Multi-step forecasting method based on recurrent neural networks model

In order to solve the problems of the traditional feedforward neural networks with a long-term prediction, an alternative neural model, multi-step recurrent neural model (MSRN), based on a partially recurrent neural network, is proposed. For the recurrent model, a learning phase with the purpose of long-term prediction is imposed, which allows to obtain better predictions of time series in future. In order to validate the performance of the recurrent neural model to predict the dynamic behavior of the series in the future, two-different data time series have been used. An artificial data time series and the vibration data measured from the real time series are used to compare the ability of multi-step prediction. The results show that the MSRN model can contribute a good accuracy of prediction.