A comparison of wood log dissimilarities to predict sawmill output with k-Nearest Neighbor algorithms
Résumé
In the sawmill industry, to predict the set of lumber that would be sawed from specific wood logs is a difficult problem. Even if they exist many sawmill simulators able to simulate the sawing process in order to predict these quantities, they can be too slow for large scale industrial problems. Replacing these simulators with machine learning surrogate models, or metamodels, is a promising avenue of research to speed up predictions. One such research direction is based on the computation of pairwise dissimilarities between logs, used, for example, by k-nearest neighbor algorithms. Interesting results have been obtained with the so-called iterative closest point (ICP) dissimilarity who has, however, several undesirable properties. This paper explores another alternative based on ensemble of shape functions.
Origine : Fichiers produits par l'(les) auteur(s)