unknown data can be identified, so as to achieve data matching.

Find the covariance, see Formula 3.

Calculate the covariance of n variables to obtain the covariance matrix, as shown in Formula 4.

A is a matrix of order n, if number λ And n-dimensional non-zero column vector x satisfies Ax= λ x. That number λ Eigenvalue called A, and x corresponding to Eigenvalue called A λ The eigenvector of. When the characteristic polynomial is equal to 0, it is called the characteristic equation of A, and the characteristic equation is a homogeneous linear equation group. The process of solving the eigenvalue is actually to solve the solution of the characteristic equation. For covariance matrix A, see Formula 5 for the calculation method of its eigenvalue (possibly multiple).

Figure 3.

Electric distribution network data fusion

The eigenvalues are sorted from large to small, and the largest k eigenvectors are selected. Then the corresponding k eigenvectors are used as column vectors to form the eigenvector matrix W (n). Calculate X, that is, project the data set X onto the selected feature vector, so that we can get the reduced dimension data set X we need. The eigenvectors are arranged into a matrix from top to bottom according to the size of the corresponding eigenvalues in rows, and the first k rows are taken to form the matrix P. Y=PX is the data reduced to k dimensions. Through feature extraction of power distribution data, solve the data features that can accurately describe power distribution data resources, and provide an accurate and effective data basis for data classification and matching.

Above all, at the business level, research and establish evaluation indicators from the logic, relevance, integration and other dimensions of the business. On this basis, research and build a multi-level, multi-directional and multi-dimensional distribution network topology data quality evaluation system to help understand the true level of distribution network topology data quality and support the deepening application and decision-making of distribution network.