We have seen this movie before. Its called "Back to the
Future". In the mid-80s, the second edition of the movie series
predicted flat screen TVs, fingerprint scanners, video conferencing and
flying cars in 2015.
However, the first edition of the movie was the most
memorable one going back to the past from 1985 to 1955.... But now that
we are in 2015, lets take a dive back to the mid-80s when the movie was
released... Incidentally, its also the time when there was another
movie playing in real life in technology: called the
Telephony-to-Internet saga.
A worldwide sprawling synchronous infrastructure called the
telephone network sized for peak demand, i.e. capacity sized to serve
the largest amount of simultaneous demand (with very high probability).
When a phone call is made, instantaneously an end to end "circuit" and
its associated capacity is reserved to allow the call to be made. A
shift from analog to digitization of the telephony infrastructure:
time-division multiplexing (TDM), optical transport of large bundles of
time-synchronized digital information (SONET). The emergence of a few
new applications (fax, email) that rode on top of the digitized
infrastructure as an overlay network. The emergence of a small amount of
buffering or storage capacity to temporarily store "packets" of demand
at the overlay nodes (both at the source and intermediate "routers"),
supporting the idea of "packet switching" which admitted asynchrony and
allowed the capacity of routers to be sized above average demand, but
well below peak demand. The emergence of end to end decentralized
control algorithms (carrier sense multiple access & randomized
controls in Ethernet, decentralized flow control in TCP) allowed demand
to be responsive and shaped dynamically to match the actual capacity on
the paths (instead of reserving peak capacity as in telephony
circuits). Application services like email which appeared to be worse
than the current highly reliable telephone voice service, but much
better and quicker than snail mail and memos. And finally the emergence
of HTTP, web browsers and the WWW. The rest, as they say, is history.
Recently Whatsapp allows users to make calls via mobile devices to other
mobile device users, tied to their phone numbers, symbolizing how voice
has transitioned to another application over the Internet.
If you look back at the abridged history, the three pivotal
technologies that underpinned the Telephony-to-Internet transformation
were:
(a) digitization of infrastructure that allowed overlay applications over a synchronous infrastructure sized for peak demand
(b) the introduction of buffering (i.e. storage), and the
notion of asynchronously switched "packets" instead of
time-synchronized switching of bits (or bit-bundles)
(c) the emergence of decentralized controls (embedded in
Ethernet, TCP/IP, and inter-domain routing policies) that allowed demand
to dynamically respond to capacity available
Peak Energy Techno-Economics
While the analogy is not perfect, the electricity grid is
displaying a number of similarities so that we can selectively learn the
right lessons from history. For instance, to understand the equivalent
analog of (a) we need to appreciate the economic implications of a
synchronous infrastructure designed for peak demand and what degrees of
freedom allow us to "overlay" flexible supply/demand sources on it. When
we look at graphs as the picture below which show rapid growth of clean
energy or renewable options (eg: solar / wind) in the future, it is
important to understand the nature of these sources (solar, wind) are
fundamentally different, amenable to IT-driven management and some
lessons from history could be valuable as analogs.
The electricity grid is sized for (an estimate of) peak
demand, and operates synchronously, i.e. when you turn on a switch for
your lamp, a signal via grid frequency is instantly conveyed to remote
electricity generators which spin up or down slightly to supply your
lamp's needs in real-time. In countries like India, where even in urban
centres, often there is not enough supply available (or economically
contracted by the utility) to meet demand, customers have to bear a
power cut. This power outage is often unscheduled, and the situation is
worse in rural settings, where either the grid does not even reach them
or even if it does, power supply is available only for a few hours of
the day. The assets deployed both on the grid side and consumer side are
essentially idle during power cuts, a huge opportunity cost. A
peak-demand sized infrastructure is quite expensive, compared to an
infrastructure that could be sized somewhere between peak and average
demand (and the difference is managed smartly). This "peaking" effect is
also reflected in wholesale spot prices of electricity in such markets,
where peak spot prices tend to be 5-10X costlier than at other times
(which is why it is economized on by utilities). Utilities also maintain
spinning generation reserves based upon natural gas or diesel to handle
peaks and they are very expensive, and used only for a few hours in the
day.
