We address the problem of multi-step ahead time series signal prediction in the energy industry, with the aim of improving maintenance planning and minimizing unexpected shutdowns. For this, we develop a novel method based on the combined use of Ensemble Empirical Mode Decomposition and Long Short-Term Memory neural network. Ensemble Empirical Mode Decomposition decomposes the time series into a set of Intrinsic Mode Function components which facilitate the prediction task by effectively describing the system dynamics. Then, Long Short-Term Memory neural network models perform the multi-step ahead prediction of the individual Ensemble Empirical Mode Decomposition components and the obtained predictions are aggregated to reconstruct the time series. A Tree-structured Parzen Estimator algorithm is employed for the optimization of the hyperparameters of the Long Short-Term Memory neural network. The proposed method is validated by considering various long-term prediction horizons of real time series data acquired from Reactor Coolant Pumps of Nuclear Power Plants. The results show the superior performance of the proposed method with respect to alternative state of the art methods.

Ensemble empirical mode decomposition and long short-term memory neural network for multi-step predictions of time series signals in nuclear power plants

Baraldi P.;Zio E.
2020-01-01

Abstract

We address the problem of multi-step ahead time series signal prediction in the energy industry, with the aim of improving maintenance planning and minimizing unexpected shutdowns. For this, we develop a novel method based on the combined use of Ensemble Empirical Mode Decomposition and Long Short-Term Memory neural network. Ensemble Empirical Mode Decomposition decomposes the time series into a set of Intrinsic Mode Function components which facilitate the prediction task by effectively describing the system dynamics. Then, Long Short-Term Memory neural network models perform the multi-step ahead prediction of the individual Ensemble Empirical Mode Decomposition components and the obtained predictions are aggregated to reconstruct the time series. A Tree-structured Parzen Estimator algorithm is employed for the optimization of the hyperparameters of the Long Short-Term Memory neural network. The proposed method is validated by considering various long-term prediction horizons of real time series data acquired from Reactor Coolant Pumps of Nuclear Power Plants. The results show the superior performance of the proposed method with respect to alternative state of the art methods.
2020
Ensemble empirical mode decomposition
Long short-term memory recurrent neural network
Multi-step ahead prediction
Predictive maintenance
Prognostics
Reactor coolant pump
File in questo prodotto:
File Dimensione Formato  
1-s2.0-S0306261920317281-main.pdf

Accesso riservato

: Publisher’s version
Dimensione 5.14 MB
Formato Adobe PDF
5.14 MB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1159022
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 66
  • ???jsp.display-item.citation.isi??? 48
social impact