To apply alternative methods and to perform statistical evaluations of forecast results, the influencing factors relevant for the forecast are set, and the forecasting model is trained, with the help of the integrated Configuration Client.
The training process involves referring back to past input data and results for the load profile and automatically modelling the process behaviour under the influence of various factors. The resultant load profile forecast is shown against the actual known variations and the identified forecast errors.
After the training process BelVis PRO calculates, with a view to potential cost and time savings, the importance of each influencing factor and the number of factors that must remain below a specified maximum mean forecast error. The system is configured to use load profiles and influencing factors in such a way that ensures that individual error values do not corrupt the results. In the event of changes in the sales or consumer structure (e.g. seasonal variations in consumption patterns) the system is retrained to take account of the changes.
The extremely high operating speed allows the system to be retrained without any disruption to the operation. For a 365-day forward forecast with 15-minute intervals the computation time required by BelVis PRO on ALN networks is less than one second!
