Architecture And Technical Realization Of Operation Analysis System Based On Multidimensional Data Of Power Plant


 Flexible analysis and use of multi-source data from power plants to accurately predict system failures and scan potential risks are of great significance for achieving efficient and accurate system operation analysis and decision-making. The article introduces the overall architecture and function design of a system operation analysis system based on massive data and demonstrates the functions and operation analysis results of each module of the system through application examples. The system integrates system data such as plant load information, equipment layout, (automation) electrical equipment control, signal and measurement systems, and relay protection configuration, and uses improved machine learning algorithms and weak point identification methods to expand the correlation analysis of target data. The function of fault risk level prediction and weak point identification is conducive to the power plant operation and maintenance management department to propose corresponding technologies and management methods for operation, maintenance, and repair work, which improves the scientificity and practicability of the existing analysis system. It has laid the foundation for the realization of informatization, intelligence, and lean operation analysis.


Introduction
With the continuous development of smart power plant construction and the continuous improvement of load information, equipment layout, (Automation) electrical equipment control, signal and measurement system, relay protection con guration and other system data, how to exibly apply the big data method in the information eld to big data analysis, realize the accurate prediction of the fault and potential risk scanning, and nd out the weak links to improve, has become an urgent problem to be solved [1][2][3][4] . The traditional information data quality, data types ,and data integrity are poor, so it is necessary to investigate and analyze the information system and Preprocess Based on the existing and reliable data information. Considering the rich information of multi-source heterogeneous massive data, it is necessary to mine features with a strong ability of fault risk prediction from massive data and numerous fault features through feature extraction and related line analysis. The weak characteristics and highfrequency transient characteristics of the fault make the fault diagnosis and processing more di cult. Therefore, it is urgent to study the intelligent fault diagnosis and prediction technology applicable to it 5 ; 6 . There are some problems in operation and maintenance, such as heavy maintenance workload, scattered equipment ,and insu cient professional level of operation and maintenance personnel. To effectively improve the operation and maintenance analysis and decision-making level, enhance the system controllability, and shorten the fault recovery time, computer operation simulation aided decisionmaking system has become an indispensable tool for modern power plant planning and construction 7 .
At present, PSASP 8 ; 9 , NETOMAC 10 , PSS / E 11 , ETAP 12 , psapac 13 , BPA and DIgSILENT 14 are used for operation and maintenance control at home and abroad. Although these software provide a useful tool for simulation calculation and analysis, they are not directly aimed at operation and maintenance, and lack of operation data correlation analysis, fault risk prediction, weak point identi cation ,and other auxiliary decision-making functions. In this paper, an operational analysis system is designed, which integrates data correlation analysis, fault risk level prediction and weak link scanning and identi cation functions. In addition to completing the data collection and preprocessing of the traditional analysis system, the improved feature selection algorithm is used to realize the correlation analysis function of multi-source heterogeneous data; the multi-dimensional and uncertain fault in uencing factors such as region, equipment status, time ,and working condition are comprehensively considered, and the fault risk level prediction function based on improved decision tree algorithm and the complex network is extended It can effectively reduce the probability of failure and risk occurrence, which is of great signi cance for e cient and accurate operation analysis.

Methods
The operation analysis system relies on the existing business system of the power plant for advanced application development, supporting the e cient and accurate operation and maintenance of the system.

A. Overall structure
The overall framework design of the system is shown in Figure 1.
1) Data layer. Based on the power plant common data model standard, the whole plant monitoring system (SIS) 15 , the power plant electrical monitoring management system (ECMs) 16 are used to access the equipment asset operation and maintenance management system and the load electricity information acquisition system. Through data integration and cleaning, according to the actual requirements of operation analysis, the fault data, operation data, working condition factors ,and other data are deeply integrated to realize the serial collection of the multi-service data systems. This layer is composed of a graphic library and attribute database, which is stored in access/oracle. It provides basic general data service and database access ability for the platform layer to call.
2) The platform layer. Through the common data/service interface, the platform layer uses web service 17 and extract transform load (ETL) 18 to access external system data from the data layer; This layer is composed of operation framework construction and development platform modeling, including open source components of the work ow, transaction processing, security system ,and other platforms. The application platform running framework is composed of an operation basic framework, interface display framework, icon display framework,and application integration framework. The development platform modeling tool consists of business modeling, report de nition, permission con guration, etc.
3) Application layer. The bottom layer uses Java, C, C + + and other programming languages, and based on echards chart library 19 , carries out front-end data visualization design, which is divided into data management, correlation analysis, fault risk prediction, weak point identi cation, system con guration and report analysis modules, providing charts, report reports,and other displays to realize multi-level user interaction.

