Identification of nuclear plant parameters using experimental data and high-order dynamic models

Large linear dynamic models for nuclear reactor systems are widely used for simulation and control system design. It is important to be able to verify these models and the parameters in them. Existing parameter identification techniques are very time consuming for use with large systems. Identification is achieved by an optimization procedure that adjusts system parameters to minimize differences between experimental frequency responses and theoretical frequency responses obtained from the dynamic model. A new method that uses a partitioned matrix technique was developed. This technique constitutes a very efficient analysis algorithm for large models when implemented on the digital computer. The work included a study of methods for assessing the identifiability of parameters by fitting dynamic test data. The Fisher information matrix was found to be useful for this purpose. It was also found that the frequency dependency of the sensitivity function is important in determining identifiability. The measurements should include frequencies where the sensitivity to the parameter of interest is largest. Also, it was found that separate, unique identification of parameters with parallel curves of sensitivity versus frequency is impossible regardless of how large the magnitudes of the sensitivities are. The method was demonstrated in a test case. It used data (from the Oconee I pressurized waterreactor) and a 29th-order model. The results demonstrated that the computational requirements are reasonable for large systems and that the procedure can identify parameters if all the necessary conditions are satisfied. In general, the work has provided a systematic method for parameter identification in systems described by large linear dynamic models