Data Mining for Soft Sensing Modeling of Power Plant Parameters

As a new modeling thought, the accurate soft sensing model of power plant parameter was established by data mining method, which obtained effective information from the large number of real-time operation data and avoided low accuracy of conventional modeling method caused by some assumption. A kind of basic modeling mode, including data preprocessing, mining model, verification model and the strategy from data to soft sensing model, was proposed in the paper. Under this mode the main steam flow was taken as an example, the soft sensing model was established based on partial least-square regression with the real-time data collecting in field. The model maximum error was -0.618%, furthermore, the model relative error was within 0.1% when 10% deviation of input variables were appended. The example results indicated that the proposed modeling thought and the mode were effective for the soft sensing, and could enhance the modeling accuracy and stability.