Sie sind auf Seite 1von 6

Proceedings of the Third International Conference on Machine Learning and Cybernetics, Shanghai, 26-29 August 2004

AGENT-BASED RETAIL ELECTRICITY MARKET:


MODELING AND ANALYSIS
JING YU', JIAN-ZHONG ZHOU', JUNJIE YANG', WEI wu', BO FU', RONG-TAO LIAO*
!
'Department of Hydropower and Information Engineering, Huazbong University of Science and Technology, Wuhan,
Hubei 430074, P.R.Chma
'Department of Computer Science and Technology, Huazbong University of Science and Technology, Wuhan, Hubei
430074, P.R.China
E-MAIL yujing77 12@ 163.com, jz.zhou @mail.hust,edu.cn

Abstract: These challenge the researchers who are interested in this


In the past decades, power systems in the whole world area a lot.
have been facing pressure to he deregulated or restructured. Although some progess in modeling electricity
Tbis researeh has designed an agent-based retail electricity market are achieved, relatively little has been done at the
market to simulate the trading procedure in modem power retail level. Customers with freedom to choose between
systems, and utilize Colored Petri nets technology to present
the communications and cooperatino nf agents in the market. products and suppliers (and conversely) chakicterize the
We design fnnr kinds of participant agents: Power Exchanger retail electricity market"'. To obtain a satisfied result of
Agent(PXA), Power User Agents(PUAs), Retail Company scheduling the consumption (generation) in day-ahead retail
Agents(RCAs) and Independent Power Plant Agents (IPPAs). electricity market, the power consumers (producers) must
PXA is supervisor of the marketplace, in charge of opening not only have good forecast ability but also good
and dosing the market, and dealing with the trading request psychology. So how to develop an efficient trading system,
from the other agents, avoiding conflicts with constraints of how to guarantee the impartiality of trading procedure, and
power transmission grids. The other three kinds of agents how to solve the conflicts lurking'in communications and
could enter the marketplace arbitrary if it opens and freely cooperation among traders are'important problems we are
select their satisfied consumerdproviders, which are the chief
distinctionsof our design. The results of analysis show that the facing.
proposed retail electricity market could increase efficiency, Agent technology, as. a useful tool in distributed
reduce operational costs and give consumers more alternatives. artificial intelligence, attracts us greatly. Some agent-based
Though it has more latently conflicts and concurrencies, which simulators have been constructed for electricity
cause a tittle complication, we successfully deal with them by markets'"~'*]. These models have hinted at the potential of
colored Petri nets. agent-based models for the analysis of electricity markets.'
However, they either consider electricity market as auction
Keywords: marketplace, bidding round after round, or view it as
Intelligent agent; colored Petri ne@ retail electricity electronic commerce; considering electricity power as
market; hybrid method; communicationand cooperation normal commodity. That. does not fully represent the
characters of retail electricity market. In retail electricity
1.' Introduction market, information flows and business procedure bill now
take place arbitrarily .between totally separate participants.
Deregulation and restructuring of power systems has Obviously, the trading procedure contains lots of
'

been an important area for the past decades, because it will concurrencies and it is'hard to avoid conflicts among
promote competition among alternative retail electricity consumers (producers or suppliers). Therefore, we will
providers, cause providers to be more cost-efficient, apply colored Petri nets, which have a semantics that builds
eliminate high-cost providers, and result in lower retail upon true concurrency and conflict'instead of interleaving'".
electricity pnces for customers'". Thus centralized system to describe the information flows and business procedure
planning and operation management must be remodeled to among trading>agents.By the constructed colored Petri net,
adapt to the new market environment. In recent years, many we could w i l y detect the conflicts, and avoid them in
countries around the world have modified their electricity model .the interactions of agents.
supply frame, and constructed types of electricity market. However, it would be impossible for paper of any !

