Degree
Doctor of Philosophy
Program
Electrical and Computer Engineering
Supervisor
Dr. Jin Jiang
Abstract
This dissertation investigates the use of a Kalman filter (KF) to predict, within the shutdown system (SDS) of a nuclear power plant (NPP), whether a safety parameter measurement will reach a corresponding trip set-point (TSP). The proposed predictive SDS (PSDS) designs aim to initiate shutdown actions at a time which is earlier than conventional shutdown initiation. These early detections are, in turn, expected to improve safety and productivity margins within the NPP. The KF-based point-PSDS design utilizes a linear time-varying (LTV) system model to predict mean safety parameter measurements for comparison against the TSP. The KF considers noise covariances that either have assumed predetermined values, or are estimated online using an adaptive limited memory filter (ALMF). The PSDS is enhanced to consider, by recursive least squares (RLS) estimation, conditions that are uncertain with respect to the assumed system model and noise properties. The result is a KF⁄RLS-based PSDS that compensates for prediction error by RLS-based estimation of deterministic disturbances to the system state and measurement. The PSDS is further enhanced to calculate confidence intervals for the predictions as a function of the propagated error covariance. This enhancement results in interval-PSDS designs that consider confidence in an impending condition by comparing predetermined confidence interval bounds against the TSP. Finally, an optimal-PSDS design is formulated to adapt the effective prediction, e.g. horizon or bias, by limiting and minimizing the probability of missed and false trip occurrences respectively using hypothesis testing methods and optimal alarm theory. In this manner, the optimal-PSDS is made aware of the quality of past predictions. The PSDS designs are compared, through simulation and experiment, against a conventional SDS in terms of response time or time-to-trip for the steam generator level low (SGLL) safety parameter under various conditions of uncertainty, e.g. parameter error or unmeasurable signals. MATLAB-based simulations demonstrate that the PSDS designs are able to reduce time-to-trip. The PSDS designs are then implemented within a Tricon v9 safety-PLC with a scan time that adheres to current nuclear industry regulations. The experimental results reveal that a reduced time-to-trip can be achieved for a real-world system with unknown system-model mismatch.
Recommended Citation
Rankin, Drew J., "Predictive Shutdown Systems for Nuclear Power Plants" (2017). Electronic Thesis and Dissertation Repository. 4635.
https://ir.lib.uwo.ca/etd/4635