Informatica 41 (2017) 121–128 121
Dynamic Protocol for the Demand Management of Heterogeneous Resources
with Convex Cost Functions
Jernej Zupančič
International Postgraduate School Jozef Stefan and Jozef Stefan Institute
Jamova cesta 39, 1000 Ljubljana
E-mail: jernej.zupancic@ijs.si
Matjaž Gams
Jozef Stefan Institute
Jamova cesta 39, 1000 Ljubljana
E-mail: matjaz.gams@ijs.si
Keywords: pricing mechanism, negotiations, multiagent system, resource demand balancing
Received: March 21, 2017
Increasing demand of resources has driven research towards design of mechanisms that address resource-
demand management, which usually focus on balancing peak demand and optimal device scheduling.
While balancing homogeneous sources is well known, the goal of this paper is to balance the demand of
heterogeneous resources with a convex cost function as well as to encourage consumers to reduce their
demand. We propose a novel dynamic protocol achieving both goals using a serial cost-sharing pricing
mechanism. The pricing-mechanism design is strategy proof and budget balanced, whereas the protocol
does not require the consumers to reveal their utility functions, and supports hierarchical organization of
consumers. We initially applied the proposed approach to a multi-agent system, where the consumers are
smart house devices controlled by a corresponding agent, which in turns negotiates the electricity price with
a smart city. Experimental results show that the proposed protocol successfully reduces the total consump-
tion and lowers the peaks while all the consumers in equilibrium are given a fair price. We further provide
experimental results indicating advantages of serial cost-sharing pricing and tariff pricing mechanisms over
the average cost-pricing mechanism that is commonly used in the resource-demand management.
Povzetek: V prispevku predstavimo mehanizem za uravnoteženje porabe heterogenih virov, katerih cen-
ovne funkcije so konveksne.
1 Introduction
The rising demand of resources such as electricity, water,
natural gas and oil along with a limited amount of natural
resources increases the costs of resource supply [1]. Con-
sumers are typically priced the same per-unit price for a
resource, regardless of whether they increase or decrease
their consumption.
Mechanisms that stimulate the use of electricity when its
generation is cheap and depreciate its consumption when
its generation is expensive, have already been proposed and
implemented in the energy sector [9].
Similar mechanisms can be implemented for any other
resource [13], although this is rarely done, since electricity
demand management is recognised as the main resource-
demand management problem. The main reason for this is
that it is inefficient to build new power generation plants in
order to meet peak demands, while there are parts of the
day, when there is barely any demand at all. While other
reports have focused on shifting the load from one part of
the day to another [9, 12], we propose a mechanism to en-
courage consumers to reduce their consumption in general.
A resource-demand management mechanism that sets
the same price for every consumer in advance does not
fully exploit demand-management capabilities. Dynamic-
pricing mechanisms were proposed in [3], [10] and [4]. In
those papers, the price for the next time unit changes dy-
namically and is the same for every consumer. Addition-
ally, such mechanisms require the consumers either to re-
port their utility function (which raises privacy concerns)
or to leave the decision about consumption to the con-
sumers themselves (which is not as efficient as it could be).
A dynamic-pricing mechanism that negotiates prices with
every individual was proposed in [2]. However, their ap-
proach assumes that the consumers will report their con-
sumption truthfully, which is not always the case. Dis-
tributed real-time electricity pricing mechanism was intro-
duced in [8], however, due to intense communications re-
quirements between the suppliers and consumers it takes
time to calculate the optimum solution. Truth incentive de-
mand management was proposed in [5]. The mechanism
described in [5] has several desirable properties besides be-
122 Informatica 41 (2017) 121–128 J. Zupančič et al.
ing truth incentive: it is proved to converge to a Nash equi-
librium, it is budget balanced (the total cost of resource
provision equals the total cost the consumers pay), and it
reaches the global optimum (minimal peak-to-average ra-
tio). However, it still raises some issues: the resource con-
sumption of every consumer is known to everyone, con-
sumers are only encouraged to shift their load to different
parts of the day and not to reduce consumption, smaller
consumption does not result in smaller per-unit price, and
real-time pricing requires price prediction capabilities.
Further, the only known general technique for designing
truthful and approximately budget-balanced cost-sharing
mechanisms with good efficiency or computational com-
plexity properties is due to Moulin [6].
