The autocorrelation descript the relation between the time series values and its h-lag period. One of the most important characteristic for the stable sequence is that the autocorrelation function \rho_h is rapidly descend to 0 by the increase of the h-lag period.
? smoothen the time series
一个观点： These recurrent connections differ from the other network connections as they connect neurons within the same hidden layer. This gives the neural network memory which makes it particularly well suited to the problem of predicting CPU demand.
Parameter sweeps revealed that any more than three hidden neurons did not lead to any increase in performance
Parameter sweeps also found that more than two inputs did not lead to any increase in performance