1994 Fiscal Year Final Research Report Summary
Application of Adaptive Filter for Criticality Surveillance Systems
| Project/Area Number |
05680424
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
エネルギー学一般・原子力学
|
|---|
| Research Institution | Setsunan University |
|---|
Principal Investigator |
YAMADA Sumasu Setsunan University, Faculty of Engineering, Professor, 工学部, 教授 (00029133)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
KISHIDA Kuniharu Gifu University, Faculty of Engineering, Associate Professor, 工学部, 助教授 (90115402)
|
|---|
| Project Period (FY) |
1993 – 1994
|
| Keywords | system identification / subcriticality estimation / ARMA model / adaptive filter / recursive ARMA model identification / criticality surveillance / inverse problem / identifiability |
|---|
| Research Abstract |
Numerical analysis
In the framework of the point reactor kinetics approximation, we have proposed a method for on-line subcriticality monitoring by recursive Auto-Regressive Moving Average (ARMA) Model identification algorithms for the time series of neutron signal fluctuation, however, the transient characteristics for estimating time-varying subcriticality was not satisfactory and also there were problems of over-and/or under-estimations in some cases. To mitigate these problems, we proposed the application of ADF (Adaptive Filter) algorithms. The research was focused on the basic analysis of applicability of ADF algorithms for time-varying subcriticality estimation and we obtained the following conclusions. Estimated parameters and subcriticalities with ADF algorithms have larger stochastic fluctuation than by the one based on recursive ARMA model identification, however, the ADF algorithms have fairly better transient characteristics and no problems of over-and/or under-estimations.
… More
Hence, ADF algorithms can be applicable for estimating subcriticality cahnge in $ units, even though there exist stochastic fluctuations.
Theoretical analysis
We first theoretically confirmed that the ADF algorithm can identify the ARMA model of a stochastic system driven by a random white noise as an inverse model of the original system. Generally speaking, this is an inverse problem. Hence, identifiability of the system is the essential in the system identification. Hence, this problems have been studied in the framework of discrete-time multiple input-output feedback system. Then, we reached the conclusion that the sufficient conditions for identifying the true system from the observed signatures are ;
(1) the feedback system must be of the minimum phase,
(2) the equivalent noise sources of the model assumed for the stochastic system have to be mutually independent.
This indicates that a feedback system of three variables with two observation variables does not satisfy the second condition even if the noise sources of the true system are mutually independent. Less
|
