Demand Forecasting in the

Mohammad E. Khodayar holds a Smart Grid Paradigm: Features

B.S. degree from Amirkabir
University of Technology, Tehran, and Challenges
Iran; an M.S. degree from Sharif
University of Technology, Tehran,
and a Ph.D. from the Illinois Institute Demand forecasting faces challenges that include a large
of Technology, Chicago, all in
Electrical Engineering. He was a
volume of data, increasing number of factors that affect the
Senior Research Associate with the demand profile, uncertainties in the generation profile of
Robert W. Galvin Center for
Electricity Innovation at IIT. He is
the distributed and renewable generation resources and
currently an Assistant Professor with lack of historical data. A hierarchical demand forecasting
the Department of Electrical framework can incorporate the new technologies,
Engineering, Southern Methodist
University, Dallas, and an associate customer behaviors and preferences, and environmental
editor of the IEEE Transactions on factors.
Sustainable Energy. His current
research interests include power
system operation and planning. Mohammad E. Khodayar and Hongyu Wu
Hongyu Wu is a Research Engineer
at the Power Systems Engineering
Center at the National Renewable
Energy Laboratory (NREL) in I. Introduction expansion planning of the
Golden, Colorado. Prior to joining generation and transmission assets
NREL, he was a Visiting Scientist at With the advent of deregulation in power systems is determined
the Robert W. Galvin Center for in the power industry, demand based on the provided forecasts on
Electricity Initiative at the Illinois forecasting has gained electricity demand (de la Torre
Institute of Technology. His research
considerable attention in energy et al., 2008). In long-term power
interests include optimization of
large-scale systems, power systems market operation and planning. system planning, overestimations
operation and planning, distributed Demand forecasting impacts in demand would result in higher
renewable energy integration in the the operational decisions on investments in generation and
smart grid, and smart home/building. generation scheduling, reserve transmission expansions, while
and reliability margins, as well as underestimations could lead to
the maintenance scheduling in deficiency in generation supply
electricity networks. Moreover, (Stremel, 1981). Similarly, in

July 2015, Vol. 28, Issue 6 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 51
 short-term power system environmental effects, have further aggregated to provide the
operation, overestimations in dramatic impacts on the models distribution network demand
demand result in an increased developed for demand forecast in profile. Several guidelines for
number of committed units to the electricity networks. investment in generation and
serve the load, which translates
into a higher system operation cost,
whereas underestimations could
I n the smart grid paradigm,
demand forecasting plays a
crucial role, as it applies to several
distribution infrastructure were
developed using information
extracted from real-time
result in deficiency in electricity practices, such as demand-side observations (Park et al., 1999;
supply and cause catastrophic management (DSM), renewable Firmo and Legey, 2002).
events as a result of real-time energy integration, and energy Energy consumption patterns
imbalances between generation storage planning and scheduling. are dependent on time, weather
and demand (Nogales et al., 2002). Smart grid deployments, such as conditions, and random behavior

E xtensive research has

addressed the challenges in
short-term and long-term
energy management systems and of customers and are further
controlled by energy
management systems to ensure a
As the number
demand forecasting in power certain comfort level for
systems considering the
of factors that customers. Examples in
interdependence of the demand contribute to residential and commercial
patterns with the spatiotemporal demand forecasting facilities include heating and
behavior of the consumers and increases, the present refrigeration, space cooling and
weather conditions (Shu et al., heating, and air conditioning. In
2009; Hong et al., 2013; Nose-Filho
framework for the smart grid paradigm, several
et al., 2011; Song et al., 2005; processing the data technologies contribute to
Taylor and Buizza, 2002; Bo and requires further variations in the demand pattern,
Li, 2009; Fay and Ringwood, revision. including renewable energy
2010). The forecast models resources, distributed generation,
developed for utilities over the smart appliances, storage
past two decades were associated advanced metering infrastructure facilities, and electric vehicles
with decent economic conditions, (AMI), provide more efficient (EVs). As the number of factors
which leads to slighter volatility real-time control to balance that contribute to demand
in the load data and parameters energy supply and demand while forecasting increases, the present
affecting customer demand. offering improved visibility for framework for processing the
However, these safe margins electricity supply and demand data requires further revision. In
were eliminated as the grid patterns. AMI provides a tool for order to provide more accurate
approaches its capacity limits, consumers to understand the and efficient demand forecasts,
congestion becomes more correlation between demand and several challenges need to be
frequent and public awareness of the price of electricity, and offers a adequately addressed including:
greenhouse gas (GHG) generation better insight into practices in (1) processing large volumes of
pushes the generation portfolio distribution networks such as data offered by smart meters
toward more renewable energy demand response, pricing corresponding to energy usage;
resources. The development of scheme, and expansion planning. (2) addressing shorter time spans
smart grid projects and energy The widespread development of for demand forecasting, i.e., five
efficiency programs, as well as smart meters provides frequent to 15 minutes; (3) identifying
increased customer awareness on and continuous readings on various elements that impact the
the energy consumption and demand patterns, which are demand forecasts in the new

