Feature Selection and Modelling of a Steam Turbine from a Combined Heat and Power Plant Using ELM

Abstract

The modelling of complex industrial processes is a hard task due to the complexity, uncertainties, high dimensionality, non-linearity and time delays. To model these processes, mathematical models with a large amount of assumptions are necessary, many times this is either almost impossible or it takes too much computational time and effort. Combined Heat and Power (CHP) processes are a proper example of this kind of complex industrial processes. In this work, an optimized model of a steam turbine of a real CHP process using Extreme Learning Machine (ELM) is proposed. Previously, with the aim of reducing the dimensionality of the system without losing prediction capability, a hybrid feature selection method that combines a clustering filter with ELM as wrapper is applied. Experimental results using a reduced set of features are very encouraging. Using a set of only three input variables to predict the power generated by the steam turbine, the optimal number of hidden nodes are only eight, and a model with RMSE less than 1% is obtained.