6G-SANDBOX

The 6G-EARN service has been tested and its performance has been evaluated in different scenarios. A number of KPIs and KVIs have been derived and evaluated for both the service itself and the 6GSANDBOX experimentation platform. Regarding the 6G-EARN service, our main findings are:

1. Aggregation of data from multiple sources as per Federated learning and with integration of external resources improves prediction accuracy by more than 100% (in cases of limited available data per client).
2. Forecasting the daily consumption of a household can be sufficiently accurate, with mean absolute percentage error MAPE <20%
3. Aggregating consumption data from too diverse data sources may have the opposite effect, and hence careful grouping of the users is necessary.
4. 6G-EARN training can be completed in less than 10 minutes even without GPU availability at end-clients

Regarding the 6G-SANDBOX experimentation platform, our main findings are:

1. 6G-SANDBOX supports the deployment of any properly-designed cloud service and experiment with a diverse set of simulated Trial Networks scenarios.
2. The connectivity and computation requirements of Energy communities are adequately captured by 6G-SANDBOX and 5G.
3. Clear separation of the control and user planesin a 5GS deployment is necessary.
4. With the cloudification of the 5G system, competition for computing resources between 5G and the hosted services should be considered and avoided.
5. For compute-intensive services (such as the training phase of 6G-EARN), monitoring of the cloud infrastructure and flexible allocation of compute resources are critical