Prediction
Prediction
Network Time-Series Forecasting
Managing complex systems composed of multiple interconnected elements that are influenced by external factors is a difficult task. Today technology allows to monitor what is happening on every single element of a complex network, but for optimal planning an additional step is necessary: making forecasts on different time horizons and with different degrees of detail.
- Description and benefits
- Application examples
We developed the Network Time-Series Forecasting solution in order to be able to make accurate estimates of future states by going into detail on the individual element, rather than just stopping at the overall system level.
Starting with data from large-scale sensor networks, our solution allows to analyze and forecast future trends in sensor values at all degrees of interconnect density. This means that a prediction of future trends that takes into account all existing information is available for each sensor measurement.
In addition, our solution provides a series of interpretative considerations of the phenomenon, such as the determining factors for a certain forecast or the strongest correlation links, so that the user can identify the most effective corrective actions, if any. The availability of an interpretable prediction also allows for the rational involvement of all key stakeholders: from those who design, to those who maintain operations, to those who use the network on a daily basis.
Maintenance planning
Ability to decide when to make an asset non-operational in order to minimize the negative impact in terms of disruption and cost
Reacting effectively to unexpected phenomena
By predicting load peaks (such as socio-political events or holidays) it is possible to make the best decisions due to the interpretability of forecasts
Increase speed of action
With this solution, planning and service continuity processes and decisions become faster and more informed
Sectors
Application examples of Network Time-Series Forecasting
IT
Network traffic forecasting and analysis to prevent slowdowns and schedule maintenance
Energy & Utilities
Real time consumption forecasting in order to optimize transport (dispatching, redelivery), network control and maintenance in a smart-grid context
Media & ADV
Monitoring and forecasting user impressions on multiple web pages
Deepening
Operating logic
The data driven agnostic models, that do not require the knowledge of the phenomena underlying the data generation, are placed side by side in an ensemble to interpretable models linked to the functioning principles or other characteristic properties of the considered system. The latter may be physical, as in the case of distribution networks, or immaterial, as in the case of financial data. Thanks to the ensemble a vision and interpretability of the results is preserved, but it is also possible to understand where this is not sufficient and where agnostic models can capture hidden and unknown dynamics.
Success story: application of Network Time-Series Forecasting solution
The solution for Terna
Terna, the operator that manages the electricity transmission grids, has reduced forecasting errors by 15% thanks to Network Time-Series Forecasting.
C.F. e P.IVA 07015910966 | Via Schiaffino 11, 20158 Milano