UDC 004.855.5
ELM NEURAL NETWORKS IN THE PROBLEMS OF PREDICTING RESIDUAL USEFUL LIFE OF DISK DRIVES
L. A. Demidova, Dr. Sc. (Tech.), full professor, professor of corporate information systems department, Institute for Information Technologies, MIREA – Russian Technological University, Moscow, Russia;
orcid.org/0000-0003-4516-3746, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
I. A. Fursov, post-graduate student of corporate information systems department, Institute for Information Technologies, MIREA – Russian Technological University, Moscow, Russia;
orcid.org/0000-0001-8583-4351, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The problem of predicting the remaining useful life of disk drives by a machine learning model using Extreme Learning Machine (ELM) tools is considered. The aim is to create a neural network model based on ELM, as well as to adapt it by including new generated features to obtain results comparable to those of other feed-forward neural networks. The SimpleRNN recurrent neural network that has already become standard as well as its improved versions in the form of neural network with long short-term memory (LongShort Term Memory, LSTM) and controlled recurrent unit (Gated Recurrent Unit, GRU) show good generalization ability, however, speed learning can be long, while the extreme learning machine does spend much less time on this process. This is especially evident in the problems where there is a lot of data, while ELM generalizes almost as well as its recurrent counterparts
Key words: residual useful life, disk storage, neural network, time series, machine learning, SimpleRNN, LSTM, GRU, ELM