Hidden state variables and AR-MA formulation of reactor noise.
Kuniharu Kishida, Hiroshi Sasakawa · Journal of Nuclear Science and Technology · 1980
Coupled Markovian Langevin equations in the conventional reactor noise theory are cast into a time series model with the aid of contraction of information, coarse graining in time, and the concept of innovation. This makes it possible to physically interpret time series data and to apply them to reactor diagnosis. An evolution equation for the variance of the innovation is found and discussed in connection with the Riccati equation used in prediction theory. The relation with the Markovian representation well-known in control theory is also discussed. A simple example is treated to demonstrate that the time series model has less information than Langevin equations.