Data Mining Techniques Applied to Power Plant Performance Monitoring

The incorporation of distributed control systems in power plants has become increasingly standard. However, particularly under alarm conditions, ease of access to plant-wide signals complicates the task of monitoring plant operation. Data mining techniques are ideally suited to dealing with such data, which tends to be highly correlated and collinear. Typically, models identify relationships under normal conditions from historical data. Subsequently, these discovered relationships form the basis for detecting unusual deviations from trained behaviour. Principal component analysis and partial least squares techniques are demonstrated, and shown to take full advantage of readily available DCS data, and provide an informative monitoring method. Copyright  2005 IFAC