IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO.

3, AUGUST 2009 1241

Purchase-Bidding Strategies of an Energy Coalition

With Demand-Response Capabilities
Daniele Menniti, Member, IEEE, Ferdinando Costanzo, Nadia Scordino, Member, IEEE, and
Nicola Sorrentino, Member, IEEE

Abstract—The implementation of demand-response programs Thermal dispersion coefficient of the residence, in

(DRP) has gained interest as a means to alleviate energy con- kW/ C.
sumption during peak-hours. Two explanations account for the
success of such programs which involve both utilities and elec- Total thermal capacity of the residence, in kWh/ C.
tricity consumers, with the latter often organized into coalitions:
the system operator meets its goal of reducing the load peak; Parameter equal to .
simultaneously, electricity consumers achieve economic benefits
when reducing consumption during peak hours. In this paper, a A real number representing the unwillingness of a con-
Monte Carlo-based algorithm has been proposed for the formula- sumer belonging to the th consumer-class to trade
tion of multiple purchase offers in the day-ahead energy market living comfort for savings at the th hour, in /( C) .
(DAEM) by coalitions in which consumers vary in their sensitivity
to DRP, manifesting different responsiveness to hourly tariffs Coefficient associated with the volatility of the spot
based on the hourly market clearing prices. Being able to monitor price.
how coalition members use air-conditioning in the presence of
variable hourly energy tariffs, the coordinator can then define
Length of the planning period (in the particular case
a purchase-bidding strategy, depending on how price-sensitive equal to 24 h).
the coalition is. Simulation results show that the presence of a
Number of coalition’s members.
price-sensitive demand leads not only to a subsequent reduction
in energy prices during peak-hours but also leads to a decrease in Number of responsive consumers, i.e., those reacting to
their inter-hour volatility. the energy price volatility of the market, moving con-
Index Terms—Demand-response, inter-hour volatility, power sumption towards those hours where prices are lower.
demand, power system economics, price volatility.
Number of normal consumers, i.e., performing an in-
termediate behavior, if compared with responsive and
NOMENCLATURE nonresponsive consumers.
The notation used throughout the paper is as follows. Number of nonresponsive consumers, i.e., not influ-
enced by price volatility, maintaining fixed consump-
Parameters: tion habits despite the higher prices.
Ideal indoor temperature at the th hour f for th con- Number of simulations in the Monte Carlo-based
sumer-class, in C. algorithm.
Outdoor temperature at the th hour, in C.
Variables:
Lower bound of the indoor temperature at the th hour,
in C.
Energy amount absorbed by the conditioning system of
Upper bound of the indoor temperature at the th hour, the th consumer-class at the th hour, in kWh, at the
in C. th iteration of Monte Carlo algorithm (output variable
Time interval of 1 h. of Monte Carlo algorithm).
Maximum amount of power absorbed from the air con- Sum over peak hours – of the hourly energy
ditioning system, in kW. amounts absorbed by the conditioning system of the th
consumer-class, in kWh, at the th iteration of Monte
Carlo algorithm.
Manuscript received October 30, 2007; revised February 23, 2009. Current
version published July 22, 2009. This work was supported in part by the Italian Sum over off-peak hours ( – and 19–24) of the
Ministry of University and Research under Grant PRIN 2005-2005015592_003. hourly energy amounts absorbed by the conditioning
Paper no. TPWRS-00796-2007.
D. Menniti, N. Scordino, and N. Sorrentino are with the Department system of the th consumer-class, in kWh, at the th
of Electronic, Computer, and System Science, University of Calabria, iteration of Monte Carlo algorithm.
87036 Arcavacata of Rende (CS), Italy (e-mail: [email protected];
[email protected]; [email protected]). Width of the th step, in kWh, of the multiple purchase-
F. Costanzo is with Omnia Energia S.p.A, Cosenza 87100, Italy (e-mail: offer (aggregate demand of the whole coalition), at the
[email protected]).
Digital Object Identifier 10.1109/TPWRS.2009.2023750
th hour.

0885-8950/$25.00 © 2009 IEEE

1242 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009

Width of the th step, in kWh, of the purchase-bid curve is therefore being dedicated to identifying how such energy
of a consumer belonging to the th consumer-class, at crises can be avoided, modifying factors regarding either the
the th hour. supply or the consumption side of the energy exchange process
[4], [6]. In fact, an analysis of the great blackouts which have
Objective function of the air conditioning consumer occurred in the past (in both the USA and Europe) shows that
optimization problem associated with the th con- while massive outages are often caused by a specific event, the
sumer-class. overall power system security is nonetheless strained by severe
Indoor temperature at the th hour for the th con- loading conditions. Fortunately, these extreme system operating
sumer-class, in C. conditions caused by high energy request usually occur only
during summer months, when air conditioning significantly
Indexes:
increases the amount of energy consumption to well above the
Index identifying the generic consumer-class, with average base-load of non-summer months.
meaning responsive consumer, nonrespon- In the past, when electric power systems were vertically in-
sive consumer, and normal consumer. tegrated, security problems, due to line overloading, were re-
solved centrally by the system operator, using programs which
Index identifying the generic hour, . either rescheduled generation or curtailed loads, with curtail-
Index identifying the generic step of a multiple pur- ment involving only large industrial consumers [2], [3]. Nowa-
chase offer, . days, in the unbundled electric power system, peak loading has
been significantly reduced through demand-response programs
Index identifying the generic iteration of the Monte
whereby household and small industrial consumers are orga-
Carlo-based algorithm, . nized into coalitions [7]–[13] who have opted to limit consump-
Miscellaneous: tion during energy peaks, but who, in exchange, enjoy lower
energy costs. Price signals, i.e., variable hourly energy tariffs,
Price generated at the th iteration of Monte have been supported by FERC in the USA and by the European
Carlo algorithm, as a function of the mean value, Commission in Europe [14] as a means to provide incentive
, and variance, , of the market clearing price for consumers to vary their consumption habits. In fact, mod-
at the th hour and of the volatility coefficient . ulated energy consumption in response to variable hourly en-
Mean value of the market clearing price at the ergy tariffs may not only reduce the chances of energy crises
th hour. during peak hours but, for cost-sensitive consumers, such vari-
able hourly energy tariffs also sensitize consumers to the need
Price associated with the th step of a multiple to modulate overall energy consumption, cultivating self-reg-
purchase-offer, at the th hour. ulating habits which economize energy [11], [15]. However,
Variance associated with the market clearing whether a consumer responds to price signals depends on his
price at the th hour. willingness to vary consumption habits. Changing such habits
thus becomes a training process whereby greater possible sav-
Variance associated with the energy consump-
ings are more likely to change consumption habits, inducing
tion of the th consumer-class, at the th hour.
users to concentrate energy use within the hours when prices
Variance associated with the energy consump- are lower [11]–[13], [15]. A direct consequence of this sensi-
tion of the th consumer-class, at peak hours. tized behavior is a reduction in excess loading during peak hours
Variance associated with the energy consump- when prices are usually highest. Interestingly, when demand re-
duces energy usage during peak hours, the energy market re-
tion of the th consumer-class, at off-peak
sponds by lowering tariffs of peak hours [11].
hours.
The reciprocal positive effect that variable hourly energy
Standard deviation associated with the market tariffs have on consumer habits is therefore important in the
clearing price at the th hour. context of energy management. The economic benefits ob-
Expected value operator. tained through consumer responsiveness to variable hourly
energy tariffs as a function of his willingness to vary con-
Variance operator. sumption habits have been reported elsewhere [16]. Here, we
Mean value operator. focus on the formulation of purchase-bidding strategies for the
day-ahead energy market (DAEM) by a coalition of consumers
Probability associated with random variable . manifesting different price-signal responsiveness [10], [17],
examining what motivates changes in consumption habits and
how consumption flexibility may influence market clearing
I. INTRODUCTION prices (MCPs). A model is then proposed for the coalition
ITH electric power systems operating close to their coordinator whose primary responsibility is to minimize energy
W security levels and the impossibility of installing new
lines which would otherwise alleviate heavily loaded circuits
procurement cost, taking into account heterogeneous condi-
tioning habits and market price volatility. To formulate the
[1]–[3], episodes of brownouts and blackouts are increasingly optimum purchase-bidding strategies, a stochastic optimization
more frequent throughout the world [4], [5]. Significant effort problem must be solved by the coalition coordinator who, being
MENNITI et al.: PURCHASE-BIDDING STRATEGIES OF AN ENERGY COALITION WITH DEMAND-RESPONSE CAPABILITIES 1243