The need for peaking capacity has only intensified over the decades
and is being further intensified due to the uptake of renewables. The
illustration below shows how the demand in ISO-NE (New England in USA)
has evolved. Think of it as a frequency histogram, and it is saying that
the top 500 hours (top percentile of load) occurs with very small
frequency, i.e. a few hours or days for the entire year, but spinning
reserve generation capacity, transmission and distribution capacity
(poles, wires, power systems) has to be peak provisioned 24 x 365 to
ensure reliability. This is why the tarriff structure in many regions
(eg: California) is tiered to discourage peak demand (or large levels of
demand, which correlates with users who contribute to peak demand).
This is also why enterprises (commercial/industrial energy users) pay
"demand charges" by KW-peak, i.e. peak-power they consume (to reflect
the grid sunk investment costs they drive), even if they cross that
level only for a few minutes in a month.
Overlay Technologies for the Energy Infrastructure:
If we can introduce "overlay" technology that is cheaper,
but allows the offsetting of this peak demand, its economic value would
be the capex/opex saved by offsetting peak capacity required otherwise.
We could achieve this either by (i) generating energy spatio-temporally
matched to when/where peak demand occurs, or (ii) storing energy (in
thermal, chemical forms), or (iii) time-shifting ("when") /
space-shifting ("where") demand or supply to arrange the "when-where"
matching of demand / supply (i.e. virtual storage), It is worth
re-emphasizing that the economically relevant comparable at the margin
is not average energy price, but the peak (or tiered) energy price
applicable to that marginal unit of energy used.
Consider method (i) where we overlay renewable generation
matched to consumption. Solar energy generated at homes in hot locations
(eg: Arizona, Middle East, India etc), tends to roughly coincide with
peak demands for energy without any further intervention. In contrast
wind energy that blows faster at night, and is remote (i.e. it has to
compete at wholesale, not retail prices) may be less valuable
economically purely from a timing/coincidence of demand perspective.
This relationship of economic value to coincidence of supply/demand
means that it is more valuable to have a low capacity-factor generator
like Solar (i.e. producing on average less than its peak rating) than a
higher capacity factor resource like Wind which may not be in production
coincident with peak demand. However it is important to temper this
point noting that as Solar penetration increases, the peak net demand
will shift to the evening time (also called the "duck curve" effect),
and the value of an overlay technology like wind or storage may be
higher in that context.
The second pattern (ii) is "overlay" technology such as
battery storage (or other forms of energy storage like thermal storage).
Tesla Energy recently announced a 10 kWh PowerWall battery for homes
priced at $3500 for 10kWh, or $350/kWh (wholesale price prior to
installation/inverter costs). Given its 10 year warranty, 365 days/year,
or approx 3500 cycles, which implies a simple levelized cost of
350/3500 = 10 cents / kWh LCOE (ignoring other costs and discounting
cash flows). If the user has sunk costs in solar at home, then at the
margin, time-shifting the solar energy to offset peak electricity prices
would be attractive if this arbitrage was worth at least 10 c/kWh. The
ability to provide other services (eg: backup power during outages) is
not factored.
The third pattern (iii) is where demand and supply do not
coincide, and beyond batteries, a set of predictive analytics and
control can be used to match them spatially ("where") and temporally
("when"). IBM Research in partnership with clients has
pioneered a number of cloud-based insights capabilities for
demand/supply management for utilities via the Smarter Energy Research
Institute (SERI). On the decentralized demand response front (i.e. the ability
to make appliance energy demands flexible), IBM Researchers developed an
innovation called nplug motivated by power-cuts in India which allows
at a plug level the ability to shift demand automatically to match
capacity (as implicitly inferred by analyzing grid voltage) without need
for any price signal, communication or coordination protocol with
centralized entities. This was also inspired by the randomized
distributed control methods used in Ethernet, but adapted to the grid.
