Importance Sampling Using Multilabel Radial Basis Classification for Composite Power System Reliability Evaluation

Doğan Urgun

doi:10.1109/JSYST.2019.2944131

2020-06-01

PRE

Doğan Urgun

Chanan Singh *

Vijay Vittal

DOI：10.1109/JSYST.2019.2944131

原文链接

原文求助

摘要

En 中文

Resource additions to electric power grids are planned in advance to maintain reliability of electric power supply. This is achieved by performing reliability studies for calculating reliability indices at the planning stage to ensure that the required levels of reliability will be met. This article proposes a new state classification approach to calculate power system reliability indices using a combination of multilabel radial basis function (MLRBF) networks and importance sampling (IS) within the framework of Monte Carlo simulation process. Multilabel classification algorithms are different from single-label approaches, in which each instance can be assigned to multiple classes. This characteristic gives MLRBF the capability to classify composite power system states (success or failure), at the bus as well as system level. Bus level indices provide useful reliability information for locational reliability, which allows more rational and equitable distribution of resources. MLRBF classification does not require optimal power flow (OPF) analysis; however, OPF is required for the training and cross-entropy optimization phases. This article shows that the scope and computational efficiency to evaluate reliability indices can be significantly increased if the proposed MLRBF classifier is used together with well-known variance reduction technique of IS. As the classifier is trained to recognize the reliability status of a system state, it can also be used in operational planning when a number of scenarios need to be evaluated in a short time for their ability to satisfy load. The proposed method is illustrated using the IEEE reliability test system for different load levels. The outcomes of case studies show that MLRBF algorithm together with IS provides excellent classification accuracy in reliability evaluation while substantially reducing computation time and enhancing the scope of evaluation.

Keyword:

Power system reliability

Reliability

Monte Carlo methods

Training

Neural networks

Planning

Composite power system reliability evaluation

cross entropy (CE) algorithm

importance sampling (IS)

Monte Carlo simulation (MCS)

multilabel radial basis function (RBF) learning algorithm