Smart Grid Challenges
John Duff
Landis&Gyr

 

The concept of the electric "smart grid" is all in the news these
days, but we have been working on making it a reality for years.
At Landis+Gyr, we have a mix of technologies that stream meter data
back to a head end system: RF, GPRS, and PLC (power line carrier),
including both one-way and two-way systems.  Most of this is electric
metering data, with some water and gas data sometimes mixed in,
depending on the utility.  An RF installation may have 4 million
meters, sending back two separate pieces of data at 15 minute
intervals, in addition to a bigger daily payload from each meter.
That's 4000000 * 2 * 4 * 24 + 4000000 = 772,000,000 data packets
in a 24 hour period.  This data gets aggregated at hundreds of
collectors at the edges of the network, and then downloaded and
processed in the head end system.  We may also have a future
requirement to collect reads at 5 minute intervals, instead of 15
minutes.  A PLC system, which is usually used in rural areas to
send the data back over the power lines themselves at a slower data
rate, could have 500,000 meters sending back hourly data in addition
to a bigger daily payload from each meter, for a total of 12,500,000
data packets per 24 hours.  And there are other technologies that
we are implementing, such as Zigbee support for in-home displays,
where the user can view their energy usage and pricing, and opt-in/out
during peak load periods where the utility needs to shed some load
on the grid.  That involves additional data we will need to pump
through our network.

Our challenges involve designing the head end and collector systems to
process, store, act, and report on these huge volumes of data.  We also
have to integrate with third party systems such as outage management and
engineering diagnostics systems.  Our current architecture is based on
an SOA platform to communicate with the collectors and third party
software using web services, coupled to a SQL Server or Oracle backend
database.  This has been adequate, but as we scale up our networks and
integration efforts, we decided it wouldn't satisfy our needs.  We are
exploring alternatives such as an enterprise service bus architecture to
replace the SOA system, and also CEP (complex event processing) to react
in realtime to the datastreams.  We need to support the ever-increasing
volume of data that we are gathering from our networks, perform our own
analytics on it, and also continually make it easier for third parties
to connect to our systems to receive our data streams (raw and/or
filtered).  Utilities commonly require SQL Server or Oracle backend
databases, but I am exploring alternatives such as Apache's Hadoop+Pig
as well.   Rather than vertically scaling up the big iron to handle the
data, I'd rather horizontally scale it out.