52 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 The Electricity Journal
 paradigm such as climate change, which affect the availability of the methodologies. For example, for
renewable energy integration, generation and transmission price-responsive demand and EV
pricing schemes and economic assets as well as demand patterns. technology, it is expected that
incentives, government policies, The demand pattern in the most of the charging processes for
and technology developments; scenarios addresses how residential EVs take place at
and (4) addressing the dynamic customer demand responds to night, which will eventually
behavior of the customers in climate changes in severe flatten the demand profile in the
response to technological weather conditions. Developing distribution network. In addition,
development, environmental such frameworks would help integration of distributed
concerns, and economic incentives. utilities to plan ahead for better renewable energy resources

T he needed tools to address

the above challenges are
required to: (1) process large
reliability and resilience of
energy supply in the electricity
introduces uncertainties to net
demand, defined as electricity
demand minus renewable
networks.
volumes of historical data in short generation. As a result of the
time spans; (2) develop new stochastic nature of renewable
frameworks for demand The increased number of energy sources, the uncertainty,
forecasting when the customer elements considered in volatility, and intermittency of net
behavior is transformed as a demand forecasting can demand are much higher
result of new practices and compared to those of electricity
technologies in the energy
be captured by demand, leading to higher
industry, and when the historical developing a set of ramping requirements for
data may no longer represent the scenarios reflecting the generation assets to follow the net
behavior of consumers; and (3) social, economic, and demand in real time. More
derive the elements that affect specifically, the elements that
demand and contribute to energy
environmental states of ought to be considered in demand
management practices in the the system. forecasting under the smart grid
smart grid paradigm. The paradigm are as follows:
increased number of elements 1. Weather: Accurate weather
considered in demand forecasting II. Key Factors in forecasts lead to more precise
can be captured by developing a Demand Forecasting renewable generation such as
set of scenarios reflecting wind and solar, and demand
the social, economic, and The demand profile in the forecasts in the distribution
environmental states of the smart grid deviates from network. The role of the weather
system, under which the demand historical patterns as a result of is further highlighted once the
forecasting framework is integrating communication, customers are experiencing
developed. In general, the monitoring, and control extreme conditions, in which a
scenario-based forecasting model technologies at the customer slight change in the weather
requires enterprise-level tools to level. These technologies include condition would result in
provide flexibility, scalability, EVs, smart and price-responsive dramatic changes in the comfort
accuracy, and dexterity for both appliances, and renewable energy level and electricity demand.
short-term and long-term resources implemented close to Figure 1 shows the correlation
demand forecasting [SaS White the demand side, which change between the demand and the
paper]. Examples of such conventional demand patterns temperature in an extreme
scenarios include severe weather and increase the complexity of condition. The demand for
conditions in electricity networks, the demand forecasting electricity in Texas increased to

2015, July Vol. 28, Issue 6 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 53
 [(Figure_1)TD$IG]

