摘要：

Inctual industrialields, dataorodellingresually generated gradually, whichequireshathe data-based predictionodelashenline learningapability.lthoughanynline learninglgorithmsaveeen proposed,he generalization performance needsoe improvedurther. Inhis paper,tructure-adjustablenline learning neural network (SAO-ELM)asednhe extreme learningachine (ELM) with quicker learningpeedndetter generalization performance is proposed.irstly, ELM ishanged intotructure-adjustable learningachine, in whichhe numberf nodes in itsingleidden layeranedjusted.hen,pecialtrategy is developedoandlehe difficultyhathe newddedidden nodes'utputsorrespondingohe discardedraining dataannotebtained.fterhat,n iterative equation is presentedopdateheutputatrix whenidden nodesredded.esultsf numericalomparisonasedn dataromheeal worldenchmark problemsndnctualontinuousasting processhowhathe performancefAO-ELMasignificantdvantagesverhatfheypicalnline learninglgorithmsn generalization performance. Inddition,AO-ELMetainsheeritf quick learningharacteristicf ELM.

展开