| Smart grid dispatching and control systems(D5000 system)play an important role in ensuring the safety and reliability of power grid and avoiding unnecessary loss.However,nowadays the operation and maintenance staff don't have effective methods for real-time monitoring and prediction of the health status of the system.To meet the actual needs of the D5000 system,this paper conducts a research on its health evaluation and devises the health degree evaluation method for the system based on experts' experience and D5000's characteristics,such as multifarious business,complex business hierarchy,business being closely related to hardware,and its special way of working;In order to increase the objectivity of the evaluation results,the evaluation method is based on machine learning and solves the problems caused by the unbalanced distribution of the system sample data.The research results are of great significance to improve D5000's operation and management and ensure its safe and stable operation.The main works of this paper are as follows:(1)The establishment of the comprehensive evaluation index system of the D5000 system.Firstly,this paper analyzes the factors that affect the system's health degree,which are hardware foundation,business support and work mechanism;Secondly,in view of the thr-ee influencing factors,this paper devises the comprehensive health degree evaluation index system that applies to D5000's special working mechanism,consisting of the hardware health degree evaluation index system and the business health degree evaluation index system;Thirdly,this paper analyzes the characteristics of the evaluation index system,such as the coexistence of qualitative and quantitative indexes,the priority of indexes,business being closely related to hardware,and its special working mechanism,which lays a foundation for devising a comprehensive health degree evaluation method for the D5000 system.(2)The study of health degree evaluation method for the D5000 system based on experts' experience.A comprehensive health degree evaluation method based on the integration of business and hardware is proposed on the basis of considering the priority of evaluation indexes and combining qualitative analysis with quantitative analysis.Firstly,according to the priority of evaluation indexes,this paper formulates corresponding priority evaluation principles;Secondly,through deep study of the multi-nodes,no master architecture and master-slaves architecture of the D5000 system,this paper puts forward corresponding method to integrate business health degree and hardware health degree;Thirdly,this paper determines the weights of indexes and evaluation standards,combining the analytic hierarchy process with experts'experience;Fourthly,this paper adopts the weighted average method and the parallel system health degree method to accomplish the comprehensive evaluation of the D5000 system.Effectiveness of this method is verified through example analysis of the system's operation data.(3)The study of health degree evaluation method for the D5000 system based on machine learning.A D5000 system health degree evaluation model based on random forest is established according to the system's sample data.Firstly,given the complex business hierarchy of the system,this paper proposes a health degree evaluation model based on multi-layer cascade architecture after analyzing the similarities and differences of business evaluation indexes and the differences of correlation relation between adjacent index layers;Secondly,to solve the multi-classification problem caused by unbalanced sample distribution of the system,this paper establishes a single random forest sub-model for the training data row and column random sampling;Thirdly,considering the D5000 system's business operation mode of horizontal integration and vertical operation,this paper adopts the calculation method of serial and parallel combination in the evaluation process;Eventually,the D5000 system health degree evaluation model based on multi-layer cascade architecture with 17 random forest sub-models is constructed.Effectiveness of this model is verified through example analysis of the system's operation data. |
PDF Full Text Request