In some ways similar to our approach is the trading mar-
ket for the smart electricity grid described in [11]. The
authors proposed continuous double auction for the smart
electricity grid. However, their mechanism is not scalable
nor it is dynamic. Further, [14] addresses similar problem
– dynamic resource-demand management and it uses se-
rial cost sharing mechanism. However, no comparison with
other mechanism were made.
To be successfully implemented in practice, a mecha-
nism has to address all the above mentioned shortcom-
ings. It has to be easily understood by consumers, it should
change the prices dynamically and adapt them fairly to
each individual according to the amount of resource con-
sumed. Further, the mechanism should be truth incentive
and should encourage lower resource consumption with-
out asking the consumers to publicly reveal their demand
or utility function. We propose a novel protocol that pos-
sesses those properties. It is based on the following three
key ideas:
1. It ensures consumer satisfaction, by using a fair per-
unit pricing of the resource;
2. It has built-in truth incentive and budget-balancing;
3. To preserve privacy, consumers do not have to reveal
their utility functions.
The rest of the paper is structured as follows. In Section
2, we describe the problem domain to which our mecha-
nism can be applied. In Section 3, we describe the pro-
posed mechanism and its properties. Results from applying
the mechanism to electricity demand management domain
are presented in Section 4, while Section 5 concludes the
paper with summary and discussion.
2 Problem domain
Our mechanism can be applied to a general resource-
demand management problem (when there are many con-
sumers that consume convexly priced resource), but for il-
lustrative purposes we will make use of smart city domain
to provide examples. The proposed protocol applies to
a multi-agent system, where devices and appliances that
consume a resource in one consumer unit (e.g., a smart
house) are controlled by one agent – house representative
agent. House representative agents interact with the nego-
tiator agent forming a multi-agent system (Figure 1).
Agents participating in the proposed protocol are defined
in Definition 1.
Definition 1. Let A be a set of all agents in the system.
Let dij be a particular device j in a house i. Let hri ∈ A
be a house representative agent that is responsible for co-
ordinating all devices dij in the house. Therefore, house
i consists of devices and a representative agent and is de-
noted as Hi.
Let nr ∈ A be a negotiator agent that is responsible for
fair distribution of resource r per-unit prices.
Every house representative agent interacts only with the
resource negotiator agent. House representative agents do
not interact among each other.
nr
hr1 d11
d12
d13
hr2
d21
d22
d23
hr3
d31
d32 d33
d34
H1H2
H3
Figure 1: Multi-agent system example to which the proto-
col applies
We make the following assumptions for our problem do-
main.
House representative agents can control the devices in
the house. House representative agents have the ability to
switch the devices on and off. Further, agents can lower
the power consumption of the device in finite number of
discrete steps if gradual decrease of power consumption is
supported by the device.
House representative agent strives to maximise con-
sumer satisfaction. It is a primary goal of the house repre-
sentative agent to consider users preferences, when nego-
tiating with the resource negotiator. It is unacceptable for
the inhabitant to pay a higher per-unit price for a resource
than he was prepared to.
Cost function of a resource is convex, strictly increas-
ing and non-constant. Examples of functions of this kind
Dynamic Protocol for the Demand Management of. . . Informatica 41 (2017) 121–128 123
are frequently used quadratic cost function and piecewise
linear cost function [5].
3 The Protocol
The protocol defines a resource-demand management
mechanism that enables the matching of resource demand
with resource supply. Supply varies because resource
provision usually depends on some independent variables
(weather, fuel costs, taxes etc.). We assume that for one
protocol run the supply and per-unit prices can be accu-
rately defined in advance. Demand varies due to the supply
and the consumer preferences. We assume that consumer
preferences can be accurately defined in advance before
each protocol run. Demand then depends only on the sup-
ply and using the proposed protocol the equilibrium can be
reached.
The protocol is designed by the following requirements:
– negotiation finishes in a finite number of steps,
– consumers are motivated to reduce resource consump-
tion by the protocol itself,
– the protocol is truth incentive,
– the pricing is fair,
– the total cost of resource provision equals the total
cost the consumers pay, i.e., the protocol is budget-
balanced.
According to those requirements and design the pro-
posed protocol is a variant of a Moulin mechanism [6] -
iterative fair sharing procedure.
3.1 Informal Protocol Description
Informal description of the protocol follows. First the smart
house owners define the maximum per-unit price they are
prepared to pay for every operating state of every device.