able to monitor how coalition members use air-conditioning and small industrial energy consumers was suggested and pro-
in the presence of variable hourly energy tariffs, is therefore moted by academy [11], [13] and government [14] and, rather
able to determine how much price signals actually modify than conceptual, it is an operational practice in Italy [18]. To
consumption habits, i.e., the demand-response capabilities of this end, different synonyms such as load aggregations [11], or
a coalition. cooperative consumers and coalitions [9] were used in the lit-
The novelty offered by this paper does consist in a systematic erature to address such collective organizations, among which
methodology for the formulation of the above-mentioned pur- the authors chose the term coalition, as used in the remainder of
chase-bidding strategies, that is of hourly aggregate step-wise this paper. Such coalitions are mainly nonprofit organizations
demand curves, as a result of a sensitivity analysis based on [8] that provide their members with economic advantages by
the observation of the reaction of consumers to variable hourly minimizing energy consumption costs by varying energy usage
energy tariffs. The proposed methodology is based on Monte according to the hourly variations in energy tariffs, with the
Carlo technique, this last suitable to solve the stochastic problem additional social benefit of decreasing the chances of system
numerically by producing a sufficiently high number of likely overload.
combinations of hourly energy tariffs, starting from real records The structure of a coalition is, in general, hierarchical, with an
of the market. After solving the air conditioning optimization upper-level player, the “coalition coordinator”, in the literature
problem (ACOP) [16] for each combination of the hourly energy also load serving entity [11] or aggregator [12], [13], operating
tariffs, the hourly probability density functions of the energy in the name and interest of the whole coalition, which is made of
consumption are then available in order to finally generate the lower-level players who are individual members of the coalition.
hourly aggregate step-wise demand curves, as deeply described The coordinator was here regarded as a nonprofit entity, with
in Section IV and applied in the Appendix with a straightfor- the main task to procure the electric energy needed by the coali-
ward numerical example. The formulation of the ACOP was de- tion at minimum cost. The authors are also investigating about
veloped starting from a proposal by Hämäläinen [9], analyzing what would happen if the role of the coordinator was played by
all possible consumers’ reactions to time of use (TOU) tariffs a profit entity. The main difference would probably consist in
based on mean market prices, thus resulting into a deterministic the fact that, in order to mature a profit, the coordinator would
problem, nevertheless applied to a problem of space heating. deduct a given percentage of the savings of the coalition or,
Section III outlines the main features of a coalition with worse, he would likely apply a flat rate, such as a TOU tariff,
demand-response capabilities; Section IV proposes the above much more advantageous for his interests, though much less for
mentioned methodology to define purchase-bidding strategies the coalition. However, additional research related to this aspect
by a coalition adopting demand-response programs within is needed, for instance to examine which is the most advanta-
the DAEM, defining and formulating first the ACOP for an geous tariff for a coalition coordinated by a profit entity which
individual consumer of the coalition and then the coordinator limits at minimum or, rather, avoids likely conflicts of interest.
problem for the whole coalition before solving with a Monte To accomplish the task to procure energy at minimum cost,
Carlo-based algorithm. Some numerical results are reported the coordinator faces the so-called optimum portfolio problem
in Section V, derived from simulations based on field data [19]–[21], i.e., he must decide how much energy must be
of historical trends in the Italian market, showing the impact purchased through long-term contracts and how much must
of variable hourly energy tariffs on the formulation of pur- be reserved for purchases in the DAEM. Nonetheless, before
chase-bidding strategies and how such strategies influence procuring energy, the coordinator must first derive an aggregate
MCPs. demand, collecting the electricity demands of all individual
members of the coalition. Demand can be classified according
II. COALITION WITH DEMAND-RESPONSE CAPABILITIES to the time span in which it can be foreseen and controlled
The participation of small domestic consumers in a liber- [12] in long-term, medium-term and short-term, respectively,
alized electricity market may be difficult and disadvantageous identified one year ahead (fitting for long-term contracts), one
due to several factors. For example, an individual consumer en- day (24 h) ahead, one hour ahead. A medium-term demand
tering the electricity market must be familiar with the system, is well suited to participate in one-day-ahead energy markets
knowing how, where and when an offer must be submitted, how as well as in reserve markets [22] and this is addressed in the
to forecast energy prices and determine which is the best price paper. It is worthwhile to note that the coordinator derives a
to offer. This requires the consumer to have managerial skills likely aggregate demand within a forecast process in which the
plus a thorough knowledge of energy market mechanisms. In coordinator reveals each member the forecasted hourly price
addition, an individual consumer participating in an electricity for the next day and receives information about the consequent
market may be at a disadvantage since supply opportunities for member consumption.
individual consumers differs from that offered to an organiza- A second issue for the coordinator is to choose the energy
tion of consumers which, with greater consumption volumes, rates for the members of the coalition. After the clearing of the
holds greater purchase power. Nonetheless, the transaction cost market, the coordinator, based on hourly clearing prices, has two
associated with trading directly in a wholesale electricity market alternatives for choosing the energy tariffs to apply to each coali-
is usually prohibitive to domestic consumers and, furthermore, tion member. For members who do not adhere to the demand-re-
domestic consumers may not consume enough energy to meet sponse program (DRP), the coordinator may apply a tariff per
the “minimum energy purchase” requirement of the wholesale kWh which is constant with the hours of the day. However,
market. Therefore, the idea to aggregate residential, commercial members who do modulate their consumption and adhere to the
1244 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009

