Data-driven reliability modeling, based on data mining in distribution network fault statistics

Power distribution fault statistics provide splendid resource for extracting experimental knowledge. The extracted knowledge includes the inherit characteristics of the network assets. Analysis and estimation of failures require a comprehensive understanding of faults in terms of the relevant effective parameters. This paper outlines a data-driven model to represent momentary failure rate in terms of the most influential factors based on the study of the recorded historical fault data as well as the expert's experiments in the Greater Tehran Electricity Distribution Company. A methodology is presented for momentary fault causes identification and model construction using artificial neural networks. Satisfactory results indicate that the developed model can easily be implemented to estimate other fault types in power distribution systems.

I. Introduction

Maintaining and improving the reliability of power distribution systems has become a core challenging activity for electric utilities, due to deregulation and privatization of the utility industry. As a result, utilities-who strive to improve their reliability indices-, need to improve their reliability estimation and monitoring methods. For this reason, electric utilities have developed modern tools for optimal asset management, based on distribution network reliability analysis. Fault statistics-treasured in IT-based management automation systems of the electric utility companies-provide splendid resources for extracting experimental knowledge. This knowledge extraction is referred to as Data Mining or Knowledge Discovery [1]. The extracted knowledge include the inherit characteristics of the network asset engineering practice within that company, which might be exclusive to its specific network, thus very valuable for that utility. Hence, the utilities record and analyze their fault statistics; as they would rather prefer to rely on their own context-specific extracted knowledge, rather than the typical values [2], [3]. Therefore, the utilities need practical and adaptive reliability indices, based on their own data, in order to direct their resources properly [3].