SVM hyperparameters tuning for recursive multi-step-ahead prediction - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Neural Computing and Applications Année : 2016

SVM hyperparameters tuning for recursive multi-step-ahead prediction

Résumé

Prediction of time series data is of relevance for many industrial applications. The prediction can be made in one-step and multi-step ahead. For predictive maintenance, multi-step-ahead prediction is of interest for projecting the evolution of the future conditions of the equipment of interest, computing the remaining useful life and taking corresponding maintenance decisions. Recursive prediction is one of the popular strategies for multi-step-ahead prediction. SVM is a popular data-driven approach that has been used for recursive multi-step-ahead prediction. Tuning the hyperparameters in SVM during the training process is challenging, and normally the hyperparameters are tuned by solving an optimization problem. This paper analyses the possible objectives of the optimization for tuning hyperparameters. Through experiments on one synthetic dataset and two real time series data, related to the prediction of wind speed in a region and leakage from the reactor coolant pump in a nuclear power plant, a bi-objective optimization combining mean absolute derivatives and accuracy on all prediction steps is shown to be the best choice for tuning SVM hyperparameters for recursive multi-step-ahead prediction.
Fichier principal
Vignette du fichier
Tuning hyperpara of recursive SVM.pdf (794.57 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-01342889 , version 1 (06-07-2016)

Identifiants

Citer

Jie Liu, Enrico Zio. SVM hyperparameters tuning for recursive multi-step-ahead prediction. Neural Computing and Applications, 2016, 28 (12), pp.3749 - 3763. ⟨10.1007/s00521-016-2272-1⟩. ⟨hal-01342889⟩
197 Consultations
163 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More