~7803-8403-2/04/$20.00@2W4 IEEE
95
Proceedings of the Tbird International Conference on Machine Learning and Cybernetics, Shanghai, 26-29 August 2004

reasonable length to cover all the aspects of retail electricity when IPPAs or RCAs service their customers, they should
market. As a result, this paper focus on constructing an enquire PXA whether there is any conflict with power grid
agent-based retail electricity market model to simulate the constraints, such as power transmission capability, balance
trading procedure and the behaviors of participants, which between supply and requirement, and so on.
will be described in section 2, then presents the information Because the day-ahead retail electricity market is open
flows and business procedure among trading agents via and closed at a certain clock (e.g. close at 12 noon the day
colored Petri nets in section 3. And the agent-based model before they desire delivery), the trading agents must
fusion with colored Petri net could solve the talent conflicts accomplish their trade in a limited time. Each trading agent
and concurrencies lurking in the communications and has a unique buying and selling strategy. They can exhibit
cooperation among agents successfully. With little anxious buying and selling behavior, rational behavior or
modifications, the model can be extended and implemented greedy behavior. See Figure 2 for an illustration of different
to the rest aspects as well. buying and selling strategies.

2. Agent-based retail electricity market

We assume that the reader has basic knowledge of


intelligent agent and colored Petri nets, so that we will not
describe the concepts of them in detail. In this section, we
firstly draw a sketch of our designed agent-based retail
electricity market, and then describe the object, knowledge
and planner in each agent, with pseudocode elaboration of
the infernal behaviors separately. At the end of this section
interactions among agents are mentioned.
',.-
Figure I . Architeerure sketch of Figure 2. Agent buying and selling
2.1.Architecture of the agent-based retail electricity retail electricity market strategies"
market
So different agents would learn and react in different
A simplified classification of the participants in retail way. As the time goes on, the trading agents who does not
electricity market follows ["I: accomplish trade will change their bid prices, or modify
a. Independent Power Plant (IPP) consumption curve, even give up some benefit (these
b. Power Exchange Companies (PX) decisions are made by planner mechanism of trading agent).
c. Retail Companies (RC) After PXA open the marketplace, the trading agent
d. Power Users (PU) could join it on arbitrarily time spot before it closed. The
According to these participants, we constructed four intemal behaviors of agents could be implemented by
types of agents: independent power plant agent (IPPA), pseudocode in section 2.2.
power exchange agent (PXA), retail company agent (RCA)
and power user agent (PUA). The agents work together to 2.2. Elaboration of participant agents
form the retail electricity markets with other agents.
Hereafter we will refer to PPAs, RCAs and PUAs as The agents in our retail electricity market are
trading agents. Figure. 1 shows the architecture sketch. intelligent agents, which has its objective (cost or profit
Trading agents could change their role as buyer or objective). They could update their knowledge '

seller as their electric power requirements change. In our autonomously according to their temporal (or final) trading
agent-based prototype of retail electricity market, each result, which lead them make mok satisfied trade-plan to
trading agent can be viewed as either inviting public succeed in their object. That is to say, an agent could learn
bidding for purchase-plan containing that quantity Qh of (update knowledge base), and plan its activities to reaction
electricity at an asking price APh for consume at each hour the events outside, and achieve its object. So we consider
h next day (assume that each time part in schedule is one the each agent is composed of three main mechanisms:
hour), or compete to sell their 24-hours power service-plan. object, knowledge base (KB) and planner.
In this paper, we view IPPAs as buyers and PUAs as sellers,
and that RCAs have dual nature (sell power to PUAs or buy 2.2.1 PXA
power from IPPAs), and PXA p!ays the role of maintaining
the power system works safely and stably. That is to say, Object of PXA is that economize the power grid to