For instance, smart house owner can have a dish washer, a
washing machine and a heat pump installed in the house.
For the dishwasher the owner is prepared to pay the per-
unit price of 0.12 EUR/kWh, for the washing machine 0.10
EUR/kWh, for the heat pump at maximum power 0.30 EU-
R/kWh since keeping the home warm is more important
than washing the clothes, and for the heat pump at half
of the maximum power 1.0 EUR/kWh since in the case of
high resource demand the per-unit price will be higher but
the owner is prepared to pay the high per-unit price in order
to maintain the border level of comfort. Other resources,
e.g. water or gas can be added in the same way. By stat-
ing all the per-unit prices the user defines the strategy of
her house representative agent and the agent can partici-
pate in the mechanism defined by our protocol on behalf of
the house owner. Since the house representative agent re-
sides in the house and does not share the information about
prices, the character of the home owner or her representa-
tive agent remains private information.
Our protocol is dynamic, meaning that it is repeated after
some period of time has elapsed from the previous run. For
some resources the per-unit price is expected to be more or
less stable over time and does not change very often. Time
elapsed between protocol runs in these cases can therefore
be longer. However, per-unit prices of some resources can
fluctuate rapidly. For instance, the per-unit price of electric-
ity can change rapidly with the increasing use of renewable
resources for its generation depends on the wind speed, so-
lar radiation, water flow rate etc. Since those variables are
hard to predict a day in advance, an hour or half an hour
time intervals are assumed and the protocol must be able
to support a large number of houses and perform all the
calculations in a short period of time.
When a new time interval in which the decision about
consumption has not been reached yet approaches, the ne-
gotiator agent triggers the protocol. First, it gathers the
information about resource supply from the suppliers in or-
der to construct a resource cost function. Then it asks every
house representative agent it has connection to as to what
is her preferred consumption. Since this is at the start of
the protocol and no price has been set yet, every house rep-
resentative agent responds with the highest required con-
sumption. The negotiator agent then computes the per-unit
prices individually for every house using some kind of a
pricing mechanism and informs the house representative
agents of the per-unit prices. It then again asks every house
representative agent about their required consumption. The
house representative agent can then either agree with the
per-unit price and it responds with the same consumption
as before or it can reduces the required consumption in the
hope of a lower per-unit price. The proposed protocol guar-
antees that the per-unit price for lower consumption is ei-
ther lower or in the worst case equal to the per-unit price
of higher consumption. The cycle ask for consumption -
respond with consumption - compute per-unit prices is re-
peated until every house representative agent is satisfied
with the per-unit price and at that point the protocol run
ends and an agreement is made stating that each house will
consume the specified amount of the resource in the ap-
proaching time interval at the per-unit price that was set by
the negotiator agent and agreed upon by the house repre-
sentative agent.
When new time interval starts the house representative
agents turns the devices and appliances in the home ON
or OFF, depending on the agreements made. Every addi-
tional consumption of the resource, beyond the agreed one,
is paid for at uniform maximal per-unit price that covers
the costs of resource providers for additional amount of re-
source provision.
3.2 The Protocol Definition
During the run of the protocol many rounds take place with
house representative agents reporting their desired con-
124 Informatica 41 (2017) 121–128 J. Zupančič et al.
sumption to the negotiator agent and negotiator agent com-
puting per-unit prices for every house representative agent.
The per-unit prices are computed individually for every
consumer using Serial cost sharing mechanism [7]. The Al-
gorithm 1 determines fair per-unit price for every consumer
– lower consumption results in lower costs. Since the re-
source is convexly priced, it also results in lower per-unit
pricing.
Algorithm 1: Serial cost sharing function
Serial_Cost_Sharing(f, c):
N = length(c)
sort c in increasing order
p = []
for i in range(N):
p[i] = cost(i, f, c)/c[i]
sort p in order to
match original order of c
return p
Algorithm 1 takes two inputs: resource cost function,
denoted as f , and desired consumption vector, denoted as c,
and returns a vector of per-unit prices p. Individual cost is
computed by a function cost(i, f, c) that takes as inputs the
following parameters: consumer i, a sorted consumption
vector c, and resource cost function f . Function cost is
defined recursively in Eq. (1).
cost(0, f, c) = 0
cost(i, f, c) =
f
(∑i−1
j=1 c[j] + (N + 1− i) · c[i]
)
N + 1− i
− (1)
−
∑i−1
j=1 cost(j, f, c)
N + 1− i
whereN is the number of consumers participating in nego-
tiation.
