ESTIMATING MFN TRAINABILITY FOR PREDICTING TURBINE PERFORMANCE

The present paper examines the relationship between the mapping nonlin earity indicators (distribution angle alpha and distribution gradient beta) of training samples and the empirical trainability of MFN (multi layer feedforward neural networks) with the model problem being the ma pping between turbine efficiencies and blade throat design. An empiric al trainability measure is defined as a means of representing the degr ee of difficulty involved in training an MFN. The raw training samples are preprocessed using two and four different options for input and o utput components, respectively. These options result in mapping cases with different mapping nonlinearities and trainabilities. The results of a numerical experiment confirm that alpha and beta can be correlate d to the empirical trainability of MFN in the context of the model pro blem. Copyright (C) 1996 Civil-Comp Limited and Elsevier Science Limit ed