96
Proceedings of the Third International Conference 00 Machine Learning and Cybemetics, Shanghai, 26-29 August 2004

deliver required quantity of electric power from trading 233 RCA


agents on condition that all the constraints are satisfied.
PXA hold a sectional tariff for the transit of electric power, Object of RCA is maximizing its profit:
which is broadcasted to every trading agent in the market PmfitofRCA= PayFmmPUA-PayTolPPA-Pay ToPX
but could only be modified by PXA itself. KB of PXA where PaytoIPPA= Knowledge base contains of tariff for
contains the data related to grid utilization status (e.g. the transit of electric power, purchase (provide) price data
resewed power delivery capability). Planner dedicate to of PUAs (IPPs). Planner dedicates to match the best pair
how to efficiently utilize every delivery line in the grid cooperators of PUA and IPPA.
according to the requested delivery quantity and avoid any
conflicts with constraints. There are lots of optimization Pseudocode of RCA:
algorithms could be applied to design the planner, .we 1. Enter the marketploee and cakulore the rest Iim it could spend on
trade (7imeForRCATmdej.
would not describe it in detail for paper length limits. 2. Addthe delivery tadffto RCA-KB
While 7imeForRCATrade >O ,do (
Pseudocode of PXA: 3. Receive donondsfrwn PUAs .
1. Open marktploce and invite trading agents. 4 Evolute andmake the mstpmfitabk PUanpPa table.
Wile 7irneofTadoO. do( 5. Smdpurchnseplan tolPPAs in mblc.
If any trading agent ewer the m r k p l o c e ( If IPPA reply occepfame (
2. Broadcast tarifl for the transit of electricpower: 1 6. Sendpmvide plon to eomsponding PUA in fable
If PUA reply a order (
If enquiry come ( 7. Enquire PXA
3. Receive; If PXA permits (
4. If enquiry not conflict constraints { 8.. Sign contract with PUA and comsponding IPP
5. Add request to PX-KBto be scheduled) ] Else reject PUA
6. Else (RejecrJ 1 1 1 ELse reiectlPPA
7. Close marketplace. IEise( '
8. Optimize power delivery schedulefor next day 9. Adj"stpumhm&pmviplan and learn
9. Learning
11. Leoming
2.22 IPPA
2.2.4 PUA
Object of IPPA is to gain more profit:
PmfitqlPPA= PayFmmPUA- PayToPX-CostlnlPP Object of PUA is minimizing its payment. ,The
Knowledge base contains data about generator units' payment is related to the price and consumption schkdule it
status (e.g. power generated and resewed), generating costs, trades in day-ahead retail electricity market.
historical purchase prices of PUAs, tariff of PX, and so on.
Planner dedicates to optimize the generation schedule and
find satisfied cooperant PUAs to maximize its profit.
PayFmmPUA=
H
2 SpotPricehx ConsumePowerh
h;l
Often the value of H is 24 or 48, denoting the division
Pseudocodeof IPPA of a day (His 24 in the example of this paper). PUA may
1. ' Enter the marketplace and calculore the rest rine it could spend on
forecast its consumption of eleitricity every period
tmdr (TieForlPPTrade).
2. Add the delivery @riffto IPPA-KB. ((ConsumePowerh, k 1 . 2 ...H)next day and after
While lbneForlPF'Trade >O , do ( accomplish trade it will get a list of price (SpotPriceh). So
3. Receive demandsf" PUAs andRCAs. its knowledge base contains the provide price data from
If demand <= genemfion cqabiliry( PPAs and RCAs, the consume schedules of itself, the
4. Evalunre and e b a e pmfitabb eollobomrors. and send them
fhe supply plan. payment data io history, and so on. Planner of PUA is also
5. Uthe collo6omtors reply with orderfoms ( an optimization mechanism evaluating the choices among
6. Enquire PXA different power suppliers and adjusting its consume curves
If PXA permits ( to be cost-effective.
7.

Pseudoeodeof PUA:
I. Enter the market and cokulafe the rest timr it could spend
on rude (.lbneForPLJATmde).
While lbneForPUATrade >O ,do(
2. BmpdeaFt demand plan LO IPPAs and RWLs in the
markelploee.
If receive service reply (

97
Proceedqs of the Third International Conference on Machine Laming and Cybernetics, Shanghai, 26-29 August 2004

3. Evolunte rhe replies and choose pmfirobleproviders (denoted by ellipse), transitions (denoted by rectangle), arcs
4. Send onlpr 10 selected providers
If receive cO"lrac1 reply [
(denoted by line with arrowhead) and color tokens. Color
5. Sign rhe eonrmcr) ' token here is a complex data structure;containing the
Else [ communication information of a corresponding agent.
6. Adjusr d e d p l o n and 1 c 3 ~ t 2 The detail-concepts are written in reference [9], we will not
got021 I
1.
8.
Exitmorker
Laming
,
2.3. Interactions of agents

