Failure forecast engine for power plant expert system shell

This paper describes a novel technique for failure forecast in a power plant controlled by computerized SCADA system. The fault forecasting engine is designed as part of development of expert system shell for power plants. It is a hybrid approach incorporating data mining, fault models, clustering and time series analysis. For real time monitoring of plant condition, graphical models are constructed by K means clustering algorithm. To build the time series value forecasting model, Multilayer Perceptron (MLP) based neural network is used. By using latest history data base of SCADA system training and testing of the models are done. Models once created, is updated in the model library for providing adaptive nature to the proposed system. The Graphical User Interface (GUI) of the forecasting engine displays the variation of all sensor values affecting a particular fault for next time instances.