renewable energy generation

portfolio. The integration of new
technologies in the smart grid
such as EVs and energy storage
devices, which have several times
larger demand capacity than
residential loads, would
Figure 1: Electricity Demand of the ERCOT on Jan. 7, 2014 contribute to the magnitude of the
disturbances in the demand
57,277 MW at 8:00 a.m. on Jan. 7, provides the headroom that pattern. The random disturbances
2014, as heaters were turned on affects the demand profile in may result in sudden imbalances
when the temperature reached its response to the changes in the between generation and demand,
lowest level. The temperature in weather conditions. which can be compensated by
the Dallas/Fort Worth area was 2. Temporal factors: Electricity using fast-response generation
21 8F, as shown in the figure. As consumption is dependent on technologies. Deploying
the temperature increased later in time of day, day of the week distributed generation (DG) with
the day, demand decreased once (weekday, weekend), the type of intelligent control mitigates such
heaters were turned off. day (holiday, special event) fluctuations locally and reduces
Several factors affect the comfort as well as the season. The the required load following and
level of residents including demand-side management and regulation services in the
clothing, intensity of activity, pricing mechanism in the smart wholesale electricity market.
individual metabolism rate, and grid can limit the fluctuation of 4. Electricity price and demand
environmental conditions such as demand and the difference response: In regulated electricity
ventilation and air velocity, as between peak and off-peak markets, the electricity price is
well as the temperature of the periods. The new demand pattern fixed to recover the cost of energy
surroundings. Among others, the is much flatter than the traditional generation, transmission, and
environmental conditions are demand pattern in the electricity distribution, while in a
controlled by the energy grid, which reduces the temporal competitive market structure,
management systems. The impact on the intraday demand price is determined based on the
outdoor temperature affects the forecast profile in the smart grid. balance between demand and
operative indoor temperature; However, factors such as seasonal supply. In the latter paradigm, the
however, a small deviation in differences, climate changes, consumers respond to variations
indoor temperature or air speed weekends and holidays would in the price of electricity by
will not affect the thermal comfort still impact the value of demand. adjusting their demand, which is
of the occupants. Switching on air 3. Volatility in generation and in turn determined by the pricing
conditioning and refrigerating demand profiles: Random schemes. These pricing schemes
systems will not impact the disturbances in consumption and ranging from time-of-use (TOU)
indoor temperature and humidity generation are consequences of to real-time pricing are also
immediately because of the heat customer behaviors and the referred by terms such as
storage capability of the building. availability of renewable energy time-differentiated retail pricing,
Therefore, the electricity demand resources, respectively. The time-variant pricing, and
of air conditioning and fluctuations in weather advanced pricing programs. TOU
refrigeration could be regulated parameters including cloudiness, pricing scheme will increase the
to provide demand response wind speed, temperature, and incentives for the customers to
services and the storage capability pressure will impact the shift peak load to off-peak

54 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 The Electricity Journal
 periods. Recent advancements in resources will impact the extent of and demand. Such technologies
metering and energy these services. The availability of increase the net demand
management technologies have these resources is dependent on uncertainties and increase net
increased the appeal of real-time such developing technologies on demand forecast errors; however,
pricing and demand response the demand side as smart the level of uncertainty is
practices. Such practices could appliances and energy storage dependent on the generation
reduce the cost of serving demand facilities behind the meter. profile and its correlation with the
in the electricity market, reduce 5. Renewable energy resources: demand profile. For example,
the GHG of generating electricity The smart grid promotes the some aspects of solar energy
and change the demand forecast integration of renewable energy fluctuation including sunrise and
models as they provide incentives resources, primarily wind and sunset are predictable, the
to reduce the demand once it solar generation. Wind and solar generation profile is close to the
exceeds supply. By integrating energy are reliable sources for demand profile as the peak occurs
distributed energy management during the day and the
systems in the distribution There is a strong intermittency is affected by the
networks, these practices can cloud coverage in the area. Unlike
complementarity
limit the costly generation solar generation, the wind
expansion plan to serve peak between wind generation pattern is more
demand in short periods. As an and solar volatile and intermittent and may
example, by implementing resources even not coincide with the demand
demand response in California, pattern. However, there is a
though the level
demand was reduced by strong complementarity between
1,500 MW, i.e., 2 percent of total
of such these two resources even though
statewide demand, at peak hours complementarity the level of such complementarity
during the summer of 2006. depends on location. depends on location (Li et al.,
Consequently, no rolling 2009). Moreover, aggregation of
blackouts were ordered during such resources would result in the
peak demand periods that electricity, are deployable at small reduction in the intermittency
summer. Analogously, ISO New scale, and can diversify our and volatility of the overall
England, Exelon Corp., and nation’s energy portfolio. The generation profile. Other
Northeast Utilities reduced their available generation resources renewable generation
respective demand by 530 MW, close to the demand side will technologies including hydro and
1,100 MW, and 95 MW by affect the net demand portfolio biomass could provide more
implementing demand response that is served by utilities. Any stable generation profiles.
practices. The total 37,500 MW deficiency in supply as a result of Generation patterns of renewable
potential resources contributed volatility in renewable resources resources including small wind
by demand response would have can be compensated by electricity generation and rooftop solar
dramatic impacts on demand purchased at a higher price which panels rely on such weather
forecasts (Federal Energy is provided by more expensive conditions as solar insolation,
Regulatory Commission, 2006). dispatchable generation assets in cloudiness of the sky,
Integrating demand response will the real-time electricity market. temperature, humidity, and wind
change demand forecast schemes Hence, developing renewable speed profile. More accurate
as the correlation between energy resources close to the weather forecasts improve the
demand and electricity price is demand side imposes further generation forecast of such
further highlighted and available imbalances between generation resources.

