Techniques for improving precision and construction efficiency of a pattern classifier in composite system reliability assessment

Pattern classifiers have been widely utilized to improve computational efficiency in composite power system reliability assessment using Monte Carlo simulation. Construction of a classifier for reliability assessment demands sufficient amount of training vectors such that success and failure states can be effectively differentiated. Typically, the training vector is obtained by sampling states according to their original distributions. Failure states are therefore hardly sampled. This raises two issues. First, a large number of samples are needed to extract sufficient amount of failure states. Second, a set of training vectors becomes highly imbalanced, leading to undesirable level of precision of a classifier. This paper proposes two techniques to address aforementioned issues as well as to enhance precision of a classifier. The first technique is based on worsening system reliability to obtain balanced amount of success and failure states for training vectors. This technique enhances construction efficiency of a classifier in general and solves imbalance issue. The second is based on relaxed decision boundary which is used to improve precision of general classifiers. Various case studies are conducted on IEEE-RTS79 in order to justify the proposed techniques. Results show that the proposed techniques improve both precision and construction efficiency of a classifier.

► We point out that in any systems, there is only one classifier that separates failure and success states regardless of system reliability.
► Using this statement, one can train a classifier much faster than traditional approach by worsening system reliability to obtain informative states.
► We classify states using a proposed relaxed decision boundary which can classify states more precisely than a traditional rigid decision boundary.