SUB-SPACE PRINCIPAL COMPONENT ANALYSIS FOR POWER PLANT MONITORING

With increasing ease of access to plant-wide process signals in many power stations, operators face a growing challenge, particularly under alarm conditions, to monitor plant operations. However, recognizing that many of the signals are both highly correlated and collinear naturally lends itself to the adoption of data mining techniques. Typically, plant models are identified under normal conditions from historical data records. Subsequently, deviations from trained behaviour are used as indicators of poor plant performance and / or process faults. Both principal component analysis (PCA) and sub-space PCA have been applied to monitoring of a combined cycle gas turbine (CCGT). The capabilities of both approaches are demonstrated following a multi-block implementation, and the influence of external ambient conditions on CCGT performance are also examined.