Development of Advanced Plant Efficiency Monitoring System Using Data Reconciliation

The thermal efficiency of power plants is determined by the measured values of instruments. It is known that the measured values have uncertainty depending on the accuracy of instrument and process fluctuations. There have been many efforts to reduce uncertainties to improve the quality of thermal efficiency management. ASME performance test codes recommend the average of the measured values to minimize random errors and the periodic calibration to exclude systematic errors. While a recent trend tends to concentrate on on-line efficiency monitoring, which is different from the conventional off-line thermal performance tests, unforeseen features became troublesome: Since the on-line efficiency monitoring is performed during daily operation, it has high possibility of including random errors and/or systematic errors. The data reconciliation (DR) has been expected to be a candidate for managing uncertainty in on-line efficiency monitoring. The DR is developed using a physical model with the first principles such as mass/energy balance. Due to this nature, it has the advantage in eliminating random errors. However, the DR based on a physical model has several problems: One issue is the accuracy of the turbine cycle model. The conventional DR algorithm is difficult to apply for turbine cycle with which a large size matrix and non-linear equations are involved. Furthermore, since the minimum number of instruments for operation is generally installed in a turbine cycle, the redundancy requirement for the DR algorithm is unlikely to be satisfied. This study aims at developing an advanced efficiency monitoring method using the simulation-assisted DR. We proposed the idea to periodically establish a turbine cycle model and set up boundary conditions to generate the multiple candidates of reconciled datasets using a developed turbine cycle model. In this study, the conventional DR algorithm was followed to find a reconciled dataset, but we introduced the concept of minimum averaged squared errors to decide a set of systematic errors and the optimal reconciled dataset. We also developed the strategy to apply the results of the proposed DR algorithm to current plant efficiency monitoring systems