Adaptive Support Vector Regression for Long-Term Prediction under Nonstationay Environments: Prognostics of Components in Nuclear Power Plants
Résumé
Prognostics and Health Management (PHM) aims at diagnosing the state and predicting the future evolution of Systems, Structures and Components (SSCs). In real industrial applications, SSCs are normally operated in nonstationary environments and evolving conditions. In our previous work, an adaptive online learning approach is proposed for the short-term (one-day ahead) prediction by Support Vector Regression (SVR), giving satisfactory performance under nonstationary environments. In this paper, this work is extended to long-term (one/two/three/four-weeks ahead) prediction. An application is presented, concerning of the leakage of the first seal of a Reactor Coolant Pump (RCP) of a Nuclear Power Plant (NPP). The results show the adaptability of the proposed approach and offer some reflections on the long-term prediction problem by data-driven approaches.
Origine : Fichiers produits par l'(les) auteur(s)