As we can see in all these examples of "overlays", information
technology (analytics, optimization, controls) needs to be interwoven
with energy technology (solar, wind, grid, battery), and with an
awareness of the economic, policy, technical context, the entire
"overlay package" should perform the "when-where" matching of demand and
supply.
Energy Storage: Ability to Absorb and Manage Uncertainty
Lets move on to (b), the impact of energy storage beyond
the "peak-demand" driven economics of synchronous infrastructure. To
understand this, we need to come back to the shift from
circuit-to-packet switching. Lets ask the question: "What is a telephone
"circuit" really?" A circuit uses signaling at call set up and locked
down (i.e. reserved) capacity end to end. From a queueing perspective
(see diagram below) this set up a D/D/1 queue time synchronously, i.e.
deterministic input feeding a deterministic output. Elementary D/D/1
queueing analysis states that if capacity is matched to (peak) demand,
the amount of buffers is just 1 unit (i.e. effectively zero). If there
was no capacity available at the time of a call, a circuit could not be
established in signaling, and a call would be rejected (similar to power
cuts in India or rolling blackouts when there is no capacity; and huge
spikes in wholesale market price for capacity 10X greater than normal
during such demand spikes). The telephony "switch" uses
time-synchronization to avoid the need for buffering, and literally
switches bits from its input to its output port instantaneously. Packet
switching's fundamental abstraction - a set of bits, with a header
allowing it to be self sufficient - allowed those set of bits to be
buffered at "routers" or "caches" or "storage" more generally.
Here is the key insight: Buffers or energy storage imply that the queueing disciplines could admit random arrivals and random departures; and packet switched networks allowed flexible networks of such queues M/M/1 being the simplest. However if the randomness
of demand/supply could be "shaped" (eg: by shaping the statistics,
truncating tail behavior) or managed/controlled end-to-end (or even
locally) to match demand/supply , the residual externality (or
mismatch) can be minimized. It is important to realize that with good
matching, the "peak" not does not just shift around, but it is
attenuated through clever spatio-temporal matching and smoothing. The
combination of storage and smart controls/optimization driven by
predictive analytics will allow energy storage to penetrate faster and
have a greater transformative effect on the grid. The degrees of freedom
for such uncertainty management, shaping and "when-where" matching of
demand/supply are numerous and specific to the nature of individual
technology options (renewables, storage, grids (DC/AC), power
electronics) which allows a myriad set of formulations and contexts. But
the demand-supply management and matching problem under uncertainty,
and with energy storage is fundamentally an opportunity for information
technology.
Decentralized Control / Management: Towards an Internet-of-Energy-Networks
The third important lesson from the Telephony-to-Internet
saga is the importance of de-centralized controls/management. At a basic
level, the decentralized controls can help locally shape the nature of
randomness of demand / supply and optimize the amount of storage or
external grid or policy support needed to accomplish the matching. We
have seen this earlier when we mentioned the nplug technology for
distributed demand response inspired by Ethernet-style controls. But
beyond this, decentralized controls along with modular technology at
lower costs (eg: as is happening with distributed solar, battery,
demand-management systems), fundamentally empower the end-user or
customer to take control of their choices (in this case energy choices).
Remember the AT&T monopoly of the telephone network that got
transformed into an interacting network of autonomous systems (with many
market participants)? We are seeing this happen in the Energy ecosystem
both at the home level (with the explosive growth of distributed
solar), and at the commercial / industrial level (eg: Apple's recent
$850M investment to procure 130 MW solar power from FirstSolar, or
Walmart's announcement to cover all its roofs to generate over 100MW).
Note that in the context of enterprises with distributed operations, we
could imagine a cloud-based service to manage the energy resources
across sites of say, Wal-Mart.
