On line identification of causal relationships between variables in the feed water system of a nuclear power plant

Abstract

On line identification of nonlinear causal relationships between variables in the feed water control loop of a nuclear power plant is reported. The knowledge about the observable variables of the application has been used in the design of the architecture for the network, in the local function of each elemental processor (quadratic expansion of inputs and recurrence) and, finally, in the selection of the supervised learning algorithm. This learning algorithm is based on the local evaluation and propagation of individual output errors for each sample in the training set. This nonlinear model with delays and quadratic expansion of inputs is compared with the more usual-linear dynamic network and a clear improvement is observed. Some preliminary conclusions on the influence of signal noise relationships and the criteria for the selection of the appropriated sampling period are also included.