EMPIRICAL PROCESS MODELING IN FAST BREEDER-REACTORS

A non-linear multi-input/single output (MISO) empirical model is intro duced for monitoring vital system parameters in a nuclear reactor envi ronment. The proposed methodology employs a scheme of non-parametric s moothing that models the local dynamics of each fitting point individu ally, as opposed to global modeling techniques-such as multi-layer per ceptrons (MLPs)-that attempt to capture the dynamics of the entire des ign space. The stimulation for employing local models in monitoring ri ses from one's desire to capture localized idiosyncrasies of the dynam ic system utilizing independent estimators. This approach alleviates t he effect of negative interference between old and new observations en hancing the model prediction capabilities. Modeling the behavior of an y given system comes down to a trade off between variance and bias. Th e building blocks of the proposed approach are tailored to each data s et through two separate, adaptive procedures in order to optimize the bias-variance reconciliation. Hetero-associative schemes of the techni que presented exhibit insensitivity to sensor noise and provide the op erator with accurate predictions of the actual process signals. A comp arison between the local model and MLP prediction capabilities is perf ormed and the results appear in favor of the first method. The data us ed to demonstrate the potential of local regression have been obtained during two startup periods of the Monju fast breeder reactor (FBR). ( C) 1998 Elsevier Science Ltd. All rights reserved.