From above, we can see that the agents could leim


from trading experience, and they have their own characters,
which lead each agent to different trading strategies. At the
same time, they could be affected by the actionsheactions
of the 'other agents, and they could leam from the others in
trading procedure. That is to say, each trading agents could
not only leamautonomously (update KB), but also interacts
each other especially on planner. So' the interactions of
agents must be considered as well as internal behaviors in
this agent-based retail .electricity market.
In the proposed system, the interactions among agents
are mostly stem from the agents' guesstimates of the other
competitors' purcbasdprovide price. One triumphal agent
must bring pressure to the other competitors. Those anxious
trading agents may react immediately (as Figure 2 showed,
increasddecrease their bidding soon), comparing to the
agents with the other characters. Analogously, the triumphal
agents are also thinking about how to offer more profitable
price-lists and schedules in the next day-ahead retail
market. Figure 3. Colored Peui net of retail elecI+icily market trading
Those guesstimates and how they 'affect the decisions p d U r e
are the other complex problems that we will present in
another paper later. In the following section, we will exhibit From Figure 3, we could see that designed retail
some of the interactions among the agents in trading electricity market coloredPetri net (RCM-CPN) exhibits a
procedure by,colored Petri nets. .. clear picture of communications and cooperation among
agents in the trading procedure, and from the movements'
: 3. Trading procedure of the retail electricity market of tokens, we could find conflicts (a place with more than
pictured by colored Petri nets one output arcs) easily. And in RCM-CPN, we solved those
conflicts by processes of transitions.
3.1. Trading procedure colored Petri nets In section 3.2, we will illustrate the net how to present
the communications and cooperation among agents in
From above sections, we can s,ee that the trading trading procedure with a representative case (1 PXA, 1
procedure contains . complex communication and RCA, 2 IF'PA and 4 PUAs).
cooperation of agents, which occur randomly and may
interlace with each other, so there exist lots of latent 3.2. Illustration of designed colored Petri net
conflicts. It often leads the developed agent system goes to
a'status of disorder, which confuses trading agents and they At a certain time, assumed 8:OO am, PXA initiates a
could not trade-off or even could not accomplish their day-ahead market by advertising that it is.going to conduct
trading plan. Therefore, we adopt an intuitive modeling tool, a free trade and specifying the close time, assumed 12:OO
colored Petri nets, to picture the trading procedure (see
figure 2).
Colored Petri nets mainly composed of places