The cost function just needs to be convex, whereas any
source, homogeneous or heterogeneous (water, gas, se-
curity and similar), can be added with the corresponding
weights:
cost =
∑
resource
ωresource · costresource, (2)
where resource is a specific resource type, costresource
is the cost of that resource type and ωresource is the weight
of that resource in the resource portfolio.
If any of the consumers does not agree with the per-unit
price it receives and wants to further reduce its consump-
tion anticipating a lower per-unit price, another round of
negotiation takes place. In the following round, the de-
sired consumption of individual consumer can be the same
or lower. Negotiation stage ends when the demand is the
same for every consumer in two consecutive rounds.
It is a requirement of the protocol that a consumer does
not respond with strictly higher desired consumption than
in the previous round.
3.3 The Protocol Properties
In this subsection we evaluate the proposed protocol ac-
cording to the requirements stated at the beginning of the
Section 3.
Lemma 1. Assume there is a finite number of consumers
and each consumer is in possession of finite number of de-
vices or appliances and every device or appliance has finite
number of resource consumption levels. Then the protocol
finishes in a finite number of steps.
Proof 1. Assume that the protocol does not finish in finite
number of steps. That means that consumption vector (a
vector of consumptions of all the consumers) is different
in each two consecutive rounds. That means that in every
round at least one of the consumers reported strictly lower
consumption than before. Since we assumed infinite num-
ber of steps at least one consumer can strictly lower her
consumption infinite times. This contradicts the premise
from the lemma. Therefore, our assumption is wrong and
the protocol finishes in finite number of steps. 
Definition 2. Let f be a resource cost function, so that
f(x) is the total cost for the provision of x amount of re-
source. Let cost be a pricing mechanism used to deter-
mine the cost for every resource consumer. Let c be a
consumption vector, denoting consumptions for all the N -
consumers participating in the mechanism. The pricing
mechanism is budget-balanced iff
f
(
N∑
i=1
ci
)
=
N∑
i=1
cost(ci).
The following two lemmas are a corollary of [7].
Lemma 2. The pricing mechanism used in the proposed
protocol is budget-balanced.
Proof 2. Let N be a number of consumers, let c be a con-
sumption vector where ci-s are increasingly ordered, let f
be a resource cost function and let the cost be a serial cost
sharing mechanism that is used in the proposed protocol.
The costs of the largest consumer can be computed as
cost(N, f, c) =
f
(∑N−1
j=1 c[j] + (N + 1−N) · c[N ]
)
N + 1−N
−
−
∑N−1
j=1 cost(j, f, c)
N + 1−N
= f
 N∑
j=1
c[j]
− N−1∑
j=1
cost(j, f, c).
Dynamic Protocol for the Demand Management of. . . Informatica 41 (2017) 121–128 125
Therefore,
N∑
j=1
cost(j, f, c) = f
 N∑
j=1
c[j]
 .

Definition 3. Let per-unit price for the consumption x and
using pricing mechanism p be denoted by p(x). Let con-
sumers A and B consume aA and aB amount of resource,
respectively. A pricing mechanism p is fair iff for each aA
and aB so that aA ≤ aB , p(aA) ≤ p(aB) also applies.
Lemma 3. The pricing used by the proposed protocol is
fair.
Proof 3 (sketch). Since resource cost function is convex
and increasing, its derivative is also increasing. Since the
derivative of the cost function represents the marginal price
(per-unit price) of the resource provision, the per-unit price
of the resource is higher when more of it is used. Serial cost
sharing mechanism divides the cost in such a way that for
the same amount of resource the consumers pay the same
per-unit price and any additional consumption is paid at a
different price. Since the per-unit price of additional con-
sumed resource is higher, due to the increasing and convex
cost function, so is the final per-unit price of a larger con-
sumer. Therefore if aA ≤ aB then p(aA) ≤ p(aB). 
Definition 4. Let the pairs {(ci, pi)}i represent the (con-
sumption, price) pairs by which the consumer can state
his preferences: for the consumption of ci amount of re-
source the consumer is prepared to pay pi per-unit price.