program would enjoy more convenient variable hourly energy curve [12]. In this paper, we present a model which expresses
tariffs, based on MCPs. In order to adjust their electricity con- the purchase offer the coalition coordinator must formulate in
sumption according to hourly price variations, members must be this typology of purchase offer profile so to maximize coali-
able to receive the price information, and must have a means to tion benefits. Within the above-mentioned model, we consider
control their usage. Recent advances in communications, infor- the coordinator problem to embed all the boundary conditions
mation technology, and power electronics have made the imple- driving the coordinator decisions which are based on the hourly
mentation of load control systems at the residential level feasible multiple offers formulation. As conditioning energy is the main
both economically and technically [23]. In addition, each DRP cause of system overload, only energy consumption of air con-
consumer would enjoy a different degree of financial benefit, ac- ditioning systems was taken into account to evaluate whether
cording to how much consumption takes place during off-peak the coordinator can indeed lessen excessive energy use by in-
hours in which energy prices are lower. A price “reactive” con- ducing changes in consumer habits [11]. Nonetheless, whether
sumer will thus save more than a less “reactive” participant, even a storage capability was available, it would allow the electric
if the latter will still benefit from being a member of an energy consumption process to be rescheduled over a specified time
coalition rather than purchase directly as a sole consumer. With span. The energy storage capacity can be internal to the load
time, the hourly tariffs will hopefully sensitize even a less “re- (i.e., the load has a device or system that stores the excess of
active” consumer and motivate a change to concentrate major electricity with batteries or other, or the converted energy as, for
energy consumption to off-peak hours. example, in the form of ice for cool storage or heated bricks for
It is worthwhile to highlight that whichever tariffs are applied heat storage). Sometimes the environment of the load has some
to coalition members, the coordinator does choose them, thus kind of energy storage availability (or external storage) as, for
acting like a monopolist. The problem of choosing a rule con- example, the one provided by the storage capacity due to the
cerning what price a monopolist should set, in order to maximize specific heat of the air, walls, furniture, etc. located in an office
social welfare, subject to a constraint on profit, was discussed or dwelling in an air conditioning/heating load. This storage ca-
in an earlier literature by Ramsey [24] in 1927 and is known as pability (external and/or internal) would allow the electric con-
Ramsey problem or Ramsey–Boiteux pricing. According to this sumption process to be rescheduled over a specific time span,
pricing rule, for any monopoly, the price markup should be in- whether necessary, according to the market conditions [12].
verse to the price elasticity of demand: the more elastic demand Formulation of the coordinator problem must then take into
for the product, the smaller the price markup. This was stated account members’ expectations of energy savings as a function
by Robinson (1933) but it was recognized later that Ramsey of their willingness to accept higher indoor temperature. We
had found the result before (1927) in another context [24]. The speak about expectations because energy prices are not certain
rule was later applied by Boiteux (1956) to natural monopolies but forecasted. Minimum deviation thus reflects how willing in-
(decreasing mean cost): a natural monopoly experiences profit dividual coalition consumers are to trade living comfort for sav-
losses if it is forced to fix its output price at the marginal cost. ings. This equilibrium between expectations and minimum de-
Hence the Ramsey–Boiteux pricing consists into maximizing viation can thus be considered a form of “welfare maximiza-
the total welfare under the condition of nonnegative profit, that tion”. The optimal solution of the coordinator problem would
is, zero profit. In the Ramsey–Boiteux pricing, the markup of therefore be the provision of a set of optimal purchase-bidding
each commodity is also inversely proportional to the elastic- strategies which can be used in the DAEM for the next 24 h.
ities of demand, but some customers with relatively inelastic
demands may acquire a strong incentive to seek alternatives if III. COORDINATOR PROBLEM
charged higher markups, thus undermining the approach. Po- The derivation by the coordinator of an aggregate demand
litically speaking, customers with relatively inelastic demands curve to be presented in the DAEM as purchase-offer suffers
may also be viewed as those for whom the service is more nec- from a number of uncertainties.
essary or vital; thus, charging them higher markups can be chal- One uncertainty affecting the process of derivation of a pur-
lenged as unfair. According to this drawback of Ramsey theory, chase-offer is nonetheless related to the estimation of the hourly
and since the main aim of an energy-consumers coalition is market prices, outcome of the clearing process of the DAEM,
actuating DRPs and maximizing the coalition benefits, the au- according to which responsiveness of coalition members must
thors decided that all coalition members pay the same variable be evaluated. A second uncertainty is price responsiveness of
hourly energy tariffs, based on the hourly MCPs. The main re- the whole coalition, as a function of the responsiveness of each
sponsibility of the coordinator is therefore to derive an aggre- member, whereas this last can be defined as the willingness to
gate demand curve based on the individual customer demand modify consumption habits in order to enjoy lower hourly tar-
curves and contract terms, which he will present as a purchase iffs. Responsiveness can depend on many factors: some con-
offer for the 24 h of the next day in the DAEM, taking into ac- sumers are willing to sacrifice reliability to reduce their price
count the forecasted prices per hour for the following day, price for electricity, referred to as price-based demand; others need
volatility and, most importantly, each member’s responsiveness a high degree of reliability and are willing to pay for it, termed
to price signals in terms of habitual hourly energy consump- must-serve demand. Nonetheless, an individual customer may
tion. Within the set of possible offers, multiple offers allow for have both price-based demand and must-serve demand [13].
the representation of flexible purchase offers in terms of market Consumers were therefore classified as responsive, nonrespon-
price as a function of energy quantities, deriving a profile of sive, or normal [9]. A consumer is classified as responsive if
purchase offer which follows a stepwise decreasing monotonic he reacts to the energy price volatility by moving his energy
MENNITI et al.: PURCHASE-BIDDING STRATEGIES OF AN ENERGY COALITION WITH DEMAND-RESPONSE CAPABILITIES 1245

consumption towards hours in which prices are lower; a con- Naturally, this is only one way of modeling cost-efficient
sumer is classified as nonresponsive if he is not influenced by thermal comfort [17]. Other distance measures could also be
price volatility, maintaining fixed consumption habits despite used, such as the absolute deviation or the maximum devia-
the higher costs; a normal consumer shows an intermediate pat- tion from the ideal indoor temperature. The degradation of
tern of behavior compared with the responsive and nonrespon- living comfort is a penalty for the consumers’ welfare and,
sive consumers. transformed into a cost when multiplying it by a coefficient,
According to the above classification, an ACOP can then be , it can be homogenized with the other costs, yielding the
formulated for each different consumer-class, characterized by generalized cost, , given in the following, and representing
a stochastic profile reflecting the volatility of the hourly market the objective function of the generic ACOP associated with the
prices. To this end, a third factor of uncertainty the coordinator th consumer-class:
faces is the estimate of how many members are responsive, non-
responsive, or normal, respectively, , being
the total number of consumers belonging to the coalition, on
the basis of historical data and of previous experience.
The coordinator problem is thus made up of a collection of (2)
, and stochastic problems, whose simultaneous
solutions must be collected in order to estimate consumption as where h is the length of the planning period (one day),
a function of the likely hourly market prices, and thus to formu- is the price generated at the th iteration of Monte
late the aggregate curves representing the optimal purchase-bid- Carlo algorithm, presented in the following Section IV-B, as a
ding strategies to be presented in the DAEM for the next 24 h. function of the mean value, , and variance, , of the MCP
In the following section, the problem of welfare maximiza- at the th hour and of the volatility coefficient , with being
tion for the single consumers class, ACOP, is first formulated. a coefficient ranging from 0 to 1, respectively, corresponding
Then a Monte Carlo-based algorithm is proposed to solve first to null or to the highest MCP volatility, is the energy con-
sumption at the th hour, and (in /( C) ) is a real number
the ACOP for each consumer class and to finally aggregate all
representing the unwillingness to trade living comfort for sav-
ACOP solutions, achieving as final result the solution of the
ings at the th hour: the lower the value of the latter, for example
overall coordinator problem, i.e., the best bidding strategy in
/( C) , the more price-reactive the consumer is. In
the face of uncertainty over the hourly price forecast. addition, what happens at the th hour is excluded from (2)
A. Air Conditioning Consumer Optimization since it would have no effect on the indoor temperatures of the
Problem (ACOP) day under consideration.
The ACOP can thus be modeled as follows:
Electricity consumers can optimize conditioning costs and
comfort by combining two strategies: increasing the desired in-
door temperature but nonetheless taking advantage of the time-
variable electricity prices. Indeed, they increase air conditioning
use within those hours when prices are off-peak so to main- (3)
tain indoor temperatures at a desired level, but then decreasing
air conditioning use within hours characterized by peak prices.
However, this optimal strategy would require the definition and (4)
solution of a multi-criteria goal-programming problem, mod-
(5)
eled starting from the proposal of Hämäläinen et al. [9], which
however was applied to a problem of space heating, analyzing (6)
all possible consumers’ reactions to TOU tariffs based on mean (7)
market prices, thus resulting into a deterministic problem. Dif-
ferently from Hämäläinen, the problem is here characterized by where is the residence thermal dispersion coefficient (in
a stochastic nature, since market prices were modeled as sto- kW/ C), is equal to , with being the total thermal
chastic variables. As input data, the ideal indoor temperatures, capacity of the residence (in C/kWh), is a time interval of
the energy consumption profile, and the historic MCPs were one hour, , and are, respectively, the outdoor
considered over the 24 h of the day after, taking into account temperature, and lower (min) and upper (max) bounds of the
MCP volatility. To model the willingness of the generic th indoor temperature at the th hour.
consumer-class (with meaning responsive consumer, The decision variables of the problem are the
nonresponsive consumer and normal con- amounts of hourly conditioning energy consumption
sumer) to trade living comfort for savings within each th hour, a and the indoor temperatures
measure of the “degradation of living comfort”, i.e., of the devi- . Constraints (3) state that initial
ation from the ideal indoor temperature, , was introduced: and final indoor temperatures and energy consumptions,
respectively, must be the same, since they are correlated to
(1) the same hour (the 0th hour matches with the 24th hour, both
representing midnight). Constraints (4) model the thermal
where and are, respectively, the actual and ideal performance of the residence, and is functionally dependent on
residence indoor temperatures at the th hour for the th con- the aforementioned sets of decision variables, and , on
sumer-class. the outdoor temperatures, , and on
1246 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009

