Failure Rate Modeling: A Non-Parametric Data Mining Approach to MV Network Field Data

2009 IEEE Electrical Power and Energy Conference, EPEC 2009 01/2009; DOI: 10.1109/EPEC.2009.5420910


Power distribution fault statistics provides valuable knowledge of system failure rate behavior, as the start point of reliability evaluations. Using this statistics, electric utilities trace and develop their reliability plans based on fault statistics. This paper considers a data mining approach to model momentary failure rate in terms of the most influential factors. A methodology is presented here, for momentary fault cause identification, using a feature selection algorithm applied to MV network of the Greater Tehran Electric Distribution Company (GTEDC). Subsequently, two non-parametric failure rate models; classification and regression tree (CART) and artificial neural network (ANN) are utilized to cope with the high non-linearity of the problem space. Results and comparisons of the characteristics of the proposed methods illustrate the advantages and disadvantages of each model.

1 Follower
32 Reads
  • [Show abstract] [Hide abstract]
    ABSTRACT: Electric power utilities develop asset management (AM) strategies, based upon their reliability studies, which provide brighter images of the utility company's performance. Failure rate models are highly instrumental in transforming from simple reliability analyses into effective AM strategies. This paper proposes an adaptive network-based inference approach, to model momentary failure rate, in terms of the most influential factors. These factors are achieved through the vast field studies and historical outage fault statistics of the Greater Tehran Electricity Distribution Company (GTEDC) network. Performance of the presented failure rate model is tested, and discussions on capabilities of the model are then presented in detail.
    Transmission and Distribution Conference and Exposition, 2010 IEEE PES; 05/2010
  • [Show abstract] [Hide abstract]
    ABSTRACT: We have introduced a novel framework for realization of Adaptive Autonomy (AA) in human-automation interaction (HAI) systems, as well as several expert system realizations of that. This study presents an expert system for realization of AA, using logistic regression (LR), referred to as Adaptive Autonomy Logistic Regression Expert System (AALRES). The proposed system prescribes proper Levels of Automation (LOAs) for various environmental conditions, here modeled as Performance Shaping Factors (PSFs), based on the extracted rules from the experts' judgments. LR is used as the expert system's inference engine. The practical list of PSFs and the judgments of GTEDC's (the Greater Tehran Electric Distribution Company) experts are used as expert system database. The results of implementing AALRES to GTEDC's network are evaluated against the exact predictions of the presented expert system. Evaluations show that AALRES can predict the proper LOA for GTEDC's Utility Management Automation (UMA) system, which change according to changes in PSFs; thus providing an adaptive LOA scheme for UMA.
    MELECON 2010 - 2010 15th IEEE Mediterranean Electrotechnical Conference; 05/2010
  • [Show abstract] [Hide abstract]
    ABSTRACT: Smart grid expectations objectify the need for optimizing power distribution systems greater than ever. Distribution Automation (DA) is an integral part of the SG solution; however, disregarding human factors in the DA systems can make it more problematic than beneficial. As a consequence, Human-Automation Interaction (HAI) theories can be employed to optimize the DA systems in a human-centered manner. Earlier we introduced a novel framework for the realization of Adaptive Autonomy (AA) concept in the power distribution network using expert systems. This research presents a hybrid expert system for the realization of AA, using both Artificial Neural Networks (ANN) and Logistic Regression (LR) models, referred to as AAHES, respectively. AAHES uses neural networks and logistic regression as an expert system inference engine. This system fuses LR and ANN models' outputs which will results in a progress, comparing to both individual models. The practical list of environmental conditions and superior experts' judgments are used as the expert systems database. Since training samples will affect the expert systems performance, the AAHES is implemented using six different training sets. Finally, the results are interpreted in order to find the best training set. As revealed by the results, the presented AAHES can effectively determine the proper level of automation for changing the performance shaping factors of the HAI systems in the smart grid environment.
    Innovative Smart Grid Technologies Conference Europe (ISGT Europe), 2010 IEEE PES; 11/2010
Show more