2015, July Vol. 28, Issue 6 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 55
 6. Energy storage: energy supply but also increase market. As the penetration of EVs
Advancements in energy storage the demand response capability of in the market increases, the
technologies enables customers to distribution networks. Such demand profile eventually
have large-scale, efficient storage storage facilities become more changes as a result of charging
facilities for electricity. Energy popular as a cost-effective solution and discharging patterns.
storage provides flexibility in both to reduce the peak demand and Charging EVs will increase
operations and planning in the charges among commercial and system load while integrating the
presence of the variable renewable industrial businesses. This market vehicle to grid (V2G) technology
generation in the grid. The is growing rapidly, as over can regulate demand in
generation pattern for renewable 850 MW of distributed energy distribution networks and
energy resources is not in direct storage will be annually deployed increase the reliability of energy
correlation with the energy supply. Nonetheless, there are
consumption. Usually, wind several regulatory, technical,
speed is higher at night, when implementation, and market
demand is low. Thus, the wind barriers for developing this
generation is stored in storage technology. Furthermore, the
units and retained for use in peak stochastic nature of customer
periods. The energy storage behavior, including randomness
facilities could help shift the in arrival time, charging time, and
demand profile from peak periods the expected charging demand,
to off-peak periods. The storage would increase the uncertainty in
unit is charged during off-peak the forecasted demand profile.
periods, when demand and The state of the distribution
electricity price are low, and network can also contribute to the
discharged during peak periods forecasted demand of EVs. For
once demand and the price of in the U.S. by 2020 (US Energy example, after prolonged outages
electricity are high. The capacity of Storage Market, 2015). One of the in a network with high
the storage facilities is dependent major challenges corresponding to penetration of EVs as well as the
on the scale of demand and such storage facilities is the limited heat and electricity storage
available renewable resources. information on their operation devices, there is a sudden increase
Utility demand is affected by the schedule. Such units are mostly in the demand profile as a result
amount of storage capacity owned and operated by customers of charging EVs and energy
developed. The energy storage to reduce their peak demand storage facilities. These sudden
facilities would change the charges as well as the cost of peaks should be addressed in the
forecasted demand profile, as the energy, and their operation forecasted demand in severe
operation scheduling of these schedule is not readily apparent to conditions after restoration of the
units relies on the electricity price utilities. Hence, such storage network. If the system is in
and their technical constraints, facilities may impose large errors normal operation, this peak can
including the limitations on in the demand forecast of be avoided by scheduling the
capacity and ramping capabilities. customers unless they are charging cycles, using the V2G
Furthermore, energy storage can monitored exclusively. capability of the EVs and
facilitate demand response 7. Electric vehicles: EVs are regulating demand using energy
practices in distribution networks. considered a favorite alternative management systems. Some
Behind the meter, storage facilities for transportation with minimum studies have indicated that large
not only improve the reliability of dependence on the oil and gas number of EVs can be