the physical features of the residence, described by coefficients

and . Constraints (5) define the maximum variation of the
indoor temperature a consumer is willing to tolerate. Equations
(6) represent a class of capacity constraints, whereby the total
energy absorbed by the air conditioning system must not
exceed the size of the system itself, . Finally, (7) state that
the amount of energy must be positive or null. The specific
parameters modeling the overall reaction of a consumer,
hereafter addressed as consumer-responsiveness, are the ideal
indoor temperature vector, , the
lower and upper temperature limits,
and , and the hourly trade-off Fig. 1. Monte Carlo simulation.
coefficients, . In the following
computations, we assume a trade-off coefficient which is
equal for all the consumers of the generic th consumer-class
and also constant for each th hour of the planning period 4) Output variables: their values provide the simulation re-
( with ). These last assumptions on were sults (the hourly probability density functions of the energy
evidently made for the sake of simplicity; nonetheless, is consumption, for each generic th consumer-class at each
naturally varying with consumer-class and hours of the day, th hour).
and an analysis of its variability might be conducted by means The Monte Carlo simulation is essentially based on the as-
of recent advances in home automation and power electronics. sumption that a direct analytic solution to a problem might often
be too onerous or, to a certain extent, impossible. The problem
B. Solving for the Coordinator Problem Using a Monte under question is then numerically solved, producing a suffi-
Carlo-Based Algorithm ciently high number, , of likely combinations of input variable
From the previous ACOP definition of air conditioning en- values (MCPs) and evaluating the corresponding output (the
ergy consumption, the coordinator problem formulation is rep- hourly probability density functions of the energy consumption,
resented by a collection of stochastic ACOPs, which are as for each generic th consumer-class at each th hour) on the
numerous as the number of coalition members, a function of basis of the model equations. To build each combination, a value
the coalition composition ( and ), of the willing- is randomly generated (i.e., “extracted”) for each input variable,
ness of each member to trade living comfort for savings ( in accordance with the specified probability density function
being the unwillingness), and of the volatility of the MCPs (the and the variables correlations. Iterating this algorithm times
coefficient). The responsiveness classification of each coali- (with quite large so as to obtain acceptable results stochasti-
tion member, the determination of the exact composition of the cally), independent values are obtained for the output vari-
coalition, and the association of each consumer-class with the ables, representing a sample of likely output values. Such a
right values of are extremely challenging tasks for the coor- sample could then be analyzed by means of stochastic tech-
dinator, and while they represent an interesting research topic, niques, to generate frequency histograms and to numerically de-
it was not possible to consider them in this paper, remaining a fine the trends of the output probability density functions.
future aim of the authors. Therefore, to be able to understand In the remainder, as before, lower-scripts rc, nrc, and nc will
how the proposed models work, the authors chose all the afore- characterize variables and parameters, respectively, of respon-
mentioned parameters according to what can be found in the sive, nonresponsive, and normal consumers. The proposed al-
literature [9]. gorithm, based on the Monte Carlo simulation, solving for the
To solve the coordinator problem, a Monte Carlo-based algo- coordinator problem, can be detailed as follows:
rithm was used to reproduce and numerically solve for the set of 1) are predefined and chosen on the basis
ACOPs, each maximizing the welfare of a single consumer. of the experience of the coordinator and of historical data
Because of the stochastic nature of the variables used and the about member behaviors. For the sake of simplicity, a
computational complexities in the solution search, it was some- unique (MCP volatility coefficient) is used for each
times impossible to obtain an analytic solution. For a generic th consumer-class. Finally, ideal temperatures, physical
problem, the main elements of the Monte Carlo simulation can features of the residence and hourly trade-off coefficients
be summarized as follows (Fig. 1): are fixed ( element of the Monte Carlo simulation);
1) Parameters: input specified by the decider/analyst (e.g., the 2) The mean value and the variance ( and ) of the hourly
Coordinator) of the problem, and thus controllable; MCPs are calculated by using MCPs from real records of
2) Input variables: depending on external events, not con- the market ( element of Monte Carlo simulation) and as-
trolled by the decider, the trend of which can, however, be suming that such prices follow a log-normal distribution;
described since they are managed as stochastic variables 3) A volatility range for the hourly MCPs is defined,
(e.g., MCPs); , where is the standard deviation from
3) Model: mathematical equations (function of the parame- ;
ters and of the hexogen input variables) describing the rela- 4) From the above defined volatility range, a random price
tions among the system/problem components and defining is generated, being the index of the generic
the link among parameters, input and outputs variables iteration of Monte Carlo simulation ( and
(e.g., ACOP); in our simulations);
MENNITI et al.: PURCHASE-BIDDING STRATEGIES OF AN ENERGY COALITION WITH DEMAND-RESPONSE CAPABILITIES 1247

