Multi-stage identification framework for steam generator modeling with gap metrics

Abstract

This paper focuses on the modeling of nonlinear multi-input, multi-output coupled steam generators in nuclear power plants. To address the problem of insufficient modeling accuracy due to the low signal-to-noise ratio of closed-loop operation sampling, a multi-stage identification framework integrating the least-squares method, instrumental variables method, and predictive error method is proposed. This framework is divided into three stages: parameter initialization, initial value estimation, and parameter optimization, aiming to progressively enhance the quality of model identification. Moreover, to address the complicated problem of piecewise linearization of nonlinear systems, a piecewise linearization strategy based on gap metrics with decreasing intervals from large to small is proposed. The strategy first uses data from large time-domain intervals for model identification, followed by refined time-domain interval data. This approach simplifies the piecewise linearization process and reduces the generation of redundant models. Through extensive simulation validation, this method has demonstrated its effectiveness in the modeling of nuclear power steam generators and valve condition monitoring. The enhancement percentages of the three steam generators are 18.95%, 10.19%, and 48.85%, respectively, highlighting the significant improvement in model accuracy. Additionally, fault monitoring simulations for the steam generator feedwater valve show significant differences between the predicted values and errors of the normal model under simulated fault scenarios. By setting safety thresholds based on historical data, the working status of the valve can be effectively monitored. This paper provides new research ideas and practical solutions for nuclear power steam generator modeling and condition monitoring, contributing to the advancement of related fields.