98
Proceedings of the Third International Conference on Machine Learning and Cybernetics, Shanghai, 26-29 August 2004

noon. All trading agents interested in participating in the UOaerRCA may concurrently occurs. In this case, only
trade can then join the marketplace and submit bids (put PUA2 has option that select whether IPPl or RCA, for
their tokens to corresponding places). For example, if there <psI.ud2> and <rsz,udz> simultaneously existing in
are four PUAs, tow IPPs, one RCA entered the electricity OrderPlan. That is to say, P P 1 and RCA select to service
marketplace at the same time, then PXA broadcast the tariff the same power consumption schedule. Now the conflicts
of delivery to them, and put the constraints token C into occur. But in our design, PUA could choose to be served by
Constraints place. PUAs offer their schedule containing who's service plan, not a passive power consumer as in the
that quantity Qh of electricity at an asking price APh for traditional power market, so the conflicts are solved by
consume at each hour h next day (those information is PUA's choice - either select one of them to satisfy all its
contained in color token D),such that they put four D demand or select one supply part of power and the other
tokens ( a d l > , ad2>,<udp and aG)in PowDemand supply the rest. For simplify, we assume that PUA2 is a
place. The seller agents, two IPPAs and a RCA, put two S small user in the power systems, it can but choose one
tokens (<PSI>, <psp) and one S token (asl>)into provider. Supposing that PUAZ and PUA4 finally ordered
SuppPlnn place and PurchPlan place separately. S token the service from RCA, PUAl ordered from P P I ,
contains the service information. That is the initial UOderIPP and UOderRCA concurrently occurs- take away
marking"' (&) of this colored Petri net. <rs2, udp. u s 2 , U@. <psi, u d p , while deposit a d z , rsp,
Now, Broadcast transition is enabled"' and could <U&. rs2> into PmvidPlan, and a d l , ps+ is deposited into
occur. After Broadcast transition occurred, D in its input GenPlan.
place (PowDemand) are all took away, and these four D Then, MatchGen occurs, depositing two tokens ad2,
tokens are duplicate deposited in the output places p s p and a&,ps2> into GenPlan. These tokens are oriented
(SuppPlane and ServPlan). Then RCA and IPPs could from a d 2 . rsp and <U&. n2>, processed by MatchGen, for
receive the demand. They will evaluate and choose to,hid avoiding letting out the business secrets. All the tokens
which ones, that is to say, to service whom. That is to say, (generation plans of IPPAs) must be checked by PXA. Here
three transitions (PurchReq, ZPPAChoose and PRejectPUA) we only take account of the grid transmission capability
are enabled. Supposed that RCA choose to serve IPP2 and constraint as an instance ( C represents the capability,
IF'P4, then PuchReq occurred - take away cud2>, <U&>, placed in CheckCaPa).
<IS> and deposit processed tokens a d z > and ad.+into Next, we assume that the order from PUA4 to RCA
SuppPlan. The processed tokens contain the purchase has some illegalities, then ZllegalRP could occur- take away
demand of RCA (purchase quantity Qh of electricity at a <I&, psp. deposit ad.+to MatchU&P and <ps+ to
purchase price PPh for delivery at each hour h). And then if ZPPA-KB. Then uQ> is decomposed by RCA, and
IF'PAI choose to service RCA, and IPP2 choose service deposited into the knowledge base of RCA (RCA-KB)and
PUAl and PUA2, then ZPPAChoose occurs - take away PUAs (PUA-KB) separately. On the other side, the
a d l > , a d z > , ad.+, adz>, <r&>2, <psl>,<ps2> and remainder orders are successfully checked by CheckCapa
deposit <psi, u d p , <psi, udz> into Orderplan and deposit transition, which is utilized to ensure that if the power
<ps2, r d p , <ps2, r&> into ServPlan. Then, PRejectPUA delivery request is over load. After that, IPPl and IPP2 will
occurs - take away the frustrated a d 3 > and deposit it into sign contract with PUAl and RCA separately, and the
PUA-KB. It will be adjusted by planner of PUA, and then relative transmission capability and < C ~ , ~ Z
whose
>,
be broadcasted to all sellers again. subscript denotes the relation ship of power consumer and
Now, we find a concurrency of PRejectPUA and provider) is subtracted (take away from CheckCapa).
PChoose in the net. But they could not conflict with each Finally, RCA match the pair of PUA and IPPA, and sign a
other, because they are under the latent process (choice) of contract with corresponding PUA2.
IPPAs. Up to now, part of the trading contents (<udl> and
Then, there are enough tokens in ServPlan, MatchPUA a d z > ) are accomplished, the remainder parts will be
could occur - take away <ps2. rdp. <ps2, r&> and deposit adjusted by their trading agents and then be traded again. In
corresponding as2, u d d , as*,ud4> to OrderPlan. a d l > fact, as our proposed example process goes on, the other
and < p s p are toke away, and deposit a <psl,udl> into trade contents are executed in parallel. We only elaborate
Orderplan. So PUAI, PUA2 and PUA4 could evaluate the one assumes occur sequence, which could be expressed by
service plan from RCA and P P I , then choose the most the following:
satisfied for themselves and send orders to each selected &:P,,Denond=(<ud,>.<ud2>.<ud3,<ud3J
power supplier - URejectZPP, URejectRCA, UOderZPP and SuppPlon=(<ps,>.<psz>)
ServPlnn=(us,>)
M i PowDenond=(<uds-,<)
EritMorker= (<ud,>.cud,>, <pst>,<ps~>,us,>)