Beyond these "self-service" models where a consumer
does-it-all by themselves, we are also starting to see the rapid
emergence alternative energy service companies like SolarCity, SunEdison
etc who package installation and management of multiple energy
resources (solar, battery, demand response (eg: NEST)) & outage
tolerance possibly in microgrid configurations, with a package of
financing and policy incentives as well. The notion of microgrid is
important because it allows the emergence of "autonomous systems" that
can interconnect. Microgrid "domains" will serve to take control of
energy choices (storage, renewables, demand management, DC/AC grid
choices) within the domain, but will be interconnected to other
microgrids and to utility grids. This is akin to the emergence of
"Internet Service Providers" (ISPs) in the 1990s like AOL, Excite@Home
and others who overlaid their internet services or access services via
modems atop the existing telephony infrastructure, and offering email
etc. Once this trend gets established and the new service providers
grow, they will necessarily need to interconnect with the existing
service providers and with each other. This tends to drive the market
structure from a single switched network to a "network" of networks
model, which was the genesis of the term "Internet". This interaction
needs to be governed so that one provider takes responsibility for the
choices they make and the externalities they impose on other providers:
this is managed through inter-domain routing protocols on the Internet.
When you examine some of the higher penetration solar regions like
Hawaii, we are starting to see externalities and instabilities being
imposed by solar on the grid operator (HECO), and the need for urgent
solutions to manage this via a combination of energy storage and
distributed controls. It is useful to emphasize here that the notion of
users defecting en mass from utility grids is a short sighted view (as
much as local area networks and enterprises defecting from wide area
networks in the Internet): the internet transformation has taught us
that interconnection by itself has greater value.
Information technology has a huge role in defining and
managing this set of uncertainties and governing the economic spillover
effects, again closely integrated with the technology options at the
energy level (renewables, grid, energy storage) and financial/policy
levels. It is important to note that "decentralized" means that no
single entity has overriding control/bargaining power, but the
granularity of decentralization could evolve as a function of
technological and economic constraints. For instance, managing a large
number of commercial sites could be done via a cloud-based service, but
under the autonomous management of an entity like Walmart with their own
selection of vendors. Similarly, data centres and cloud providers could
choose to go towards renewable-powered DC grids locally; or the
emergence of micro-grids that are DC-based typically in a community- or
corporate-ecosystem setting (ranging from Rural Electrification plays to
managed EV-fleets such as e-Bike/e-Scooter share programs, e-Taxi
fleets, or e-Logistics fleets etc). A large range of integrated energy
products that embed IT / financial innovation (eg: solar lights with
e-financing/payment via mobile phones in Africa to a distributed fleet
of e-transport options that personalise public transportation, with the
ability to space-time shift renewable energy).
Analog of Internet's Fiber-optic Transformation in Clean Energy:
Finally, one phenomena we saw in the Internet era was that
the new ISPs built their own long distance networks, driven by falling
costs in optical networking. When renewables can be firmed up with
storage, and the costs of solar/wind/storage continue to drop, while
land costs start to become important, and mismatches in time and space
matter, it may become economical to locate renewable farms in places
like deserts in other countries (eg: Australia, Middle East, Africa),
and have "cheap" long distance transmission networks "wheeling" the
firmed renewable power to population centres. A few of these concepts
(eg: DesertTec, and ours) are illustrated below: the basic idea is to be
agnostic to the specific combination of renewables (subject to economic
/ technical feasibility) and build large interconnection links that
span time zones - especially in the east-west directions so that
time-of-day differentials will create a market (eg: India's morning
power supplied by Australian sun, and evening power supplied by a
combination of Australian wind and Middle Eastern sun. The new utilities
like SunEdison, NRG etc will also start investing in their own
transmission networks just like in the Internet era. We will first see
this kind of transmission networks within countries; multi-country grids
are being deployed in Europe currently as well.
No comments:
Post a Comment