Predictions and prognostics

Transcription

Predictions and prognostics
Chaire
« Sciences des Systèmes et
Défis Energétiques (SSDE) »
Fondation Europé
enne pour les Energies de Demain — É
lectricitéde France (EDF)
RESEARCH LINE(S):
Prediction and prognostics
Jie LIU
Valeria VITELLI
Ronay AK
RESEARCH TOPICS:
Prediction with uncertainty quantification.
Kernel-based pattern recognition methods for prognostics of nuclear
components’ Estimated Time To Failure (ETTF).
Nonlinear regression methods for prognostics; prediction of functional
patterns.
RESEARCH METHODS AND APPLICATIONS:
 CURRENT
Feature selection methods and correlation analysis for the pretreatment of input data.
Multi-objective genetic algorithm approach for the estimation of Neural Network (NN)-based Prediction Intervals (PIs) (application to wind speed data).
Probabilistic Support Vector Machine for fault prognosis of electrical components.
Local Gaussian Processes (LGP) for multivariate nonlinear trend estimation.
 FUTURE
Ensemble methods for improving accuracy and robustness of NN-based PIs.
Bayesian Neural Networks and Particle Filtering for prognostics.
Functional prediction and uncertainty quantification models for high dimensional signals.
RESEARCH RESULTS:
 OBTAINED
Data-driven method for determining the optimal values of the parameters of a NN maximizing coverage
probability (CP) and minimizing the prediction interval width (PIW) in Pareto optimality sense. Evaluation of
the method via a synthetic case study of literature, and a real case study of short-term wind speed prediction.
State-of-the-art review of existing probabilistic support vector machines methods for ETTF prediction.
 EXPECTED
Definition of a general strategy for input data pretreatment.
Real case studies: wind power forecasting, load prediction, etc.
Energy components ETTF prediction.
COLLABORATIONS:
Fig. 1: Estimated PIs for one hour ahead wind speed prediction (dashed lines), and
actual wind speed data (solid lines).
SPONSORS:
EDF R&D, Département Simulation et Traitement de l’information pour l’Exploitation des systèmes de Production (STEP).
China Scholarship Council (CSC).