B. Functional architecture
The core functions of the operation analysis system are realized by six modules in the application layer.
Compared with the traditional analysis system, the system extends the functions of data correlation analysis, fault risk prediction,and weak point identi cation, and generates reports and visualizes the analysis and identi cation results.The system functional architecture is shown in Figure 2.
1) Data management module mainly realizes data integration, data query,and data processing.
2) In the correlation analysis module, the advanced algorithm is used to carry out the correlation analysis of multi-source heterogeneous data to realize the selection of optimal fault features and the screening of weak point identi cation indicators.
3) According to the fault data, operation data,and working condition factors, the fault risk prediction module carry out the fault setting; analyzes the cause of power failure, the scope of power failure, and the level of power failure; calculates the fault levels of various faults in regions, stations and feeders; analyzes the fault level, fault location, transfer,and optimization of the results; It realizes the functions of fault prediction, emergency repair decision-making and so on.
4) The weak point identi cation module comprehensively considers the network topology and operation characteristics to realize the analysis, scanning,and risk early warning of the weak link of auxiliary power. 5) The system con guration module mainly completes the automatic risk analysis calculation setting and risk analysis evaluation parameter setting.
6) The report analysis module mainly completes the correlation analysis, fault risk prediction,and weak point identi cation results query, generation,and editing, and then gives the transfer, optimization and equipmeIII.

A. Data integration technology
The ow chart of data integration between an auxiliary power operation analysis system and part of auxiliary power information system 20 is shown in Figure 3.
1)Integrated design with application data of electrical primary and secondary equipment. The data integration of auxiliary power operation analysis system and electrical primary and secondary equipment system mainly includes standard code, le information of metering point, meter, low-voltage equipment, distribution transformer, etc., as well as the relationship between auxiliary power equipment, etc. Analyze the transmission data volume, transmission frequency,and other elements required for data integration with electrical primary and secondary equipment, and adopt the technical route of basic data platform of electrical primary and secondary equipment. The basic data platform is replicated by Ogg (Oracle Golden Gate) 21. Fig. 2 ow chart of system function realization application data of electrical primary and secondary equipment, auxiliary power operation analysis system obtains data from marketing basic data platform of a data center. When users, metering points, meter attributes, metering boxes are added, changed,or deleted, the basic data platform passes through Ogg as required The auxiliary power operation analysis system uses Java database connectivity (JDBC) mode to call and update data in realtime.
2)Data integration design with load information acquisition system. The acquisition system writes the power data and the bottom indication of electric energy meter into the platform through the standard interface of massive data, deploys the calculation service module of auxiliary power operation analysis system, and extracts the bottom indication of electric energy meter by the standard interface of the massive data platform.
3)And SIS data integration design. SIS pushes the data information of power transmission and distribution to the data center, and the auxiliary power operation analysis system extracts the equipment account information and network topology information required by the analysis system through ETL. 4)Data integration design with the plant automation system. The operation analysis system integrates the auxiliary power equipment and topology information, and the dispatch and control center generates the CIM format le 22 , and the data center is responsible for data analysis. 5)Data integration design with the meteorological information system. The meteorological information system writes the temperature, humidity, wind speed and other weather data through the interface of the massive data platform.

B. Correlation analysis technology
Considering that many factors affecting the failure of auxiliary power, there are many redundant and irrelevant fault features. Multi-source data integrated by data management module is adopted Combined data of machinery, working condition, equipment,and electrical system, equipment location data, distribution transformer capacity, real-time load data, monthly maximum load data and outage time, outage times, etc., determine the fault factor sample set and conduct fault feature screening. The traditional ltering feature selection algorithm relief (relevant features) assigns different weights to features according to the relevance of each feature and category and sets the weight threshold to select features. ReliefF algorithm extends the ability of relief algorithm to deal with multi-classi cation problems 23 . The correlation analysis technology proposed in this paper, after the feature selection based on ReliefF algorithm is completed, the correlation analysis of auxiliary power fault characteristics is carried out, and the redundant operation of fault characteristics is realized by improving the grey correlation analysis method, and nally,the optimal fault feature set is output (hereinafter referred to as grelief algorithm). The speci c process of correlation analysis of fault risk characteristics is shown in 1)The main fault characteristic matrix of auxiliary power is constructed, and then any feature is selected as "reference sequence", and the remaining feature is taken as "comparison sequence".
2)The correlation coe cient ξ (k) between the comparison sequence and the reference sequence at k time is solved respectively.

3) The correlation degree -ξ (k) between fault features is solved and the correlation matrix is established.
4) The correlation threshold is set to output the optimal fault feature set.