Finally, it is important to note that the Monte Carlo-based

algorithm does not provide an exact analytic formulation of the
output variables (the hourly probability density functions of the
energy consumption, for each generic th consumer-class at the
th hour), rather a sample of values, whose frequency allows
to derive an approximate indication of the probability density
functions of these output variables. With an invocation of the
central limit theorem [27], it was demonstrated that the higher
the number of simulation, , the more numerous the sample of
output variables, thus the finer the precision and accuracy.
On the other hand, increasing the number of simulations to
Fig. 2. Typical discretization of the probability density function of the energy
reach an approximate solution, the nearest to the corresponding
consumption for the generic ith hour and the generic k th consumer-class. exact solution, could however present problems of compu-
tational tractability, which are currently overcome through
recourse to efficient computational facilities such as parallel
5) For each th consumer-class, the ACOP ( element of the computing.
Monte Carlo simulation) solution is performed for the dif-
ferent consumer-classes, as a function of the random price IV. NUMERICAL RESULTS
generated in step 4, yielding triplets of en-
ergy consumption at each th hour; This section presents the simulations which considered the
6) Iterating times steps 4 and 5, a sample of triplets of uniform national prices (PUNs in Italy), i.e., national purchasing
energy consumption prices, equal to the average of the zonal selling-prices weighted
at each th hour is obtained (i.e., three probability on the zonal consumption, of the Italian DAEM for summer
density functions, one for each consumer-class) and 2005 [26] using the maximum outdoor temperature of one of the
their mean values and variances hottest days of summer 2005 (27/07/2005) as reference. Firstly,
are determined. the consumption strategies for each consumer-class (respon-
The continuously-valued load forecast just developed, effect sive, normal, and nonresponsive) were analyzed and compared
of the forecasted prices, is not practical to formulate market from an economic viewpoint as a function of different values
bids, since only discrete bids are allowed. As a consequence, it is of the MCP volatility coefficient, . The algorithm outlined in
more reasonable to consider an approximation whereby the con- Section IV-B which solves for the coordinator problem was then
tinuous probability density functions of energy consumptions applied to formulate multiple purchase offers in the DAEM for
and hourly prices are discretized (sampled) through a number a generic coalition, taking into account the uncertainty over the
of representative “slices” [25]. Here, four intervals are consid- hourly price forecast. Finally, the economic benefits were pre-
ered, being this the particular case of the Italian DAEM [26], sented for multiple purchase offers in the market.
from where historical data have been taken for simulations. Each
of the intervals were considered one half of the corresponding A. Input Data
load/price forecast standard deviation-wide. In detail, input data involved in the coordinator problem for-
The probability density function slicing process produces dis- mulation can be distinguished into input variables and parame-
crete-valued random variables and , for prices and en-
ters. Input variables are represented by PUNs, while parameters
ergy consumptions, respectively, with . For each
can be divided into three categories, as detailed in the following:
slice and are defined at the center of their respec-
•) Physical parameters and characteristics of a residence;
tive interval, with corresponding probabilities evaluated using
•) Habit parameters (set of desired ideal temperature,
standard techniques [27]. For example, in Fig. 2, the four dis-
crete multiple purchase-offers of the energy consumption for the , lower and upper bounds of the indoor tem-
generic th hour and the generic th consumer-class are peratures ( and , respectively) and trade-off
coefficients );
, and so on. •) Environmental parameters (outdoor temperatures,
Multiple purchase-offers are thus finally determined at each ).
th hour aggregating the solutions obtained for the different con- The physical parameters correlated with a consumer-class are
sumer-classes: represented by the heat capacity of the residence, , giving
, the loss coefficient, , and the maximum power of
the air conditioning system, . Empirical values assumed
by these parameters for a typical Italian residence are reported
in Table I [28]. Each th consumer-class is characterized by
where , with the upper and lower bounds of the expected
and . indoor temperature, an ideal level of the indoor temperatures
For the sake of clarity, a straightforward numerical example and a willingness to trade living comfort for savings,
was presented in the Appendix, applying the above algorithm with trade-off coefficients , which remain constant over the
step by step and showing how multiple purchase-offers can be reference day . It is worth to underline
determined. that, for the sake of simplicity and computational tractability,
1248 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009

TABLE I
DESCRIPTION AND ESTIMATION OF PARAMETERS
OF A TYPICAL ITALIAN RESIDENCE

Fig. 4. Variance of hourly PUNs over the three months of summer 2005 (June,
July, and August).

Fig. 3. Mean value of hourly PUNs in the three months of summer 2005 (June,
July, and August).
Fig. 5. Reference outdoor temperature (July 27, 2005).

outdoor temperatures, , were here considered among envi-

ronmental parameters, though they would be classified as a sto- The hourly outdoor temperature for the day under considera-
chastic variable at each th hour, as authors did elsewhere [16]. tion is reported in Fig. 5.
Responsive consumers exhibit trade-off coefficients B. Consumer-Class Responsiveness as a Function
near to zero, with values ranging from 0.01 to 0.2 /( C) , of Price Volatility
whereas the trade-off coefficients of normal and nonresponsive
consumers, and , ranged from 0.4 to 2 /( C) and Responsive consumers, more than normal and nonresponsive
from 4 to 50 /( C) , respectively. The values associated to consumers, enjoy lower purchase costs since they mainly pur-
trade-off coefficients were taken by the paper of Hämäläinen chase electricity within off-peak hours [16]. Here, we examined
[9], which the reader can refer to for more details. In the the influence of electricity price volatility on the consumption
Monte Carlo-based algorithm, indoor temperature is allowed to habits of these three consumer-classes, applying steps 1 to 7 of
vary between 21 and 23 C over the day, with a constant ideal the algorithm in Section IV-B.
temperature level of 22 C and a size of the air conditioning The Monte Carlo simulation was carried out as a function of
system, , equal to 9.0 kW. The daily consumption habits varying values of volatility coefficient to obtain the expected
of consumers using air conditioning were analyzed, taking into values of the mean quantities of electricity, as calculated in (8),
account the volatility of the PUNs over the three hottest months which are to be purchased by the coalition coordinator for the
of summer 2005 (June, July, and August). Mean value and vari- different consumer-classes within peak and off-peak hours (see
ance of the hourly PUNs in the DAEM were also calculated on Figs. 6 and 7).
these three months and their probability density functions over For this, the hypothesis that the output variables
the 24 h of the hottest day (July 27, 2005) are shown in Figs. 3 follow a normal distribution was assumed,
and 4. Without loss of generality and applicability, the hourly thus corroborating with the literature (see f. i. [30] and [31], the
correlation of PUNs was not taken into account systematically, later affirming that a normal distribution on the load, produces
f. i. by means of a covariance matrix, but through approximate a log-normal distribution on price, creating a nice link to
conditioning [29], whereas constraints were imposed on the traditional models, which is the case of our work):
values which PUNs may assume in order to simulate a quite
realistic trend. In fact, as a result of a statistic analysis of the
Italian market operator [26] hours 20–24 (late-evening peak) are
(8)
highly correlated among each other and show small correlations
with the other hours of the day. The same pattern is presented where represents the probability associated with the
by hours 5–9 (early-morning shoulder) and 11–14 (afternoon output variables (frequency of in the solutions of the
peak) and the physical reasons for the above correlations lay on ACOPs). Responsive consumers purchase energy exclusively
the structure and the composition of the load. during off-peak hours (from 1:00 to 9:00 and from 19:00 to
MENNITI et al.: PURCHASE-BIDDING STRATEGIES OF AN ENERGY COALITION WITH DEMAND-RESPONSE CAPABILITIES 1249

Fig. 8. Log-normal distribution of PUN during a peak and an off-peak hour for
increasing values of the volatility coefficient A.

Fig. 6. Peak hours: expected quantity of energy purchased by consumers for

different values of volatility coefficient.

Fig. 9. Mean and variance values of the generalized cost as a function of dif-
ferent values of the volatility coefficient (A) for a responsive consumer.

Fig. 7. Off-peak hours: expected quantity of energy purchased by consumers

for different values of volatility coefficient.

24:00; see Fig. 7), slightly reducing consumption when price

volatility is maximum (from 26.896 kWh, when to
25.731 kWh, when ; see Fig. 7). Indeed, as a conse-
quence of the highly volatile PUN (i.e., when ) during
peak hours (10:00–18:00), responsive consumers purchase Fig. 10. Mean and variance values of the generalized cost as a function of dif-
only 5% (1.172 kWh in Fig. 6) of their total daily consumption ferent values of the volatility coefficient (A) for a normal consumer.
( kWh). An increase in PUN volatility had a
greater effect on the consumption habit of a normal consumer.
In fact, the greater the price volatility (increasing values of ), The mean value (9) and the variance (10) of the generalized
the greater the probability to have lower prices during peak cost for each th consumer-class were also calculated
hours, when consumers also expect higher temperatures. This (see Figs. 9–11), having assumed that output values follow a
behavior might be explained observing that normal consumers normal distribution:
perform a different behavior compared with responsive, who
still concentrate their greater consumption during off-peak
hours though volatility increases, which corroborates with the
latest literature [15]. The greater increase in energy consump- (9)
tion by normal consumers during peak hours is thus strictly
related to the values assumed by the respective coefficients
, which differentiate consumer-class consumption habits,
(10)
and, obviously, to temperatures. As said in Section V-A, the
authors assumed outdoor temperatures, , among parame-
ters, though they are stochastic variables, as they were modeled Fig. 9 shows how, with an increment of PUN volatility, the
elsewhere [16]. generalized cost decreases. The same trend of the generalized
From an examination of the historical data (Fig. 4), it is ev- cost is observed for normal and nonresponsive consumers
ident that PUN variance is lower within off-peak hours than (Figs. 10 and 11). The cost reduction, ,
within peak hours. Therefore, with an increase of the volatility defined as the difference between the generalized cost for
coefficient, consumers not only purchase at prices lower than the and , is approximately equal to 8% of
mean value but do so with a higher probability [area (a) of Fig. 8] for responsive and nonresponsive consumers, and equal to 20%
during peak than during off-peak hours [area (b) of Fig. 9]. for normal consumers.
1250 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009

