POET: PRA Operating Experience Tool

[Rennhack]

POET: plant-specific PRA inputs, built from the operating record

A new tool on NukeWorker turns the U.S. fleet's day-by-day operating history into initiating-event frequencies and data-analysis parameters, with the Bayesian work shown. If you know someone who does risk or licensing, this is the one to forward.

Every commercial reactor files a daily report on its power level. Stacked up over years, that record is the raw material behind a lot of probabilistic risk assessment: how often plants trip, how long they stay down, how much time they spend in each operating state. The hard part has always been getting from that raw record to numbers you can defend in a model. A new tool on NukeWorker, POET (the PRA Operating Experience Tool), does that step.

Point it at a plant and it returns plant-specific inputs drawn from operating experience:

• initiating-event frequencies for unplanned (forced) outages, split by duration into short, medium, and long
• exposure on the right basis, both reactor-calendar-year and reactor-critical-year
• plant operating state fractions, time-based, from the actual record rather than assumed
• refueling frequency and duration

For the PRA crowd, the part that matters is how the numbers are handled. Each frequency carries a Bayesian posterior on the gamma-Poisson model with a Jeffreys prior, the same conjugate approach the NRC/INL RADS calculator uses. You get the mean, the 5th and 95th, the median, and an error factor, so a value drops straight into a lognormal basic event as a median with an error factor. Exact frequentist (chi-square) confidence intervals sit right beside the Bayesian ones for comparison.

It does not stop at the plant. POET compares each plant's figures to the NRC/INL SPAR industry averages (the NUREG/CR-6928 series), runs a Bayesian update of the SPAR prior with the plant's own experience, and fits an empirical-Bayes population prior across the fleet for a NukeWorker-native reference. It also runs the prior-versus-plant consistency check before leaning on a generic prior, the kind of step a reviewer asks about.

A few things PRA folks will want to know up front. POET supplies inputs, not a PRA, and it is explicit about that. Where its plant value is a proxy (the general-transient initiating-event rate is inferred from forced-outage behavior, with bias in both directions), it says so plainly rather than quietly merging numbers. Every figure traces back to the operating record it came from, with the window, the exposure basis, the duration thresholds, and the source stamped on it. The method is written up as a Newly Developed Method, with a method basis and verification-and-validation page and a crosswalk to the Initiating Event and Data Analysis supporting requirements of ASME/ANS RA-Sa-2009, as endorsed by NRC Regulatory Guide 1.200, Revision 3. The goal is to be ready to take into a focused-scope peer review, not to claim a conformance a tool cannot have on its own.

When you are ready to use the numbers, POET exports a parameter table formatted for SAPHIRE, CAFTA, or RiskSpectrum, carrying both the gamma parameters and the lognormal median and error factor. Analysts can also bring their own data: there is a self-service import for INPO performance indicators, starting with fuel reliability, so the figures are your fleet's and current.

You do not have to do PRA to find it worth a look. Viewing one plant's inputs is free, and it is a genuinely interesting read on how the fleet actually runs. The benchmarking, uncertainty ranges, year-by-year trends, exports, and your-own-data import come with the Analyst level of the outage subscription. And if you have a friend who lives in SAPHIRE or sits through peer reviews, send them to it. That is who it was built for.

You will find POET under Outages, in the Fleet Data and Stats menu.

Here is a sample for Diablo Canyon:

https://www.nukeworker.com/outages/poet/?plant=Diablo+Canyon+1