作者: Frank Dieterle , Stefan Busche , Günter Gauglitz
DOI: 10.1016/S0003-2670(03)00338-6
关键词:
摘要: Abstract In this study, an algorithm for growing neural networks is proposed. Starting with empty network the reduces error of prediction by subsequently inserting connections and neurons. The type element location where to insert determined maximum reduction prediction. builds non-uniform without any constraints size complexity. additionally implemented into two frameworks, which use a data set limited in very efficiently, resulting more reproducible variable selection topology. applied binary mixtures refrigerants R22 R134a, were measured surface plasmon resonance (SPR) device time-resolved mode. Compared common static all implementations show better generalization abilities low relative errors 0.75% 1.18% R134a using unknown data.