Fig. 12. Convergence of the mean value of the generalized cost to the “exact”
Fig. 11. Mean and variance values of the generalized cost as a function of dif- m
value as a function of the number of simulations, .
ferent values of the volatility coefficient (A) for a nonresponsive consumer.

on a Pentium 4, 1.8 GHz with 1056 GB of RAM using version

The trends of the generalized cost are characteristic of con- 4.0 of Microsoft Fortran Power-Station. The solution time was
sumer-class responsiveness since, the greater the increase in of 15 min for 1000 iterations, therefore suitable for practical
the volatility coefficient of PUN, the higher the probability of implementations.
reaching medium-low values of PUN during peak hours. As a From results of Section V-B and based on the aforemen-
consequence, the various consumer-classes distribute consump- tioned assumptions made about coalition composition, calcu-
tion over the 24 h according to their different reactions to price lations showed that the generalized cost is equal to 172.289
volatility. ( 1.72289*100 from Fig. 11) and 143.5437 (
As an example, for (high PUN volatility), a responsive 1.72289*50), for homogeneous and heteroge-
consumer spends 0.86078 while a normal and a nonresponsive neous coalitions, respectively. Applying the algorithm outlined
consumers spend and 1.65119, respectively, indi- in Section III-B, the coordinator is able to estimate the hourly
cating that when PUN volatility is high, responsive consumers, consumption of the coalition based on PUN values and thus de-
who maximize their consumption to off-peak hours, enjoy lower termine the couples (price, quantity) to submit in the DAEM for
costs. Conversely, normal or nonresponsive consumers prefer to the next 24 h. Figs. 13 and 14 depict these couples distributed
spend more to reach the desired level of living comfort. There- over 24 h, illustrating how there is a different hourly consump-
fore, each consumer-class maximizes welfare, although they tion in the presence of price-sensitive consumers (Fig. 13), the
have different expectations of energy savings due to different last choosing to enjoy lower tariffs rather than maximize their
degrees of willingness to accept higher indoor temperatures. living comfort. Fig. 15 illustrates the stepwise trend of a multiple
offer at 12.00 am of Fig. 14, derived by aggregation of the cou-
C. Purchase-Bidding Strategies for a Coalition ples (quantity, price) of the four offers from the most to the least
Using the algorithm developed in Section IV-B, it was pos- costly, giving rise to a typical step-wise trend. From a compar-
sible to define the multiple offers which would be presented in ison of the two proposed cases, the solution to the coordinator
the DAEM for July 27, 2005, the hottest day of the year. Ac- problem thus demonstrates that the multiple offer from a het-
cording to step 1, the number of coalition members was arbi- erogeneous coalition shifts and maximizes energy consumption
trarily fixed to , with a base case, in which the coali- towards off-peak hours (Fig. 13) with respect to the same ty-
tion is homogeneous, i.e., made up of nonresponsive consumers pology of offer presented by a homogeneous coalition of only
only , and a generic case, in which the coali- nonresponsive consumers (Fig. 14).
tion is heterogeneous, with responsive, The initial choice on the numbers of consumers belonging
normal, and nonresponsive consumers, representing to the coalition, in fact, greatly influences the purchase-bidding
a coalition with more responsiveness than the homogeneous strategies. In particular, for a heterogeneous coalition, the in-
base case, however with a dominance of normal and nonrespon- fluence of responsive and normal consumers ( , and
sive consumers. The value at which is set is of particular im- ), comprising 50% of total number of consumers
portance since, the day ahead for the day after, PUN volatility , yielded offers which are shifted to concentrate
should not be high and although it is possible for to approach consumption to off-peak hours when PUNs are medium-low.
1, an was used, having verified through a sensitivity Referring to Figs. 13 and 14, we addressed so far the problem
analysis that this value is a good estimations to represent a con- of the formulation of a purchase bidding strategy. Consequent
dition of minimum risk. to this problem, as a function of the market clearing process for
Moreover, an increase of the number of iteration, , leads to a each hour, the quantity of energy accepted by the market will
convergence of the output variables (the hourly probability den- represent the energy “contracted” by the coalition. Nonetheless,
sity functions of the energy consumption, for each generic th in real time, the quantity of energy required by the coalition may
consumer-class at the th hour) to the relating analytic values. differ from that contracted. The responsibility for such a differ-
Fig. 12 shows the trend of the values assumed by a function ence between actual and contracted hourly energy is pertinence
of the output variables, the mean value of the generalized cost of the balance responsible party, which is required to pay for this
function which, after iterations, converges to a final difference if it is positive, whereas it is refunded by the market
solution. The Monte Carlo-based algorithm was implemented operator otherwise, according to prices of the balancing market
MENNITI et al.: PURCHASE-BIDDING STRATEGIES OF AN ENERGY COALITION WITH DEMAND-RESPONSE CAPABILITIES 1251

Fig. 14. Multiple purchase offers over 24 h for a homogeneous coalition, with
energy quantities given by the histograms while prices are represented by the
dotted line.
Fig. 13. Multiple purchase offers over 24 h for a heterogeneous coalition, with
energy quantities given by the histograms while prices are represented by the
dotted line.
the responsible end-users. The more the coordinator improves
the capability of forecasting the needed energy as a function of
[32]. The balance responsible party may be the coalition itself, the forecasted prices, the lower the final price of energy for the
or any other dispatching user signing, as a third party, a contract whole coalition is, and this trend is assured by the nonprofit na-
for settlement of imbalances with the TSO. ture of the coordinator himself. There is therefore a particular
The coordinator is then responsible for such a difference be- need to focus on market price forecast, however, out of the scope
tween actual and contracted hourly energy and, in the name of of this paper, and authors are already working in this direction.
the coalition, he is required to pay for this difference if it is pos- Finally, the benefits in terms of reduction of inter-hour price
itive, whereas he is refunded by the market operator otherwise, volatility due to the presence of an elastic demand were pre-
according to prices of the balancing market [32]. These eco- sented by means of an example. Analysis which compares the
nomic unbalances can be handled by the coordinator by using effect of multiple purchase offers in the DAEM during an off-
load profiling techniques, so that he is able to attribute the de- peak hour (2 a.m.) and peak hour (12 p.m.) was thus under-
viations between actual and forecasted energy consumption to taken. For each th hour (2 a.m., 12 p.m.), a selling bid curve
1252 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009

Fig. 17. Different PUNs for different purchase offers during a peak hour.
Fig. 15. Stepwise trend of a multiple offer at 12:00 p.m. by a homogeneous
coalition.

