Sense demystify the data demands of grid edge computation

Role of the cloud

Cloud computing does have an important role to play in supporting NGSMs. Whilst tasks such as real-time device detection need to be performed at the meter using the high-resolution data feed, these detectors are very house specific and need to be learned over time. For that reason, pre-processed power data from each home needs to be exported to the Sense cloud, and this is readily transmitted over the homeowner’s WiFi.

The key here is that the data transmitted has been pre-processed by the grid-edge computer and as a result, the bandwidth requirements are much more modest than one might expect based on the high resolution of the raw metrology data. In fact, on a typical rolling basis it amounts to 5-10MB per day over WiFi (rising temporarily during the first month of intensive learning), which is a lot less than a typical video doorbell or security camera. These data volumes are directed to the Sense cloud, not DNO systems. Consumers receive real time device updates when they open their app, but DNO’s receive insights on a need to know basis, so the levels of data received by their systems is very manageable. Event signals need to be real-time in order for them to be of use to the homeowner (or DNO or retailer) but the processed power data for AI learning is not time sensitive and can be batched in order to avoid WiFi congestion.

Periodically, learnings from the cloud-based AI engines, such as house-specific device detection models, need to be sent back to the grid-edge processor by way of an OTA (over the air) update. Again, this does not have a real-time requirement and so can be performed as and when the WiFi network allows.