Let A = (cA, pA) and B = (cB , pB) be two possible out-
comes of the protocol. The consumer preferres A over B
iff
1. cA > cB and pA ≤ pi where ci = cA, or
2. cA = cB and pA < pB ≤ pi where ci = cA.
In other words, the consumer prefers to consume the
maximum possible amount of resource she can afford. In
this case the following lemma applies.
Lemma 4. Assume the definition of preference as in Defi-
nition 4. Further, assume the consumers are infinitely risk-
averse. Then the proposed protocol is truth incentive and
strategy-proof.
Proof 4 (sketch). During the protocol the consumer can
strategize by responding with the same level of consump-
tion in two consecutive rounds of the protocol although she
can not afford the per-unit price (according to her real pref-
erences). That is because the per-unit price of the resource
can lower when all other consumers lower their consump-
tion. If the per-unit price lowers enough the per-unit price
could become affordable for the strategic consumer. How-
ever, since no consumer has the full knowledge of all the
consumers preferences or the resource prices, there is a
positive probability that the per-unit price will not become
affordable. If the consumer is infinitely risk-averse, she
wants to avoid the worst possible results and therefore, she
will not strategize. In every round the consumer will re-
spond according to her preferences, which will guarantee
a good outcome for her. Therefore, the proposed protocol
is truth incentive and strategy-proof. 
4 Experimental Results
Application of the protocol in electrical energy consump-
tion domain is presented in this section. We carried out
the experiment to examine the amount of electrical energy
consumed and the average per-unit price of the electrical
energy when using the proposed protocol.
4.1 Model description
While there are many variants for house representative
agent implementation, the important properties of the
house representative agent are:
1. It can reason about desired resource consumption in
every round of the protocol, and
2. it can decide whether the offered per-unit price is sat-
isfiable for the house.
In our experiment house representative agent possess the
information about the electrical energy consumption of the
devices and the information about consumers settings –
maximal per-unit price for operating each device. This in-
formation is private to each house representative agent and
sent neither to other house representative agents nor the re-
source negotiator.
The parameters for each house in our experiments were
the following:
– Number of consumption levels: uniformly drawn in-
teger number from the interval [3, 15],
– Consumption level size: uniformly drawn real value
from the interval [0.1kWh, 1.5kWh],
– Acceptable per-unit price for every consump-
tion level: drawn from normal distribution with
mean 0.15EUR/kWh and standard deviation of
0.055EUR/kWh.
A convex, strictly increasing piece-wise linear cost func-
tion was used for electrical energy pricing (suggested in
[5]). It can be represented by a list of tuples (Bi, pupi),
where Bi is the amount of energy available for the per-unit
price of pupi. To obtain a convex function, we assume that
lower priced energy is supplied first.
The parameters for cost function were the following:
– Number of tuples (Bi, pupi): 50 tuples were used;
– Block sizes: on average each block was four time the
size of total minimal level consumption;
126 Informatica 41 (2017) 121–128 J. Zupančič et al.
– Block per-unit prices: drawn from normal distribution
with mean 0.3EUR/kWh and standard deviation of
0.055EUR/kWh.
4.2 Electricity Consumption Experiment
In order to make a comparison of the serial cost sharing
mechanism used in our protocol to standard cost sharing
mechanism, the average cost sharing mechanism was also
implemented [5].
Definition 5. Let f be a resource cost function and let c
be a consumption vector. The average cost sharing mecha-
nism for consumer i is defined as:
averageCost(i, f, c) =
f
(∑N
k=1 c [k]
)
∑N
k=1 c [k]
· c [i] .
The protocol was run 2000 times with serial cost shar-
ing mechanism and average cost sharing mechanism for ev-
ery number of houses we specified – specifically 100, 500,
1000, 1500, 2000. The main variables we observed were
the percentage of total possible consumption that was re-
alized after the protocol run and the average per-unit price
the consumers have to pay.
Since there was a large difference between the two pric-
ing mechanisms a third mechanism was implemented that
combines the average and serial cost sharing mechanisms.
Definition 6. Let T be a number of possible different per-
unit prices of the resource. Let c be a consumption vector,
where consumptions are orderer from smallest to largest,
and let N be a number of consumers. Now group the con-
sumers into T equal sized groups where in t-th group there
are all consumers with index from (t−1)·
⌊
N
T
⌋
+1 to t·
⌊
N
T
⌋
.