C. Fault risk prediction technology
The typical data prediction algorithms in machine learning theory include neural network algorithm, support vector machine (SVM), decision tree algorithm and lifting algorithm. The AdaBoost algorithm mentioned in this paper is a typical lifting algorithm, which can predict the risk level by upgrading the "weak classi cation algorithm" algorithm for the same training set. The algorithm ow is shown in the red dotted box in Figure 5. The weak classi er is trained by decision tree algorithm. Through the deviation of the weak classi er, the weight of the sample is updated, and then the strong classi er is obtained through repeated iterations. Finally, the fault prediction effect is improved. In the classi cation of fault risk level, considering that the scale of different power supply areas is very different, three kinds of fault risk indicators are selected, including the frequency of weekly fault outage, the proportion of outage duration and the proportion of lack of power supply. The outage risk degree of auxiliary power failure is divided into three levels: general, moderate and severe. The optimal feature set for fault risk level prediction is screened out by correlation analysis technology in Section 2.2, and the input sample set required in the prediction stage is effectively determined on the basis of this set. Firstly, the auxiliary power fault sample set is determined and imported, and the sample data is de dimensioned after "normalization"; the fault sample data is split randomly according to the proportion of 7:3, 70% of the data is used for training model, and 30% of the data is used for prediction; back propagation (BP) neural network, SVM algorithm and ada-dt are selected in machine learning The algorithm is used to predict the auxiliary power fault respectively, and the algorithm with the best effect is determined to train the fault risk prediction model by comparing the prediction accuracy.

D. Weak point identi cation technology
Considering the safety, economy and network topology of auxiliary power. In this paper, the network elastic performance change risk, branch load rate change risk, node voltage deviation risk and branch line loss rate change risk are introduced as key node identi cation indicators. Combined with subjective experience and objective data, the comprehensive evaluation method is used to analyze the index weight, and a multi-level and multi angle comprehensive evaluation model of auxiliary power key nodes is constructed. The ow chart of identi cation method of auxiliary power book vulnerability is shown in Figure A1 .The main steps are as follows: 1)Rread the topology data and operation state data of the auxiliary power network, calculate the power ow of the initial stable state of the auxiliary power system, and conduct node n-1 Fault simulation test; 2) Calculate the index values of network elastic change value, node voltage deviation risk, branch load change risk and line loss change risk in the initial stable state of auxiliary power;  Prediction of failure risk level.The failure risk prediction module uses neural network algorithm, SVM algorithm, and Ada-DT algorithm to predict the risk level. The accuracy of the prediction is shown in Table  II . Finally, the ADA DT algorithm with high index values is selected to train the fault risk level prediction model.The prediction results are shown in TABLE ."Level 1" in table represents the general failure risk of auxiliary power; "level 2" represents the moderate failure risk of auxiliary power; "level 3" represents the severe failure risk of auxiliary power; "comprehensive" is the overall prediction accuracy of three types of samples. It can be seen from  2) Identi cation of weak points of auxiliary power. The topology of power plant A is shown inTABLE . Since the nodes numbered 1 and 2 in the actual auxiliary power system are located at key positions, that is, the power node, only weak point identi cation veri cation except for nodes 1 and 2 is performed. The results of the weak point identi cation of auxiliary power in this area are shown in  A1 stands for network elasticity index A2 represents the risk of node voltage A3 represents the risk of branch road change A4 represents the risk of branch road change

B. System application effectiveness
The auxiliary power operation analysis system has been applied in a power plant. By analyzing the massive operation data of auxiliary power, the main in uencing factors of the fault are determined, and the fault risk level is predicted for the auxiliary power of the power plant. It provides technical support for the scienti c development of emergency repair plan, identi es and manages more than 500 weak links of auxiliary power in advance, and avoids direct economic loss. The identi cation and analysis of system weak points is applied to the auxiliary power of power plant A. through the analysis of the selected auxiliary power system, the names of the identi ed weak equipment are sorted and visualized to assist the staff to formulate the maintenance plan.

Conclusion
The auxiliary power operation analysis system and method based on big data, machine learning and complex network theory, adopts advanced computer technology and auxiliary power operation simulation technology, and constructs the auxiliary power operation analysis system which integrates data management, data correlation analysis, fault risk prediction, weak point identi cation, report analysis and other functions, so as to accurately predict the auxiliary power failure risk and The foundation of rapid identi cation of weak points can provide technical support for the establishment of auxiliary power operation analysis decision-making scheme. At present, the research and application of the operation analysis system based on the large amount of auxiliary power data is still in the initial implementation stage, and further simulation veri cation and eld demonstration of the operation analysis method of auxiliary power at the big data level are needed.