A Soft-sensing Model for Feedwater Flow Rate Using Fuzzy Support Vector Regression

Thermal nuclear reactor power is typically evaluated with secondary system calorimetric calculations based on feedwater flow rate measurements. The feedwater flow rate should therefore be measured accurately. Venturi meters are currently used to measure the feedwater flow rate in most pressurized water reactors (PWRs). The long-term operation of a nuclear power plant causes a buildup of corrosion products near the orifice of the meter. This fouling increases the measured pressure drop across the meter, thereby causing an overestimation of the feedwater flow rate. Whenever calorimetric calculations are conducted during an operating cycle, the thermal reactor power must be decreased to match the false feedwater flow rate overestimated by the Venturi meter. This requirement means that a nuclear power plant must be operated at a lower-than-planned power level. The fouling phenomenon of Venturi meters is the most significant contributor to PWR derating, which ranges from 0.5% to 3%. The most common practice for resolving this problem at PWRs is to inspect and clean the Venturi meters during every refueling period. However, fouling can reappear in as quickly as a month. With time, the accuracy of the existing hardware sensors becomes degraded due to the fouling phenomena of the Venturi meter. Many researchers have therefore been interested in resolving the inaccurate measurements of the feedwater flow rate [1-4]. Hence, in this study we developed a soft-sensing model for predicting the feedwater flow rate. Recently, many researchers have paid considerable attention to soft-sensing, which uses other readily available on-line measurements. This type of soft sensor can either replace existing hardware sensors or be used in parallel to provide redundancy and to verify whether the sensors are drifting [4-10]. The problem can be resolved by using learning and soft computing techniques if the process dynamics for evaluating the process variables is a priori unknown or difficult to model. We therefore developed a fuzzy support vector regression (FSVR) model that can increase the thermal efficiency of a nuclear power plant by making accurate on-line predictions of the feedwater flow rate.