IEEE Power & Energy Magazine - May/June 2018 - 40
✔ in a centralized power market, generators are controlPrice
figure 5. The demand/supply curves in a local P2P energy
market (Li, 2015).
* Probability forecasting: This will serve to estimate
the likely energy (both generation and consumption)
over the bidding periods from seconds to minutes to
hours. Critically, it will estimate the probability distribution of each supply and demand level for each
bid period to measure the uncertainty of the estimate.
* Ultrafast settlement. Because there is potentially significant uncertainty between predicted energy and real-time trading, the associated surpluses or shortages
would need to be reconciled and settled. depending
on the size and time block of the p2p market, big data
analytics could support fast trading and settlement.
Market Operation: Forecasting, Pricing, and
during the market operation stage, all potential offers and bids
would rely on real-time forecasting and pricing. The trading
system would then find the best matches between supply and
demand. There is a particular need to improve forecasting,
pricing, and matching algorithms to take into account products, flexibility, and uncertainties, and, ultimately, maximize
the value of local resources.
real-time forecasting, pricing, and matching are interdependent, which requires highly efficient data processing to
cope with a potentially very large number of offerings. The
quality and speed of forecasting are prerequisites for the
quality of pricing to ensure that the market reflects the availability and condition of the offerings. Forecasting and pricing
will then determine the quality of matching, i.e., the degree
of value that can be delivered to local energy resources.
energy forecasting techniques thus play a critical role
in the traditional centralized market. The merits of these
techniques are challenging at local markets for the following reasons:
✔ at the local level, electricity demand is volatile and
difficult to predict. at an aggregated level, the diversity between customers makes aggregated demand for
a centralized market easier to predict and correlate
with explanatory variables such as days of the week
ieee power & energy magazine
lable and can be used to balance demand forecasting
error. in contrast, distributed generators are usually
intermittent and uncontrollable. The two-way forecasting errors of generation and demand would bring significant uncertainty to the settlement of a local market.
✔ The deployment of low-carbon technologies will significantly challenge existing forecasting models. For
instance, the established average load profiles will become inappropriate when individual households are
equipped with pVs, own eVs, and individually have
widely varying demand profiles.
in particular, all forecasting techniques rely on the selection of dependent or explanatory variables that are then used
to derive algorithms to forecast demand. To support the local
market, the selection of variables has to go far beyond the current searching space, which is based on engineering experience. This is because, at the granular level, individual demand
and generation are markedly influenced by the explanatory
variables. For example, the output from roof solar energy is
mostly determined by meteorological factors, while household demand is heavily influenced by geodemographic factors; life patterns, driving behaviors, and charging locations
are likely to be key influencing factors for eV demand.
examining all possible internal and external variables would
be far too laborious and time-consuming for real-time applications. instead, local energy markets require a new breed of
forecasting technique that can quickly identify dominant variables from large volumes of input data and make fast predictions with a reasonable degree of confidence. essentially, new
variable selection and forecasting techniques are required to
predict key energy information concerning energy availability
and variability in a timely manner, thus informing all subsequent market operations, i.e., pricing and matching.
at the real-time trading stage, automatic matching between
intermittent generation and flexible demand would maximize
the value of local resources. Coping with a very large number
of transactions over time and locations requires large-scale
and efficient searching algorithms that consider physical
uncertainty in forecasting and financial risks in pricing. To
this end, prosumer segmentation would enable fast matching processes by controlling similar prosumers as a group to
speed up the search algorithms.
Shared Network Access
Sna aims to integrate flexible demand in a cost-effective manner. The major benefits of Sna over conventional business
models can be seen by examining the commercial and cash
flow relationships among stakeholders. more importantly, as
shown in Figure 6, the Sna scheme provides an incentive to
the incumbent dno to give up its exclusive access to the network and lease the spare or backup capacity to licensed independent parties. The ownership of assets would be retained
by the incumbent dno, but competition would be introduced
through auctioning or contracting out the rights to the spare