METHOD FOR COMPUTER-AIDED TRAINING OF ONE OR MORE NEURAL NETWORKS

摘要

FIELD: information technology.;SUBSTANCE: rate of change of the initial data time series is calculated to obtain derivative values. These derivative values undergo Empirical Mode Decomposition, which is a well-known technique. Modes extracted via said Empirical Mode Decomposition, as well as some delayed values of the time series are used as inputs for neural networks, while the output of these networks are future values of the time series, which must be predicted. These networks are trained and used to predict future values of the time series. The invention provides good prediction results by using derivative values as inputs, while past values are also considered as inputs. The invention can be used to predict values of any time series and, particularly, for predicting the behaviour of engineering systems, for example, vibrations arising in an engineering system.;EFFECT: broader functional capabilities of neural networks by further enabling processing of data time series, having value trends, particularly vibration diagnosis data.;11 cl, 12 dwg