TABLE II
MEAN VALUE AND VARIANCE OF THE HOURLY MCPS

Fig. 16. Different PUNs for different purchase offers during an off-peak hour.

was arbitrarily chosen and two purchase-offer curves were de-

rived, representing a homogeneous coalition exhibiting, respec-
tively, elastic and inelastic demand habits. The inelastic demand
is represented by a fixed quantity of consumption, regardless
of energy cost, resulting from the solution of the ACOP as a
function of a deterministic price for that hour, , whereas the
elastic demand is derived from the previous results for a homo-
geneous coalition. The PUN corresponding to the clearing
point is the clearing price which intersects the selling bid curve
and the inelastic purchase offer curve; conversely, the PUN of TABLE III
TRIPLETS OF ENERGY CONSUMPTION AT PEAK AND OFF-PEAK HOURS
the clearing point is the clearing price derived from the inter-
section between the selling bid curve and the multiple purchase
offer curve. These results clearly demonstrate that an elastic de-
mand makes it possible to obtain a higher PUN during off-peak
hours (Fig. 16) and, likewise, lower PUN during peak hours
when price is higher (Fig. 17). Elastic demand therefore signif-
icantly reduces PUN during peak hours compared with the case
of inelastic demand (Fig. 17), which corroborated with the liter-
ature [11]. In addition, and more importantly, this decreases the
inter-hour volatility of PUN.
Nonetheless, for the sake of completeness, it should be
noted that volatility of day-ahead prices might not necessarily
decrease with the introduction of elastic demand. In fact, many
studies in the literature (see f.i. [33] and [34]) shown that
volatility of day-ahead prices can actually increase when the of England and Wales (EPEW) has shown that a market based
start-up and no-load costs of generating units are considered on multiple bids minimizes the production cost rather than
in the market clearing mechanism and this because electricity minimizing the price of electricity for a given level of demand
producers do not assume the risk associated with the start-up as an efficient market would and this cost minimization can
and no-load costs or this risk is passed on to the pool. Moreover, result occasionally and unexpectedly in large price increases
as argued in [33] and [34], experience with the Electricity Pool over a short period.
MENNITI et al.: PURCHASE-BIDDING STRATEGIES OF AN ENERGY COALITION WITH DEMAND-RESPONSE CAPABILITIES 1253

TABLE IV
MEAN VALUE OF ENERGY CONSUMPTION AT PEAK AND OFF-PEAK HOURS

TABLE V
VARIANCE OF ENERGY CONSUMPTION AT PEAK AND OFF-PEAK HOURS

TABLE VI
FOUR DISCRETE MULTIPLE PURCHASE-OFFERS OF ENERGY CONSUMPTION AT
PEAK AND OFF-PEAK HOURS

Fig. 18. Multiple purchase offers over peak and off-peak hours with energy
TABLE VII quantities given by the histograms while prices are represented by the dotted
FOUR DISCRETE MULTIPLE PURCHASE-OFFERS OF ENERGY CONSUMPTION line.
AND PRICE AT PEAK AND OFF-PEAK HOURS

with the literature. In addition, it was shown that the presence

of an elastic demand leads not only to a subsequent reduction
in prices during peak-hours but also simultaneously decreases
inter-hour PUN volatility.
Though the proposed methodology can be easily extended to
several other electric loads, however, a demand response pro-
gram has a sense only when the electric load in consideration
V. CONCLUSION allows a potential modulation of consumption as a function of
This paper proposed a Monte Carlo-based algorithm solving varying hourly prices over the 24 h. Besides air conditioning,
for the problem of a coordinator of a coalition of energy other electric loads with features of consumption shifting are for
consumers sensitive to demand-response programs, who must instance represented by electric household appliances (laundry,
determine the best purchase-bidding offers in the day-ahead oven, dishwasher, iron) which cause a consistent percentage of
energy market in the face of uncertainty over the hourly price electricity consumption on the bill of an end user. Always ac-
forecast, this last based on the hourly purchasing market cording to the willingness of a consumer to trade living comfort
price in Italy (PUN). Field records of the Italian electricity for savings, the use of these electric appliances may be shifted
market were used as input stochastic variables of the proposed during the night, when electricity prices are commonly lower.
Monte Carlo-based algorithm, which solved numerically the The algorithm in Section IV-B would still apply, having mod-
coordinator problem. The solution of the above-mentioned ified the ACOP of Section IV-A which must model new con-
problem consisted in the derivation of stepwise purchase-bid straints such as the energy consumption of the th group of
curves which would be presented in the DAEM so to max- consumers at the th hour must equal the sum of energy con-
imize energy savings, as a function of willingness to accept sumptions by all appliances, with each appliance consumption
higher indoor temperatures of coalition consumers, and of bounded by the relating size, like for air conditioning. The new
PUN volatility. Though Monte Carlo method is very easy to load consumptions would not be influenced by the internal tem-
be implemented, however, it could present problems of com- perature, thus not driven by the minimization of the term degra-
putational tractability, which are currently overcome through dation of living comfort in the objective function, though still
recourse to efficient computational facilities such as parallel valid for air conditioning. Nonetheless, the objective function
computing. Numerical results also provide a discussion on the should comprise new addends suitable to simulate the willing-
lowering effects on the hourly wholesale electricity prices a ness of different groups of consumers to shift their consumption
peak shifting behavior conducted on an aggregated level of as a function of the hour of the day, then of price in that hour,
demand response programs may produce, which corroborate and of the appliance in consideration. As to lightings, though
1254 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009

this load could be embedded into the formulation, it would not where ,
be driven by price variability, since light is a primary need in with and , as pro-
particular hours of the day, not storable and not substitutable
vided in Table VII.
with other services, rather controllable by home automation.
The four discrete multiple purchase-offers of
As future aim, authors are studying the relating shifting capa-
energy price at peak and off-peak hours were,
bility of different electric loads, in addition to the importance of
respectively, derived as the average over peak
investments in well-designed control schemes that enable dis-
tribution-side demand responsiveness. and over off-peak hours of the four offers cal-
culated over the 24 h with the same procedure
APPENDIX applied for the four offers of energy quantity.
NUMERICAL APPLICATION OF THE PROPOSED Fig. 18 depicts the four discrete multiple pur-
MONTE CARLO-BASED ALGORITHM chase-offers distributed over peak and off-peak
hours.
1) Let us choose , and
for each th consumer-class. Other input parameters,
such as physical features of the residence, hourly trade-off REFERENCES
coefficients and ideal temperatures were fixed as in [1] D. Menniti, A. Pinnarelli, N. Scordino, and N. Sorrentino, “UPFC
sizing by means of an economically correct secure economic dispatch
Section V. in a deregulated power market scenario,” Eur. Trans. Elect. Power,
vol. 18, no. 1, pp. 65–78, Jan. 2008.
[2] J. J. Carpentier, D. Menniti, A. Pinnarelli, N. Scordino, and N. Sor-
2) The mean value and the variance ( and ) of the hourly rentino, “A model for the ISO insecurity costs management in a dereg-
ulated market scenario,” in Proc. IEEE PowerTech2001 Conf., Porto,
MCPs, respectively, reported in Figs. 3 and 4, are here Portugal, 2001, vol. 1.
provided in Table II. [3] J. J. Carpentier, D. Menniti, A. Pinnarelli, N. Scordino, and N. Sor-
rentino, “Secure economic dispatch: An “Economically correct” ap-
proach in a deregulated market scenario taking into account the pres-
ence of FACTS devices,” in Proc. Conf. Bulk Power System Dynamics
3) A volatility range for the hourly MCPs is defined, and Control V, Onomichi, Japan, 2001.
, where is the standard deviation from [4] [Online]. Available: http://www.UCTE.org.
. [5] [Online]. Available: http://www.ferc.gov.
[6] D. S. Kirschen, “Demand-side view of electricity markets,” IEEE
Trans. Power Syst., vol. 18, no. 2, pp. 520–527, May 2003.
[7] [Online]. Available: http://tdworld.com.
4) From the above-defined volatility range, a random price [8] [Online]. Available: http://www.energycoalition.org.
is generated, being the index of the [9] R. P. Hämäläinen, J. Mäntysaari, J. Ruusunen, and P.-O. Pineau, “Co-
operative consumers in a deregulated electricity market-dynamic con-
generic iteration of Monte Carlo simulation ( sumption strategies and price coordination,” Energy-Int. J., vol. 25, pp.
with ), thus obtaining ten samples of 24 hourly 857–875, 2000.
prices. [10] Finnish Energy Industry Federation (FINERGY) and Finnish Elec-
tricity Association (SENER): “E.L.E.C.T.R.I.C.I.T.Y. in Finland:
Generation—Consumption—Transmission—Market”, Brochure,
1997.
5) For the sake of simplicity, Table III provides triplets [11] M. Ilic, J. W. Black, and J. L. Wats, “Potential benefits of implementing
load control,” in Proc. IEEE PES Winter Power Meeting, New York,
and of energy consumption yielded by ACOP solution for 2002, vol. 1, pp. 177–182.
6) each th consumer-class as a function of peak hours [12] C. Álvarez, A. Gabaldón, and A. Molina, “Assessment and simulation
– and off-peak hours ( – of the responsive demand potential in end-user facilities: Application
to a university customer,” IEEE Trans. Power Syst., vol. 19, no. 2, pp.
and 19–24) rather than as a function of each th hour 1223–1231, May 2004.
(i.e., sum of consumptions over peak [13] H. Oh and R. J. Thomas, “Demand-side bidding agents: Modeling and
simulation,” IEEE Trans. Power Syst., to be published.
hours and sum of [14] [Online]. Available: http://ec.europa.eu/research/energy/pdf/smart-
consumptions over off-peak hours): with associated mean grids_en.pdf.
value and and variance and [15] Faruqui and A. S. Sergici, “The power of experimentation:
New evidence on residential demand response,” in Discus-
provided in Tables IV and V. The four discrete multiple sion Paper, Brattle Group, May 11, 2008. [Online]. Available:
purchase-offers of the energy consumption at peak and http://www.brattle.com/_documents/UploadLibrary/Upload683.pdf.
off-peak hours for the generic th consumer-class are [16] D. Menniti, F. Costanzo, N. Scordino, and N. Sorrentino, “Coalition of
consumers in a deregulated electricity market: Analyses of consump-
calculated as , , tion dynamics of air conditioning users,” in Proc. IEEE PES Power
, and , with Systems Conf. Expo (PSCE’06), Atlanta, GA, 2006, pp. 1174–1181.
respective values shown in Table VI. [17] R. P. Hamalainen and J. Mantysaari, “Dynamic multi-objective heating
optimization,” Eur. J. Oper. Res., vol. 142, no. 1, pp. 1–15, Oct. 2002.
[18] D. Menniti, A. Burgio, A. Pinnarelli, N. Scordino, N. Sorrentino, and
G. Costanzo, “The energy markets from a small scale consumer per-
7) Multiple purchase-offers are thus finally determined at spective: The CRETA experience,” in Proc. 5th Int. Conf. European
Electricity Market (EEM 2008), Lisboa, Portugal, May 28–30, 2008.
peak and off-peak hours aggregating the solutions ob-
[19] D. Menniti, R. Musmanno, N. Scordino, and N. Sorrentino, “Portfolio
tained for the different consumer-classes: selection problem of a big consumer in the Italian electricity market,”
in Proc. 3rd Int. Conf. Electric Utility Deregulation and Restructuring
and Power Technologies (DRPT 2008), Nanjing, China, Apr. 6–9,
2008, pp. 316–322.
MENNITI et al.: PURCHASE-BIDDING STRATEGIES OF AN ENERGY COALITION WITH DEMAND-RESPONSE CAPABILITIES 1255