Total consumption for every group can be computed and
new consumption vector for groups can be assembled
cT =bNT c∑
i=1
c [i] ,
2·bNT c∑
i=bNT c+1
c [i] , . . . ,
N∑
i=(T−1)·bNT c+1
c [i]
 .
For each group the costs can be calculated using the se-
rial cost sharing mechanism and a cost vector can be ob-
tained costT = (cost1, cost2, . . . , costT ). The per-unit
price for consumer i in group t can be calculated using the
average cost sharing mechanism
pricei =
costt∑t·bNT c
k=(t−1)·bNT c+1
c [k]
· c [i] .
We call this mechanism a tariff pricing mechanism.
We have made the same experiments as before using the
additional pricing mechanism, where we grouped the con-
sumer into two equal sized groups. The final results are
100 500 1000 1500 2000
numberOfHouses
0.00
0.05
0.10
0.15
0.20
a
v
e
ra
g
e
P
ri
c
e
protocolType
average
serial
tarrifs
Figure 2: Bar plot with number of houses and average per-
unit prices on axes
100 500 1000 1500 2000
numberOfHouses
0
5
10
15
20
25
n
o
n
Z
e
ro
C
o
n
s
u
m
p
ti
o
n
protocolType
average
serial
tarrifs
Figure 3: Bar plot with number of houses and percentage
of possible total consumption on axes
presented in Figures 2, 3 and 4. Every point or a bar repre-
sents the average of 2000 runs of the protocol.
From the experiments we concluded that a number of
houses has a negligible effect on the observed variables
when model like ours is used. The largest effect has the
type of the protocol used: protocol with serial or average
cost sharing mechanism. Since the consumers are different
and some are prepared to pay more for the same amount
of resource the consumption increases when using the pro-
tocol with serial cost sharing. It is the advantage of the
serial cost sharing to produce personalized per-unit prices
according to the amount of resource used. For that reason
also, the average per-unit price increases when using se-
rial cost sharing. However, every consumer still receives
the amount of the resource she is prepared to pay for. And
according to definition 4 maximal consumption within the
limits is the preferred result. Therefore, serial cost sharing
Dynamic Protocol for the Demand Management of. . . Informatica 41 (2017) 121–128 127
Figure 4: Scatter plot with average per-unit prices and per-
centage of possible total consumption on axes
mechanism is the preferred mechanism.
By using only two groups the results of the protocol us-
ing tariff pricing are much more similar to the protocol us-
ing serial cost sharing than the average pricing. Consump-
tion and average per-unit price increase due to the flexible
pricing mechanism used.
5 Discussion & Conclusion
We have proposed a truth incentive mechanism that encour-
ages the reduction of consumption of convexly priced re-
sources and can be easily understood by consumers. Fur-
ther, the mechanism scales linearly with the number of con-
sumers, since the per-unit prices can be easily computed.
The novel protocol ensures consumer satisfaction when
the consumer is truthful and deals also with heterogeneous
resources. But there are also some problems associated
with our approach. First, since one can not predict the con-
sumption of others and does not know the exact resource
cost function, one can not predict the possible outcome
of the protocol. Therefore, it is not rational to agree to a
higher per-unit price than one is willing to pay. Second,
the algorithms that deals with homogeneous resources [6]
find optimal solutions and have several desired properties.
Our algorithm meets some of the desired requirements but
needs several rounds to achieve that, however the number
of cycles is still less than in general Moulin mechanisms.
Since serial cost sharing was used the proposed mech-
anism is budget-balanced. The stopping of the protocol
was discussed in Subsection 3.3. However, one of the main
assumptions used to prove the stopping was that after ev-
ery round of the negotiation the consumer does not de-
mand strictly higher desired consumption. Although not
addressed in the proposed version of the protocol, addi-
tional mechanism could be implemented to release this as-
sumption and still maintain the the stopping property.
Further work will include the analysis of the consumer
behaviour when the protocol is applied repeatedly with the
same consumers and the same resource negotiator. Addi-
tionally, other agent organization structures will be consid-
ered in order to enable the protocol implementation in real-
life scenarios.
The proposed protocol demonstrates various desired
properties beside being truth incentive and scalable. Due
to its efficient resource reduction capabilities it is consid-
ered to be applicable to the sustainable smart city domain.