56 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 The Electricity Journal
 accommodated without electricity demand according to the energy management of such
increasing the peak demand of the the electricity price signals. facilities can lead to significant
distribution network by Conventional appliances, such as savings for the utility that is
coordinating the charging pattern air conditioners, refrigerators, further shared with customers. In
of the EVs. The peak demand can and dishwashers, could be smart grids, the utilities are
further decrease by integrating converted into smart appliances required to understand the
V2G technology in the system by adding monitoring and control behavior of the customers in
(Rassaei et al., 2014). The price modules. Shifting demand to off- operating their air conditioners,
signals from the utility provide peak hours would yield a pool pumps, TV sets, dish
incentives for the EVs to charge reduction in energy costs with washers, washers and driers,
during off-peak hours, which TOU or real-time electricity water heaters, EVs and other
increases total demand, and to appliances and utilize the gained
discharge (if required) at peak knowledge to provide more
hours, which reduces demand sophisticated forecasting
and price. A study in PG&E methodology, considering the
showed that the TOU program abundance of data inputs from a
influenced the charging pattern of data-rich smart grid environment.
EVs and the customers changed Advances in the technologies
the utility rate subscription as a result in a decrease in demand
result of obtaining EVs (EV, nd). growth as energy efficiency
Moreover, the charging/ measures are improved among
discharging cycles are customers. Increasing the number
coordinated with demand of STAR appliances or more
response practices in distribution efficient LED lighting facilities
networks to reduce the peak contributes to reduction in the
demand and decrease the loading pricing schemes. Over the past growth of energy consumption,
of the transformers in the years, smart appliances with which in turn reduces peak
distribution network, while sophisticated control and demand. This would also affect
maintaining an acceptable monitoring interfaces that could the demand forecast measures in
comfort level for the customers be installed on cell phones and the electricity networks. The
(Shao et al., 2012). computers were introduced by objective of demand management
8. Smart and efficient appliances: several companies; yet the in the industrial sector is to
The residential and commercial smartness in energy consumption maximize the system load factor,
sectors consumed 70 percent of is only associated with lowering and reduce power losses and peak
total electricity generation in 2013 home energy bills. The crucial demand. Similar energy
(Annual Energy Outlook, 2015). need for introducing smart management techniques are used
The developed energy appliances is further highlighted in this sector with corresponding
management technologies in in controlling large appliances. requirements of the facilities.
these sectors bring intelligence to Electricity consumption by 9. Virtual power plants,
energy consumption. In heating, ventilation, and air microgrids and nanogrids: DG
residential level, home energy conditioning (HVAC) and provides new opportunities to
management systems (HEMS) refrigeration systems accounts for serve the demand with low-
communicate with smart home a large portion (50–60 percent) of capacity electricity generation
appliances and smart electric the annual energy consumption technologies placed close to end
meters to help curtail and shift of commercial buildings. Hence, customers. As local generation

2015, July Vol. 28, Issue 6 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 57
 serves demand behind the meter,
the demand forecast schemes
have to incorporate such
the power grid. While microgrids
and VPP share critical character-
istics, they are different in several
N anogrids are small
microgrids that serve a
single load and support a wide
controlled generation resources. If aspects. Microgrids can operate range of applications from
the DGs are renewable energy autonomously by islanding in case emergency to basic power supply.
resources, the stochasticity in the there is an incident in the utility Components of a nanogrid are a
generation profile will make the grid, while VPPs are always controller, loads, storage (if any),
forecasting framework more grid-connected. Moreover, and gateways. Loads in a
complicated, as the outcomes of microgrids rely on storage nanogrid can be any electrical
the net demand forecast are facilities, and are connected to device, commonly in the range of
dependent on the forecasted distribution networks, while VPPs 1–100 W in power demand
renewable energy profile. In (Nordman and Christensen,
smart grids, the virtual power 2013). Electricity usage varies in
plant (VPP) is referred to as a response to the function or state of
group of customers under a the devices. The level of power
specific demand response or supplied to its loads is controlled
pricing program who respond to by the nanogrid controller, which
the energy market as a unified is able to negotiate with other
flexible source. Incorporating nanogrids and microgrids
VPPs allows for treating the through gateways, set the local
demand entities as another electricity price, and manage
operational plant with its internal storage. The load can be
respective attributes, including fully served, partially served, or
maximum capacity, minimum denied by the nanogrid controller.
capacity, ramp-up and ramp- The controller uses the past
down capability. The demand may not have any storage and can energy usage pattern to
portfolios of VPPs are regulated participate in the wholesale determine the load preferences
by leveraging the local electricity market. Hence, and sets the decision variables to
dispatchable generation assets. microgrids can also form VPPs correspond to the supply assets. If
Increasing energy efficiency while VPPs are not necessarily an energy storage facility is
through development of microgrids (Asmus, 2010). attached to a nanogrid, the
combined heat and power (CHP) Integrating microgrids controller stores or withdraws
technologies provides improved provides the required flexibility, electric energy as needed within
flexibility of energy supply for accessibility, and economic and the storage capabilities (Nordman
demand with different reliability benefits introduced by and Christensen, 2013).
characteristics (i.e., electricity and the smart grid and promotes the Nanogrids are considered to be
heat) in the distribution networks. distributed intelligence to operate the building blocks of the more

