Research on hierarchical differentiation intelligent examination method for operational health of low voltage distribution network

The operational health assessment of the Low-voltage Electricity District is the basic task of grid and refined management of low-voltage distribution network. Considering of the large number of stations and the lack of standardized management constraints, using the big data platform and multi-source heterogeneous data management technology, a kind of intelligent medical check-up method for the health of district station is proposed. Firstly, the self-organizing neural network clustering technology is used to classify the station area into 15 categories, which must take into account of geographical attributes, economic attributes and so on. Then, for each major type of area, the analytic hierarchy process is used to realize the subjective weight of the differentiated subject indicators. The entropy weight method is used to realize the objective weight of the three-level subdivision index, and the distance map method is used to realize the percentile mapping of the three-level subdivision index. Finally, the subjective and objective comprehensive weights are given to get the running health score of the station. Based on the actual platform data of a power supply unit in Shan Xi Province, the accuracy of the confusion matrix of the health classification of the station is about 92%, which verifies the effectiveness and applicability of the method.


Introduction
The low-voltage distribution network area is an important part of the power grid, and it has the closest connection with power users. With many factors such as the development of social economy, the increase of electricity load, the rapid increase of supply and sales of power in the low-voltage distribution network, the line loss and voltage qualification rate, the promotion of the assessment of the sub-area, the management requirements for the Low-voltage Electricity District increasingly urgent [1] [2] However, the number of low-voltage distribution network stations is large, and there are many decision-making problems in the management of Electricity District. At present, the data of the Low-voltage Electricity District is scattered among various departments. The type, format and content of the data are diverse and difficult to handle. On the other hand, there is no effective method and means to make a reasonable evaluation of the operating status of the Low-voltage Electricity District. As a result, it is difficult to judge the operational status of the station area, and difficult to provide a basis for the refined hierarchical management decision.
Therefore, how to standardize the data of many Low-voltage Electricity District and quickly judge the operational health of the station area is an urgent problem to be solved. In the evaluation method  [3] established an evaluation index system adapted to the characteristics of high-voltage distribution network planning, and proposed an analytic hierarchy process based on Delphi method for the weight of indicators at all levels; the entropy weight method is used to modify the index weights determined by the analytic hierarchy process, and the meta-element extension model is studied from four aspects: economic benefit, social benefit, technical benefit and environmental benefit in literature [4]. From the node vulnerability, the objective data information and the node value coefficient are comprehensively considered to obtain the utility risk entropy weight in the literature [5].In the research of evaluation index system, some literatures at home and abroad have proposed the smart grid evaluation index system [6] , but the index system proposed for the smart distribution network is still relatively small. For the evaluation of low-voltage distribution network, the literature [7], From the voltage characteristics of the low-voltage distribution network to establish an evaluation index system, based on the order relationship -open the grade weighting method to determine the index weight; literature [8][9][10][11]Based on the analytic hierarchy process, a comprehensive evaluation system for the new rural low-voltage distribution network was established. The 0-10 scale method was used to determine the comprehensive weight of the individual indicators.

Cluster description
Low-voltage distribution network is large and widely distributed in various regions. Each geographical environment of it is different, so it is necessary to cluster all the districts in Shanxi through Self-organizing Maps (SOM) into several major categories, and each subsequent is quantified by different analytic method index importance expert judgment matrix. The most important reason for optimizing SOM clustering is that the SOM has the ability to capture nonlinearly combined representations between features.
In the data feature selection of cluster analysis in district, 101 subdivision indicators of four dimensions of station user, station area geography, station area operation and station area management are considered. On the SOM cluster output, 15 rough classification types considering the economic and geographic attributes of the Low-voltage Electricity District are designed: economic attributes include operational, living and comprehensive, and geographical attributes include mountain villages, rural areas, townships, urban and rural areas, city.

SOM algorithm
The computational layer of the SOM clustering algorithm selects the 2-dimensional network form, because we expect to cluster the clusters into two-dimensional spatial representations, consider the economic and geographic attributes of the District region. For example, from the perspective of District's economic attributes, comprehensiveness is the intermediate transitional form of business and life; from the perspective of the geographical attributes of District, from mountain villages, rural areas, towns, towns and cities to cities, it is gradually adjacent. The essence of the SOM clustering algorithm is to find the calculation layer unit that is the shortest distance from the target unit in the learning process, that is, the winning unit, and update it. At the same time, the weights of the neighboring regions are updated so that the output node maintains the topological features of the input vector.  Table 1 shows the SOM clustering results of 107,179 stations in Shanxi Province. It can be seen that the city's comprehensive Low-voltage Electricity District accounts for a maximum of 18.91%, and the mountain village operating area accounts for at least 0.03%. For the 15 clustering clusters, the observation samples close to 1% were randomly selected, and the accuracy of the SOM clustering results was estimated to be about 92% by manual verification comparison.  mainly comes from the electricity information collection system, marketing business application system, fee control system, PMS system.

District Health Check-up Modeling
The process of the health check score model in district is shown in Figure 2.

TOPSIS method index score mapping
The technique for Order Preference by Similarity to An Ideal Solution (TOPSIS) determines by determining the relative distance between the index vector of each evaluated object and the positive and negative ideal solutions. If the object being evaluated is close to the positive ideal solution and away from the negative ideal solution, the evaluation effect is better; otherwise, it is worse. Some indicator scores will take into account the expert experience to make rule corrections.

Entropy weight method index objective weighting
Based on the entropy weight method, the information contribution size of the mapped index data, such as automatic meter reading ratio, zero user share, voltage pass rate, topological relationship automatic recognition rate, and acquisition coverage rate Objective empowerment.

AHP method theme indicator subjective weight
The Analytic Hierarchy Process (AHP) was used to calculate the subjective weight of the non-underlying indicators of the health of the station. In the actual data analysis process, based on the actual data availability and expert business judgment, the evaluation index system is simplified and then hierarchically summarized into the three-layer structure shown in Figure 3.   Table 2 shows the results of the subjective weight calculation of the AHP of the second-level indicators of the health status of the station (acquisition, fee control, line loss, and equipment dimensions).

Conclusion
Considering that there are nearly 107,200 Electricity District areas in Shanxi Province, and the factors such as the structure and personnel composition of the District network are complicated, this paper proposes a set of methods to quantitatively evaluate the operational health of the Low-voltage Electricity District. Firstly, through the SOM clustering technology, the station area is roughly divided into 15 categories considering geographical location and economic attributes. Then, for each type of station area, the customized analytic hierarchy process, entropy weight method and superior and inferior solution distance method are used to complete the evaluation index. The weightiness of the operating fitness is a percentage score. The innovation of this research is also to comprehensively utilize the multi-dimensional perspectives around the management of the Low-voltage Electricity District to design two sets of indicator systems. One set includes 4 items and 101 items of indicators for the SOM clustering data characteristics, and the other set includes 6 large class of 15 subjects and 40 third-level indicators are