6 Acknowledgements
The research was sponsored by ARTEMIS Joint Undertak-
ing, Grant agreement No. 333020, and Slovenian Ministry
of Economic Development and Technology.
References
[1] Edward B Barbier. Economics, natural-resource
scarcity and development: conventional and alterna-
tive views. Routledge, 2013.
[2] Frances Brazier, Frank Cornelissen, Rune Gustavs-
son, Catholijn M Jonker, Olle Lindeberg, Bianca Po-
lak, and Jan Treur. Agents negotiating for load bal-
ancing of electricity use. In Distributed Computing
Systems, 1998. Proceedings. 18th International Con-
ference on, pages 622–629. IEEE, 1998.
[3] Ahmad Faruqui and Sanem Sergici. Household re-
sponse to dynamic pricing of electricity: a survey of
15 experiments. Journal of Regulatory Economics,
38(2):193–225, 2010.
[4] Koen Kok. Multi-agent coordination in the electric-
ity grid, from concept towards market introduction.
In Proceedings of the 9th International Conference
on Autonomous Agents and Multiagent Systems: In-
dustry track, pages 1681–1688. International Founda-
tion for Autonomous Agents and Multiagent Systems,
2010.
[5] A-H Mohsenian-Rad, Vincent WS Wong, Juri Jatske-
vich, Robert Schober, and Alberto Leon-Garcia. Au-
tonomous demand-side management based on game-
theoretic energy consumption scheduling for the fu-
ture smart grid. Smart Grid, IEEE Transactions on,
1(3):320–331, 2010.
[6] Hervé Moulin. Incremental cost sharing: Characteri-
zation by coalition strategy-proofness. Social Choice
and Welfare, 16(2):279–320, 1999.
[7] Herve Moulin and Scott Shenker. Serial cost sharing.
Econometrica: Journal of the Econometric Society,
pages 1009–1037, 1992.
128 Informatica 41 (2017) 121–128 J. Zupančič et al.
[8] Toru Namerikawa, Norio Okubo, Ryutaro Sato,
Yoshihiro Okawa, and Masahiro Ono. Real-time pric-
ing mechanism for electricity market with built-in in-
centive for participation. IEEE Transactions on Smart
Grid, 6(6):2714–2724, 2015.
[9] Sarvapali D Ramchurn, Perukrishnen Vytelingum,
Alex Rogers, and Nick Jennings. Agent-based con-
trol for decentralised demand side management in
the smart grid. In The 10th International Confer-
ence on Autonomous Agents and Multiagent Systems-
Volume 1, pages 5–12. International Foundation for
Autonomous Agents and Multiagent Systems, 2011.
[10] Pedram Samadi, Hamed Mohsenian-Rad, Robert
Schober, and Vincent WS Wong. Advanced de-
mand side management for the future smart grid using
mechanism design. Smart Grid, IEEE Transactions
on, 3(3):1170–1180, 2012.
[11] Perukrishnen Vytelingum, Sarvapali D Ramchurn,
Thomas D Voice, Alex Rogers, and Nicholas R Jen-
nings. Trading agents for the smart electricity grid.
In Proceedings of the 9th International Conference
on Autonomous Agents and Multiagent Systems: vol-
ume 1-Volume 1, pages 897–904. International Foun-
dation for Autonomous Agents and Multiagent Sys-
tems, 2010.
[12] Perukrishnen Vytelingum, Thomas D Voice, Sarva-
pali D Ramchurn, Alex Rogers, and Nicholas R Jen-
nings. Agent-based micro-storage management for
the smart grid. In Proceedings of the 9th Interna-
tional Conference on Autonomous Agents and Mul-
tiagent Systems: volume 1-Volume 1, pages 39–46.
International Foundation for Autonomous Agents and
Multiagent Systems, 2010.
[13] Fredrik Wernstedt, Paul Davidsson, and Christian Jo-
hansson. Demand side management in district heat-
ing systems. In Proceedings of the 6th international
joint conference on Autonomous agents and multia-
gent systems, page 272. ACM, 2007.
[14] Jernej Zupančič, Damjan Kužnar, Boštjan Kaluža,
and Matjaž Gams. Two-stage negotiation protocol
for lowering the consumption of convexly priced re-
sources. In Proceedings of the 2014 Workshop on In-
telligent Agents and Technologies for Socially Inter-
connected Systems, page 5. ACM, 2014.