M icrogrids are defined as

aggregated demand and
generation sources that operate as
the electricity network efficiently.
Providing the monitoring and
control requires real-time
complex microgrids or
utility-dominated smart grids.
Since nanogrids serve a small
a single system with specified information on generation and number of loads, they leverage
boundaries. The monitoring and demand patterns, which further fast local monitoring and control
control assets in the microgrid calls for online adaptation of for energy management. The
facilitate autonomous generation and demand forecast provided control at this level
responsiveness to disturbances in frameworks. could benefit the developed

58 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 The Electricity Journal
 [(Figure_2)TD$IG]

Figure 2: The Hierarchical Structure of Nanogrids and Microgrids in Power Systems [designed by Freepik]

III. Challenges and

demand forecast framework as
demand is being controlled at very Solutions T here are several challenges
associated with forecasting
model development in the smart
fundamental levels with available
generation resources. At the The current forecast methods in grid. A major one is associated
higher level, the microgrid the power network are based on with the large volume of available
conventional generation data for electricity consumption
controllers will perform energy
technologies and passive in small time periods provided by
management to regulate demand
customers. However, in the smart the smart meters (i.e., five to
in the distribution network.
grid paradigm, the framework to 15 minutes), which requires effi-
Accordingly, the forecasting provide demand forecast should cient data processing algorithms
approach for generation and change significantly to address the and effective computing assets to
demand is affected by the customers’ awareness on energy process such large volumes in a
developed monitoring and control consumption and their reasonable amount of time.
at each level. Such controls will participation. The forecast of Another challenge corresponds to
shape demand and impact the energy consumption patterns is the number of elements that affect
forecasting schemes accordingly. performed at three levels: (1) very electricity demand at the distri-
Figure 2 shows the hierarchical short-term load forecasting bution level. Such factors include
structure of the nanogrid and (VSTLF), which has short time environmental, economic, and
microgrid in distribution spans from seconds to hours; (2) social factors that increase the
short-term load forecasting (STLF), complexity of the forecast model
networks. At a higher level, the
which has a longer time horizon and may lead to over-fitting the
forecasted demand is dependent
from hours to weeks, and (3) model. The third challenge is the
on historical data of net demand,
long-term load forecast (LTLF), intricacy in determining which
which is further regulated at lower which has a time horizon from factors contribute to the changes
levels. Developing the energy months to years. Hence, the in electricity demand at each time
management framework at each developed demand forecast models period. An increase in off-peak
level will impact the demand should improve forecasting demand might be a result of fac-
forecasting respectively. efficiency and accuracy at all levels. tors such as charging EVs and

2015, July Vol. 28, Issue 6 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 59
 [(Figure_3)TD$IG]

Figure 3: The Hierarchical Structure of Developing the Demand Forecast in the Smart Grid Paradigm

storage facilities behind the For example, in order to realize impact on the demand and
meters deployed by customers. the customer behavior with generation profile in a distributed
Similarly, a reduction in peak respect to the variation in the environment. The effect of such
demand might be caused by price of electricity, the required factors is less realized in the
discharging storage facilities, historical data to model demand aggregated demand profiles. For
EVs, or the exercise of demand and pricing schemes may not be example, the load profile of an
response. The fourth challenge is commonly available; such defi- independent system operator
forecast errors in measurements, ciencies would be reflected in the (ISO) has a much smoother profile
including temperature, humidity, training sets used to develop the than that of distributed environ-
renewable generation, and the model. ments (distribution network
customer participation level in
demand response. The accuracy
of the forecasted data is decreased
I n short, the emerging smart
grid environment requires
distributed intelligence and new
feeder, microgrid, nanogrid, etc.).
The demand profile in the
distributed environment is more
as more uncertain factors models based on artificial volatile; however, it reflects more
contribute to forecasted values in intelligence and neural networks information on the correlation
the smart grid paradigm. Last but (ANNs), expert systems, fuzzy between environmental factors
not least, the deficiency in the logic, and support vector and electricity demand. Hence,
volume of the data to fathom the machines, for accommodating the differentiating the effect of
impact of different elements and new contributors that affect the environmental factors on
behaviors on electricity demand forecast. The required electricity demand is more
consumption pattern is a big forecasting models address the difficult on the aggregated
challenge to ensure the accuracy climatic, social, economic, and demand profile as it covers
of the developed forecast model. habitual factors that have direct diverse geographic areas with

