Research on standardized evaluation model of status detection data on power equipment based on AHP

The power equipment produces a large number of status detection data during daily operation and maintenance, and most of the data is unstructured. So it makes against to the efficient processing and the deep mining of data. This article illustrates an empirical evaluation model for the status detection data of power equipment, and the model is modified by entropy method, then sorts schemes by index weights. A new evaluation model is provided to the standardization and efficient utilization of data for status detection data of power equipment, which provides strong support and guarantee for status assessment and diagnosis analysis of power equipment.


Introduction
With the construction of the smart grid and the in-depth development of the state maintenance, it has become a normalization work that real-time and dynamic analysis carried out on the equipment by the power company. So the data of power during operation and maintenance is increased in exponential speed. However, most of those data comes from manufacturer or experience record, which causes a lot of sources of information and poor normalization. Therefore, the redundant data is not conducive to high-efficiency utilization and secondary analysis. [1,2] In order to improve the grid's control ability, strengthen the specialization and lean management of the power equipment professional, and reduce security risks, it is necessary to standardize the status detection data of power equipment, and improve the level of automatic analysis of the data, which is beneficial to find the abnormal in the massive status detection data and give the corresponding preventive measures.

The sources and sort of status detection data
There are a variety of sources in the status detection data of power equipment. Now the production management system(PMS) built by National Electric Net Ltd established a uniformly standard on data's process, access and control. PMS also provides information such as display, early warning, analysis, diagnosis, evaluation and prediction. [3] Because lower rate of PMS access, there are a lot of data which comes from production management system and condition monitoring system, such as 2 1234567890 ''"" SCADA, outgoing inspection report , spreadsheets or text documents of routine text and so on. Therefore, in order to realize the integration of the status data of the power equipment, it is necessary to solve the problem that the semi-structured data of the status detection data of the power equipment is transferred to the structure.

Classification by the period of data generation
According to the general regulations and requirements of the power industry, the information sources of power equipment's status data usually include information before operating, operation information, maintenance information, and other information. Some information sources are shown in the table.

Classification by the frequency of data generation
According to the frequency of data update, we can divide those information into three categories: static data, dynamic data and quasi dynamic data.

Classification by data characteristics
According to the characteristics, it can be divided into structured data, semi-structured data, unstructured data and other types. Structured Data is what can be divided into fixed elements and can be represented by one or more two-dimensional tables. Unstructured data is the data outside of the structured data, which structure is not fixed, and can be hold in a variety of forms. such as pictures, images, audio and video. In addition, semi-structured dada is between structured data completely and totally unstructured data. They are not filled with the same specification and require intermediate transformation process, such as HTML and XML.
All the information is generated in Table 3.

Establishing the standardized evaluation model
In this paper, four typical indicators, top oil temperature, signals of dissolved gas-in-oil, partial discharge signals and casing loss data, are selected to be the data quality assessment objects. Combined the AHP and entropy weight, then we will get the standardized evaluation model.

AHP
3.1.1. The determination the weight based on AHP. There are four steps of applying AHP to make a decision. [4] The first step is to stratify the complex problems and form a hierarchy substructure diagram consisting of the target layer, the criterion layer, the index layer, and the scheme layer is formed. The second step is to use the "1-9 scale" to form a judgment matrix. Then we can verify the consistency of the matrix. Finally, by means of layer by layer, the weight of the scheme layer can be obtained, and the most important scheme is the optimal solution to the problem.  (1) Build hierarchical structure Applying AHP to solve the problem, first of all, it is clear to analyze the problem of decision making, and make it rationalize and stratify. And it is generally composed of the target layer, the criterion layer and the measures layer. As is shown in Figure 2.
We hope to improve the utilization and quality through the assessment of power equipment detection data. Therefore, the goal can be set as "assess data quality scientifically and improve data quality level". And in this paper we choose completeness, correctness, timeliness, accuracy and effectiveness as five guidelines. And there are no sub-codes. Finally, based on those guidelines, we  Figure 2. Hierarchical structure (2) Construct a judgment matrix The key step is to structure the judgment matrix in the research question by using the AHP. In this paper, we will get a judgment matrix in the beginning. Then this matrix will be compared between every two factors by experts in status detection of power equipment and get the sort order by weightiness of all factors. The degree of importance will be assigned by the Numbers 1 to 9. [5]  It represents the two elements are of equal importance 3 It represents the former is more important than the latter 5 It represents the former is obviously more important than the latter 7 It represents the former is deeply more important than the latter 9 It represents the former is extremely more important than the latter 2/4/6/8 It represents the median of the above judgment reciprocal If the importance ratio of i to j is aij, then the ratio of j to i :aji=1/aij Consulting the broad experts, the judgment matrix for the transformer is shown in Table 5 If CR<0.1, it is considered that the judgment matrix satisfies the consistency test. And if CR>0.1, it is considered that the judgment matrix does not meet the requirement of consistency, which is necessary to revise. When the judgment matrix meets the requirement of consistency, the weight of each index is calculated. And the expectations of the indicators for the next calculation is determined.

Amend the weight by entropy weight method
The subjective factors of AHP have great influences on the final evaluating conclusion. The weight of evaluation index was determined by improved AHP and was modified by entropy method. The concept of entropy in information theory comes from the one in statistical thermodynamics. And statistics on the self-information of all symbols are averaged. So the entropy of information is also called the average amount of information. [7] The entropy in information is used to measure the uncertainty of random variables and to solve the measurement of information. The smaller the value of the entropy in information is, the greater the amount of information is, and the higher the weight is.
The correction step of entropy weight method is as follows. [8] The original data matrix is set up with m pending objects and n evaluation indexes.

 
To an evaluation index: