. da ta - a n ] 1 7 M ar 2 00 3 The power law character of off - site power failures

A study on the behavior of off-site AC power failure recovery times at three nuclear plant sites is presented. It is shown, that power law is appropriate for the representation of failure frequency-duration correlation function of off-site power failure events, based on simple assumptions about component failure and repair rates. It is also found that the annual maxima of power failure duration follow Frechet distribution, which is a type II asymptotic distribution, strengthening our assumption of power law for the parent distribution. The extreme value distributions obtained are used for extrapolation beyond the region of observation. 1. Introduction Estimation of off-site power failure characteristics is important for the safe design and operation of nuclear power plants. The emergency power supply requirements are based on the ability to forecast the maximum credible loss of off-site power (LOSP) duration during the life time of the plant. The observed data is available only over a period of less than two decades and scarce on long duration failures. We address two questions: first, what is the nature of the parent distribution, i.e., the form of frequency-duration correlation function and second how to extrapolate for failure durations longer than the observed maximum. First we show, based on simple assumptions about the off-site power supply system component failure and repair rates, that the frequency-duration correlation function is a power law of the form, y ∝ x −k. Next, it is established that the distribution of annual maxima, of observed failures follow Type II asymptotic distribution (Frechet distribution, see Johnson 1970). This result strengthens the premise, that the parent distribution for power failure duration-correlation is power law in character. Extreme value analysis is also used to make extrapolations for more number of years than observed, for instance to estimate the most probable loss of off-site power duration in say, 50 or 100 years.