60 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 The Electricity Journal
 various environmental resources will contribute to the model for shorter time spans as
conditions. net demand profile of the system. the data is not available for such
Figure 3 shows the hierarchical The main feedback from the periods. Consequently, utility
structure of the demand forecast wholesale electricity market to the forecasters need to analyze the
in the smart grid paradigm. It distribution networks and the time span for the forecast model
shows how the demand forecast nanogrid/microgrid level is the in view of the mixed consumption
at each level leverages the pro- price of electricity, which impacts pattern of the two categories of
cured forecast at its lower level. the forecasted demand profile at customers. As the amount of
For instance, the demand forecast those levels. available data in the smart grid is
at the microgrid/nanogrid level quite large, the forecasting model
procures the demand served by should be also able to process
local generation resources; and IV. Transition to the large volumes of data by
the demand profile of the Smart Grid Era providing accurate data filtering
nanogrid/microgrids realized and by pre-processing and using
from the utility side reflects the In order to achieve an a combination of statistical and
intelligence to perform the energy acceptable forecast accuracy, data driven techniques to increase
management at this level. At the demand forecasters try to the accuracy of the demand
higher level, the developed understand the behavior of forecasts. Accurate load
forecasting methods for the customers in response to forecasting using the large
distribution network takes into economic and environmental amount of historical data is very
account the historical data of the factors. In the transition toward critical in building the
customers that are not clustered the smart grid via the hierarchical model for demand
in microgrids, the forecasted implementation of smart meters, forecasting in the smart grid
demand of the nanogrids/ there exist customers with paradigm. The increase in the
microgrids, the available traditional meters in the network. amount of sampling data
distributed generation resources Since the behaviors of these provided by smart meters would
including renewable customers are different from increase the efficiency and
technologies, as well as the those equipped with smart accuracy of the short-term load
topology of the network, all of meters, the demand profile of forecasting. Moreover, an
which correspond to the available overall customers is a increase in the number of
generation and demand resources combination of the demand sampling points would improve
in the distribution network. In the profiles of both types of the forecast model as the data
wholesale electricity market, the customers. Therefore, any correspond to the assets that
net demand profile is determined forecast model should be able to contribute to the demand forecast
by aggregating the demand differentiate the type of were collected exclusively. For
profile of the distribution customers with their respective example, measuring the dispatch
networks and the forecasted characteristics. The time span for of energy storage facilities
renewable energy resources. At recording and processing the installed behind the meters in
this level the demand forecast demand for smart meters is as response to the price of electricity,
performed by distribution small as 15 minutes. Thus, the and collecting data on demand
network operators can be used by forecast model can be developed response capability and behavior
demand aggregators to procure to forecast demand for the next 15, of the customers in response to
the demand profile of the 30, or 45 minutes; yet for the the price of electricity
wholesale electricity market, and traditional meters, it is not independently, would highlight
the forecasted renewable energy possible to develop the forecast the role of each asset and practice

