Classification of Transient Events of Nuclear Reactor Using Hidden Markov Model

Abstract

This article describes a part of on-line system for a residual life of a pressure vessel shell. In this system there appears a need for determining of a real history of a pressure vessel described as a sequence of so called transient events. Each event (there are 23 events now) is given by its template. It is theoretically necessary to compare data measured in a real history with all possible sequences of transient events. This solution in intractable and that is why a more sophisticated solution had to be used. Because this task is very similar to task of an automatic speech recognition, models and algorithms used to solve speech recognition tasks can be efficiently used to solve our task. Of course there are some different circumstances caused by the fact that the transient events take much longer than words and their number is much smaller than the number of words in speech recognition system's vocabulary. But the inspiration from speech recognition was very useful and the known algorithms rapidly decreased the design time.