Reduction of uncertainty of thermal performance calculations in nuclear power plants

The OECD Halden Reactor Project (HRP) has taken an active role in facilitating implementation of technology advances and in particular application of condition monitoring techniques for operation support. TEMPO [1] is a system based on physical models for thermal performance monitoring and optimization developed at the HRP. The system aims at satisfying information needs associated with condition monitoring, on-line calibration monitoring of plant measurements, process fault detection and diagnosis. Data reconciliation [2] combines a first principles model and process measurement data to calculate the most likely process state. The use of data reconciliation requires an analytically redundant set of measurements, i.e. that information about measured entities can be deducted from other measurements through the model, e.g. from mass and heat balances. Uncertainties of the computed results depend on both the uncertainties of the measurement data and the inaccuracy and uncertainty of the data reconciliation model. The TEMPO system has been in daily use in the analysis of thermal performance of the secondary side of the Loviisa NPP turbine cycle for several years [3]. This paper presents an approach where distributions of measurement errors are estimated from observed time series data. Principal Component Analysis (PCA) is used to identify process variability modes observable in several measurements, which are then subtracted from observed data to estimate the distributions of measurement errors. Uncertainty estimates for the reconciled values are then computed using Monte Carlo simulation. The approach described above is applied to a turbine section of a nuclear power plant. The estimated distributions of measurement errors are utilized in both reactor power monitoring and sensor drift detection. The results indicate that the measurement uncertainty in the process examined is not too significant a source of uncertainty affecting the results computed with data reconciliation.