A successful use case of cognitive networking consists in the reconfiguration of core virtual network topologies (VNTs) based on traffic predictive models obtained by applying data analytics to monitored traffic data. This use case entails long times (several days) to collect enough traffic monitoring samples data at core nodes to allow traffic modelling algorithms (usually at the core controller) to produce accurate models. Notwithstanding, that requirement could not be achieved in the case that metro controllers re-route metro-flows for metro-scope re-optimization purposes. In that case, some metro-flows suddenly change its node entering the core VNT, which drastically impacts on core traffic behaviour. In this paper, we present core-flow traffic models based on the aggregation of metro-flow traffic models. We consider that metro controllers generate traffic models based on monitoring the traffic of the metro-flows and those models are available in a shared repository for the core controller to access them. Moreover, the announcement of metro-flows re-routing from metro controllers to the core controller is assumed to allow fast core-flows models adaptation. Such aggregated models are then used to generate inputs for cognitive core VNT re-optimization purposes.