Overview
The DiMalloc project (acronym for techniques for DIstributed Markets, negotiation,
and resource ALLOCation) focus on market mechanisms for resource allocation
in computer systems, with energy systems management as main application.
DiMalloc is sponsored by NUTEK (as
part of the PROMODIS program) and EnerSearch
AB.
The project was started July 1st, 1998 and the first phase consists
of 18 months.
Participants
| Project leader: |
Prof. Arne
Andersson, Computing
Science Department, Uppsala
University |
| Senior researcher: |
Dr. Fredrik
Ygge, EnerSearch
AB. |
| PhD students: |
Per
Carlsson, Computing
Science Department, Uppsala
University and Department
of Computer Science, Lund
University. |
|
Maria Karlsson, Computing
Science Department, Uppsala
University and Department
of Mathematics, Statistics and Computer Science, Växjö
University. |
|
The participating organizations are EnerSearch
AB, Uppsala University,
Lund University, and the University of
Karlskrona/Ronneby.
Main Scientific Content
The interest in automated negotiation among different parties represented
by computational agents is continuously increasing. We study such negotiations
from mainly two aspects; (i) how markets can facilitate efficient resource
allocation in distributed environments, and (ii) fundamental properties
of market protocols. These two aspects are discussed further below. Both
aspects are of vital importance for tomorrow's energy utilities. The research
includes theoretical studies as well as field tests in close cooperation
with EnerSearch AB and its owners.
Efficient Market Protocols and Resource Allocation Algorithms
On a deregulated market, many energy utilities have declared an interest
in providing services rather than merely pure energy. One such service
could be to provide a comfortable indoor temperature in a public building.
As prices vary significantly during a day (cf. the balance services of Nord
Pool, the major market place for electricity in the Nordic countries), it
is important to consider energy costs while providing the services. Furthermore,
for the same reason it often makes sense for the utility to establish contracts
with customers allowing the utility to temporarily control loads at the
customer site for some compensation to the customer. As energy utilities
typically support a large number of customers (up to some millions), the
management of the energy system is a complicated optimization problem, both
from a performance point of view (in terms of communication and computation)
and from a software engineering point of view. The situation is further
complicated by its dynamic nature; customers may leave and new customers
may enter.
We have already developed new efficient algorithms for settings of this
kind as well as a conceptual framework where the energy system is modeled
as a multi-agent system. Some initial field tests have already
been performed and others are currently under implementation.
The main research issues of this project are:
- further development of algorithms for distributed markets and distributed
resource allocation in general;
- further development/incorporation of more complicated load models;
Knowledge transfer of current state of the art and new results to the industrial
partners will be fascilitated through publication of reports and papers,
and through seminars and/or workshops.
The following demonstrators/prototypes are planned:
- field tests of the developed methods;
- building of prototypes of tools for integration of the load management
system into the control rooms of power utilities;
Self-Interested Agents in Market Negotiations
Another important field of study is negotiation protocols for markets with
divisible goods, such as electricity. Interesting properties of such protocols
is how vulnerable they are to manipulation (i.e. "false" bids) and how efficient
the outcome is. For example, the major current electricity spot-market for
Nordic countries (Nord Pool) is based on bids treating each of the coming
24 hours separately. Practically, this means that when a utility makes a
decision for the amounts to buy or sell at different prices at a future
hour, this bid has to be based on some predictions of other prices, i.e.
the decision of whether or not to start a spare plant is typically based
on some expectation on the average price the coming day. Presumably the
efficiency of the market can be improved by letting the utilities use more
expressive bids, containing dependencies between the different hours. Furthermore,
the integration of production costs and more advanced pricing mechanisms
for transmission and distribution may also increase the efficiency of the
energy system. At the same time, aspects related to the vulnerability to
manipulation of such more extended protocols have to be well understood.
Our aim is to compare and analyse the use of different market protocols
for these (and other) applications and to investigate how agent-based electronic
markets can help in this respect. One advantage of agent-based markets is
that they can use protocols that are not manually practical (e.g. relying
on very fast iterations and/or negotiating with a large number of different
agents before closing a deal).
A clear goal for our future work is to contribute to, and possibly develop,
better market systems - for electricity trade and trade with other goods.
In particular, we believe that the expertise we will develop in designing
and understanding market protocols will be essential for Swedish power
utilities on a future deregulated (international) electricity market.
In this phase this part of the project is a pure research activity and
the construction of demonstrators and/or prototypes will depend on the
characteristics of the research results. Knowledge transfer of current
state of the art and new results to the industry will be fascilitated
through publication of papers and report and seminars and/or workshops.
A progress report to NUTEK, one of the main sponsors, is found here.
|
