This module (Artificial Neural Network Time Series – ANN-TS) uses modeling potential of a neural network to predict and forecast future values of a univariate time series. A univariate time series is a series of n measured values y_i (i = 1,.., n) assuming  that

• values y_i are measured in regular time intervals (time can be replaced by index without loss of information), or y_i describe quantitatively time-sequential units or events without leaving out any from the series.
• values y_i are preferably real-valued (real valued predictions make sense)
• values y_i are in some way related to one or more previous values y_i–k , y_i = G(y_i-1, y_i-2, … y_i-r) + ε_i, or, shortly G(i, r) + ε_i, where r < n is called the depth of the model.

Examples of such a series can be sampled variables in technological processes, periodical outputs from stable or unstable processes, parameters of natural processes in life sciences, geo sciences, physical or chemical sciences, series of financial or economical indices, prices, rates and so on.

Neural Time Series (ANN-TS) - Pdf manual
Neural network - Pdf manual

Protokol: Task name Data Type of model Model depth Forecast length Validation Independent variable Transformation type Dependent variable Transformation type Layer, Neurons Sigmoid steepness Moment Training speed Terminate when error < Training data (%) Termination conditions Optimization report No of iterations Max training error Mean training error Max training error Mean training error Weights Layer / Neuron Relative influence Time series

Grafy: Plot of the time series with ANN-TS prediction and validation or forecast (as selected) Plot of residuals, or differences measured minus predicted including validation data, if selected Graphical representation of the network architecture Relative influence Plot of the training (network optimization) process