Self-Organized Maps in the Identification of IRIS Reactor Transients
B. D. Baptista F., A. C. O. Barroso
Instituto de Pesquisas Energéticas e Nucleares
Comissão Nacional de Energia Nuclear - CNEN
São Paulo, SP, CEP 05508-000 Brazil
Phone: +55-11-3816 9199/262, Fax: +55-11-3816 9423, E-mail: bdbfilho@ipen.br
Abstract - IRIS, the International Reactor Innovative and
Secure, is an international program that aims to develop
an integrated primary system reactor with innovative features
that can meet most of the requirements considered
in the Generation IV Roadmap Study as a Near-Term Deployment
System. IRIS philosophy is the Safety by Design
Approach. The IRIS concept raises many challenges
that only can be addressed by an international cooperative
effort. This paper analyzes and presents many results of a
possible solution for one of these challenges, the transients’
classification and identification that may be used
to help the safe operation of IRIS. The approach studied
in this paper is based on self-organized maps (SOM), a
specialized class of artificial neural networks (ANN). The
results presented complement previous reports and show
that SOM is a quite promising tool in the identification of
initiating transients of the IRIS reactor. The ability of this
kind of ANN in promptly identify the deviation from
normal operation, with the aid of only few process sensors
signals, may be used to develop the IRIS Transient
Identification System (TIS).
I. INTRODUCTION
IRIS is an international cooperation effort to design a nuclear
energy system capable of meeting many of the requirements
for the new generation of nuclear power
plants. IRIS takes advantage of its integral configuration
to implement a safety by design approach to meet challenging
safety goals [Carelli et al., 2001]. In this philosophy,
every possible source of “safety vulnerability” has to
be addressed and adequately coped with [Packer, 2002].
Safety by design and the safety barriers recommended by
the principle of defense in depth can eliminate some and
attenuate many of the consequences of mechanical, electrical
and human fails, but to provide a really confident
plant, modern tools to provide “good information” for the
operation team are required.
This paper describes a promising tool able to identify
transient events, classifying them into normal or abnormal
transients. The system under study can provide operator
support, helping the safe operation of IRIS. The approach
is based on self-organized maps (SOM), a special class of
artificial neural networks (ANN), operating on-line with
the reactor instrumentation. This kind of representation
can allow to the operator to watch how a given transient is
evolving with respect to its severity, as the time path of
the activated units is migrating towards the border of a
new class of transients.
The central idea is to develop a system capable of identify
and classify a transient type in its early stage. This system
was first described in 2003 [Baptista and Barroso, 2003],
and after this time it was improved and tested with more
data considering a greater set of transient simulations. The
data consist of sets of eight key process variables recorded
in the first 30 seconds of several types of transients, normal
and abnormal and also in steady state conditions.
II. IRIS BRIEF DESCRIPTION
IRIS is a modular, integral, light water cooled reactor, designed
for a power of 335 MW(e)/module. The most relevant
technical characteristics of IRIS are discussed in detail
in references {[Carelli et al., 2000], [Carelli et al.,
2001a], [Carelli et al., 2001b], [Petrovic et al., 2000],
[Oriani et al., 2001], [Conway et al., 2001]}. Its “safety by
design” approach, where accidents are “designed out” to
the maximum extent possible, instead of engineering how
to cope with their consequences is presented in [Carelli et
al., 2001].
The IRIS integral vessel houses the reactor core, its support
structures, upper internals, control rod drives, eight
steam generators, internal shields, pressurizer and heaters,
and eight reactor coolant pumps (Fig.1). Hot coolant rising
from the reactor core to the top of the vessel is
pumped into the steam generators annulus. The integral
vessel configuration is essential to the safety by design
approach as shown in [Conway et al., 2001] and thus it is
key to satisfy the enhanced safety requirement.
III. SELF-ORGANIZED MAPS
Self-organized maps are artificial neural networks with a
single layer where the units are placed in a 1-D or 2-D
grid. In the 2-D SOM, the units are placed in a square or
hexagonal lattice (Fig.2). The training of the SOM is
based on the competitive learning concept: units compete
with each other to be activated when a specific pattern is
presented and the result is that just a single unit is really
active at a given moment. The original idea of the competitive
learning –winner takes all– was proposed in 1958
.....