Detection of anomalous data in fossil fired power plants by using informational hierarchical charts

Based on inherent properties of data sets, a method with hierarchical clustering is being proposed for the detection of anomalous data. Some data, acquired in power plants during operation, show strong correlativity. Making use of this correlative property, hierarchical sensor charts are constructed by the hierarchical clustering technique, and some kind of stable pattern strong correlative probing sets may be found. By comparing hierarchical charts, built up by normal data, with chart of actual operational data, some discrepancies may be found, which reflect variations of the sensors' correlative patterns. Herewith anomalous behaving sensors may be detected. Theoretical reasoning and analysis of test results show that this method is a simple and visual method for detecting anomalous data, applicable for engineering practice.