99
Proceedings of the Third International Conference on Machine Laming and Cybernetics, Shanghai, 26-29 August 2004

occur step f” MOto I&: implement the trade in a distributed way; Further studying
S,=(Bm?dcnstJ
SZ=[PurchReq. IPPAChoose, PRejectPUA)
the other latent problems in communication and
S~=(MorchPUAQUALeom,UOderlPP,OderRCA) cooperation of agents, then implementing a more robust
S.=(MatchCen, ReqCheck) trading system.
Ss=(IllegolRP. SignPUA, SignRC4 1 -
&={MarchP&LI.IPPALeam, RSginPUA)
SFIRCALeam.PUALeamI References
The sequence is only one of the numerous possible
trading sequences, which is not controllable and arises [l] Michael L. Walden, “Will Retail Electricity Market
randomly. Each time our agent-based market model Deregulation be a Boon or Bane for Consumers”,
program executed, the final trading outcome variant. But Consumer Interests Annual, Vol. 46, 2000.
the more times it executed; the better the results gained - http://www.consumerinterests.org/publiJaill.
less execution time, more profit (less cost) trading agents Pdf
get (spend). [2] K. Livik, “From passive consumer to active customer,
challenges in the retail market focusing on suppliers,
4. Conclusions products, metering, billing, communication, and data
management”, Power Engineering Society winter
In this paper we have developed a prototype to show Meet&, Singapore, pp. 835-837, January 20%.
how fusion agents and colored Petri nets technology to U1 Jens Grosskalags and Carsten Schmidt, “Artificial
model a retail electricity marketplace. This demonstration is Software Agents on Thin Double Auction Markets- A
not meant to be a prototype of an actual marketplace. Human Trader Experiment”, Proceedings of the
Because of budget and schedule constraints, the scope of IEEElWIC Intemational Conference on Intelligent
this demonstration is quite limited. But the research has Agent Technology (IAT’03), Halifax, pp.40047,
shown that fusion of colored Petri nets technology and October 2003.
intelligent agents could provide a powerful problem-solving [41 Palo Alto, EPRI - URL, “Auction Agents for the
mechanism that is ideally suited for use in the retail Electric Power Indusiry”, Available:
electricity market, and it is easily to be extended to general http://www.agentbuilder.com/DocumentationlEPRU
retail marketplace. [51 S. Harp, S. Brignone and B. Wollenherg, “ SEPIA:
Because the new restructured and deregulated power Simulator for Electric Power Industry Agents” , IEEE
systems will utilize network communications and the Control Systems, V01.20, No.4, pp. 53-69.2000.
Internet extensively, so our model is easy to be transferred I61 1. Bower and D. Bunn, “Agent-based Simulation - An
Application to the New Electricity Trading Arrangements of
and implemented in real power systems. Though our EnFland and Wales” , IEEE Transactions on Evolutionary
proposed retail electricity market has complicated Computation, Vo1.5, No.5, pp. 493-503, Oct. 2001.
information communications and processes, it could bring
[71 F. R. Monclar and R. Quatrain, ” Simulation of Electricity
modem power system powerful activities. Markets: A Multi-Agent Approach” , Proceedings of 2001
In summary this research has shown: Intemational Conference on Intelligent Systems Application
* Retail electricity market is vivid. It gives the to Power Svstems. nn. 207-212.2001.
participants more free choices and could benefit them all. 181 Felix Wu: Chris’i’eung, “A multi-agent approach to
* Agent-based model could fully simulate and the deregulation and restructuring,of power industry”,
implement the behaviors, guesstimates, discursions. and Proceedings of the 31‘‘ Annual Hawaii Intemational
interactions of each participants in our designed‘retail Conference on System Sciences. Part3(of 7). Vo1.3,
electricity market. Big Island, Jan 1998.
* Colored Petri net show good performance in [9] Jensen K., Colored Petri Nets, Vol. 1, 2nd corrected
printing, Springer verlag, Berlin, 1997.
presenting the communications and cooperation of agents.
And the trading procedure could be described in an intuitive [lo] R.H. McClanahan, “Electric deregulation impacts on
graphical way, which exhibits the concurrencies and metering, billing and customer information systems”,
conflicts latent in the procedure clearly. Rural Electric Power Conference, Little Rock, pp.
The fusion of colored Petri nets and agent approach is Al/l-A1/8, April 2001.
promising. But there are still lots of works we should do in
the future: How to fully utilize the formal analysis methods
by which propelties of CPN can be proved, thereby
improving the agents properties; Appling mobile agent to

100

Das könnte Ihnen auch gefallen