2015, July Vol. 28, Issue 6 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 61
 in shaping the demand profile to incorporate the corresponding with electricity load demand. In: Proc.
IEEE conference on Industrial Elec-
and improve the developed de- characteristics. Such a hierarchi- tronics and Applications, May, pp.
mand forecast models accord- cal structure could be extended to 3623–3628.
ingly. address short- and long-term de- Nogales, F.J., Contreras, J., Conejo, A.J.,
Espinola, R., 2002. Forecasting next-
mand forecasts in the smart grid.&
day electricity prices by time series
models. IEEE Trans. Power Syst. 17
References
II. Conclusion (May (2)) 342–348.
Nordman, B., Christensen, K., 2013. Lo-
Annual Energy Outlook 2015 with pro- cal power distribution with nano-
More advanced demand fore- jections to 2040, U.S. Energy Infor- grids. In: Proc. of the International
cast techniques would result in mation Administration, April 2015. Green Computing Conference, June.
Asmus, P., April 2010. What is the Nose-Filho, K., Lotufo, A.D.P., Min-
improvement in the operation working definition of a virtual pow- ussi, C.R., 2011. Short-Term multi-
and maintenance of electricity er plant? Navigant Research Blog. nodal load forecasting using a
networks, and provide safety, Bo, R., Li, F., 2009. Probabilistic LMP modified general regression neural
forecasting considering load uncer- network. IEEE Trans. Power Deliv.
security, sustainability, and ener- tainty. IEEE Trans. Power Syst. 24 26 (October (4)) 2862–2869.
gy efficiency to the electricity in- (August (3)) 1279–1289. Park, Y., Won, J., Park, J., Kim, D., 1999.
frastructure. Introducing local de la Torre, S., Conejo, A.J., Contreras, Generation expansion planning
J., 2008. Transmission expansion based on an advanced evolutionary
variables that have a direct impact
planning in electricity markets. IEEE programming. IEEE Trans. Power
on the demand and generation Trans. Power Syst. 23 (February (1)) Syst. 14 (February (1)) 299–305.
profile will increase the com- 238–248. Rassaei, F., Soh, W., Chua, K. A statis-
plexity of the forecast models in ‘‘How do PEV owners respond to time- tical modelling and analysis of resi-
of-use rates while charging EV Proj- dential electric vehicles’ charging
the smart grid. Several factors that ect vehicles?’’ [online]. http:// demand in smart grids[37_TD$IF], available
contribute to the demand fore- www.theevproject.com/ [online] arXiv:1408.2320v1 [cs.[38_TD$IF]CE]
casting in the smart grid para- cms-assets/documents/ 11 Aug 2014.
125348-714937.pev-driver.pdf. Shao, S., Pipattanasomporn, M., Rah-
digm include distributed energy Fay, D., Ringwood, J.V., 2010. On the man, S., 2012. Grid integration of elec-
resources installed at the cus- influence of weather forecast errors tric vehicles and demand response
tomer side, storage facilities be- in short-term load forecasting mod- with customer choice. IEEE Trans.
els. IEEE Trans. Power Syst. 25 (Au- Smart Grid 3 (January (1)) 543–550.
hind the meters, and hierarchical
gust (3)) 1751–1758. Shu, F., Methaprayoon, K., Wei-Jen, L.,
measurement and controls in the Federal Energy Regulatory Commis- 2009. Multiregion load forecasting
distribution networks, for which sion, 2006, August. Assessment of for system with large geographical
Demand Response & Advanced area. IEEE Trans. Ind. Appl. 45 (June
the forecasting model at the ag-
Metering, staff report. (4)) 1452–1459.
gregated levels relies on the out- Firmo, H.T., Legey, L.F.L., 2002. Gener- Song, K., Baek, Y., Hun Hong, D., Jang,
come of the forecast demand on ation expansion planning: an itera- G., 2005. Short-term load forecasting
the lower levels. This hierarchical tive genetic algorithm approach. for the holidays using fuzzy linear
IEEE Trans. Power Syst. 17 (August regression method. IEEE Trans. Pow-
structure of demand forecast can (3)) 901–906. er Syst. 20 (February (1)) 96–101.
efficiently addresses the effect of Hong, T., Wilson, J., Jingrui, X., 2013. Stremel, J.P., 1981. Generation system
local variables on the aggregated Long term probabilistic load forecast- planning under load forecast uncer-
ing and normalization with hourly tainty. IEEE Trans. Power Appar.
demand profile in the smart grid. information. IEEE Trans. Smart Grid Syst. PAS-100 (January (1)) 384–393.
The electricity demand is depen- 5 (September (1)) 456–462. Taylor, J.W., Buizza, R., 2002. Neural
dent on the sector (industrial, ‘‘How does forecasting enhance smart network load forecasting with weath-
grid benefits?’’ White Paper, SaS er ensemble predictions. IEEE Trans.
commercial, and residential), and
available online: http://www. Power Syst. 17 (August (3)) 626–632.
the response of demand at each ciosummits.com/media/pdf/ ‘‘US Energy Storage Market to Grow
sector to such environmental and solution_spotlight/sas_forecasting- 250% in 2015’’ [online]. http://
economic factors is different; enhance-smart-grid.pdf www.greentechmedia.com/
Li, Y., Agelidis, V.G., Shrivastava, Y., articles/read/us-energy-storage-
therefore, the forecasting model 2009. Wind-Solar resource comple- market-grew-400-in-2014.
developed for each sector ought mentarity and its combined correlation

62 1040-6190/# 2015 Elsevier Inc. All rights reserved., http://dx.doi.org/10.1016/j.tej.2015.06.001 The Electricity Journal