[20] H. Levy and M. Sarnat, Portfolio and Investment Selection: Theory and Ferdinando Costanzo was born in Cosenza, Italy, on
Practice. Englewood Cliffs, NJ: Prentice-Hall, 1984. December 10, 1980. He received the management en-
[21] M. Liu and F. F. Wu, “Managing price risk in a multimarket environ- gineering degree in 2005 from the University of Cal-
ment,” IEEE Trans. Power Syst., vol. 21, no. 4, pp. 1512–1519, Nov. abria, Cosenza, Italy.
2006. His current research interests concern the op-
[22] N. S. Rau, “Assignment of capability obligation to entities in compet- timization of models of deregulated electricity
itive markets: The concept of reliability equity,” IEEE Trans. Power markets, taking into account the bidding and pur-
Syst., vol. 14, no. 3, pp. 884–889, Aug. 1999. chase strategies of market operators.
[23] J. W. Black and M. Ilic, “Survey of technologies and cost estimates for
residential electricity services,” in Proc. IEEE PES Summer Meeting,
Vancouver, BC, Canada, Jul. 2001, vol. 1, pp. 255–260, 2001.
[24] F. P. Ramsey, “A contribution to the theory of taxation,” Econ. J., vol.
37, no. 145, pp. 47–61, Mar. 1927.
[25] J. R. Birge and F. Louveaux, Introduction to Stochastic Program-
ming. New York: Springer, 1997. Nadia Scordino (M’03) was born in Varese, Italy,
[26] [Online]. Available: http://www.mercatoelettrico.org. on February 13, 1975. She received the management
[27] A. Papoulis, Probability, Random Variables, and Stochastic Processes, engineering degree from the University of Calabria,
3rd ed. Boston, MA: McGraw-Hill, 1991. Cosenza, Italy, in 1999 and the Ph.D. degree in
[28] Internal Report, Faculty of Engineering, Dept. Mech. Eng., Univ. Cal- electrical engineering from the University of Naples,
abria, Cosenza, Italy. Naples, Italy, in 2002.
[29] B. Yanga and J. E. Kolassa, “A refinement to approximate conditional Her current research interests concern integrated
inference,” Statist. Probab. Lett., vol. 72, no. 2, pp. 103–112, Apr. solutions of optimal power flow and steady state se-
2005. curity analysis, taking into account FACTS devices
[30] A. Seppala, “Statistical distribution of customer load profiles,” in Proc. also in a deregulated energy market and the modeling
Int. Conf. Energy Management and Power Delivery, Singapore, 1995, and controlling of FACTS devices for power flow reg-
vol. 2, pp. 696–701. ulations in transmission networks.
[31] P. Skantze, M. Ilic, and J. Chapman, “Stochastic modeling of elec- Dr. Scordino is a member of the IEEE Power Engineering Society.
tric power prices in a multi-market environment,” in Proc. IEEE PES
Winter Meeting, 2000, vol. 2, pp. 1109–1114.
[32] [Online]. Available: http://www.mercatoelettrico.org/en/MenuBib-
lioteca/documenti/20041027ElectricityMarketManual.pdf. Nicola Sorrentino (M’99) was born in Cosenza,
[33] D. S. Kirschen, G. Strbac, P. Cumperayot, and D. de Paiva Mendes, Italy, on October 26, 1970. He received the man-
“Factoring the elasticity of demand in electricity prices,” IEEE Trans. agement engineering degree in 1994 and the Ph.D.
Power Syst., vol. 15, no. 2, pp. 612–617, May 2000. degree in computer and system engineering in 1999
[34] C. Liang Su, “Optimal demand-side participation in day-ahead from the Electronic, Computer, and Systems Science
electricity markets” Ph.D. dissertation, Faculty of Engineering and Department of the University of Calabria, Cosenza,
Physical Sciences, Univ. Manchester, Manchester, U.K., 2007. Italy.
[Online]. Available: http://www.eee.manchester.ac.uk/research/ He is an Assistant Professor in the Electronic,
groups/eeps/docs/su_PhD_2007.pdf. Computer, and Systems Science Department of the
University of Calabria. His current research interests
Daniele Menniti (M’97) was born in Susa (TO), concern the decentralized control and analysis of
Italy, on September 23, 1958. He received the electrical power systems.
electrical engineering degree from the University Dr. Sorrentino is a member of the IEEE Power Engineering Society.
of Calabria, Cosenza, Italy, in 1984 and the Ph.D.
degree in electrical engineering from the University
of Naples, Naples, Italy, in 1989.
He is an Associate Professor in the Electronic,
Computer, and Systems Science Department of the
University of Calabria. His current research interests
concern the analysis of electrical power systems,
real-time control, and automation.
Prof. Menniti is a member of the IEEE Power Engineering Society.