Future States for a Present-State Estimate, in the Contextual Perspective of In-Core Nuclear Fuel Management

We discuss AI-based and related approaches to problem-solving for an allocation design problem, with which fuel managers at nuclear power plants are faced on a peri-annual basis: how to position fuel-units in the reactor core (whose planar section is a symmetric grid), to achieve better. And usually better performance during the operation period up to the next EOC point (end of cycle), when over a new downtime period the allocation problem will have to be solved again. Forecasting in this domain is not accurate enough to enable the preparation of robust solutious before the reactor has actually been shut down, and the depletion degree of the fuel-units, let alone their possibly damaged status. can be ascertained. Various approaches exist. Westinghouse's LPOP is based on backcalculation from a target power-distribution. In contrast, the FUELCON expert system (Galperin and Nissan, 1988) applies hyperrecursion on a heuristic ruleset to generate alternative candidate solutions by the hundreds, these being then simulated for parameter prediction, and visualised as "clouds" of dots in the plane of power peaking and cycle length. Its successor, FUELGEN (Zhao, 1996 sqq), applies evolutionary computing, again by hyperrecursion. Arguably anticipation - and thus hyperincursion because of its being joint with hyperrecursion - apply to both FUELCON and FUELGEN: during the operation cycle (other than at downtime periods), state observability not obtaining for such variable which require direct inspection, current state estimates are based on those parameters which are observable, along with the forecast that was obtained by simulation at downtime, so that current state estimates partly depend on future states.