Project code: CES
Research group: Multi-agent and Adaptive Computation (SEN4)
Project coordinator: Han La Poutré

Modern society hinges on a sufficient and dependable supply of energy.
The gradual depletion of easily recoverable carbon fuels and increased
political pressure to reduce the products of combustion and dependence
on foreign natural resources, combined with a rapidly increasing
energy demand from growing economies, make the energy supply one of
the foremost technological challenges in the coming decades. Within
the CES project, three subprojects are carried out in the SEN4
research group.

1) Adaptive Agent Systems for Smart Energy Networks

We develop techniques and solutions in the areas of multi-agent
systems, computational intelligence, and sensor networks, for durable
energy systems.

Electricity networks have until now been used in a top-down manner.
The demand for energy can be forecasted reasonably well, and the
generation of energy is mainly prescheduled based on this, to be
supplied by large power generators. In the future, durable energy will
become more and more important. This is usually generated by many
small generators (at homes, farms, industry), which generate in a
decentralized fashion. Also, such durable energy generation typically
comes from  wind, solar, micro-CHP, biomass energy. So, this form of
energy generation is hard to predict.  In order to match energy supply
and demand, also the energy demand should be controlled. In the future
Smart Grids (smart electricity networks), decentralized control of
supply and demand is envisioned, in order to deal with the rising
amount of durably generated energy or to deal with new energy
intensive devices, such as electric cars. Important aspects are the
matching of energy supply and demand and the efficient usage of the
grid infrastructure. For these, market and planning mechanisms with
adaptive software agents need to be designed and developed. In
addition, smart planning systems should be developed at the consumer
side, which schedule the delayable energy consumption (of e.g.
boilers, freezers, electric cars, etc.), also based on the available
energy supply.

2) Exploring the role of rich context information in Smart Energy
Systems

We investigate how context information from smart physical (e.g.
homes, offices) and virtual (e.g. online presence) environments can be
harnessed to improve energy efficiency. The idea is that sensor
networks will find their way into homes and offices where they will
contribute to the creation of smart environments that are aware of the
occupants and their activities. In a parallel but related development,
an ever-increasing part of our social and professional interactions
are digitally documented and tracked online (e.g. agendas, time
schedules). The combination of these manifold sources of information
opens up the possibility of adjusting energy consumption (and
production) based on the urgency of current and planned activities, as
well as forecasts for energy availability and price. An important
aspect is to investigate how to link statistical techniques (needed to
extract patterns from numeric sensor data) to the symbolic and
agent-based learning methodologies that underpin complex interactive
systems.


3) Predictive Maintenance (or Management) in Smart Energy Networks

We focus on the role of sensors, knowledge aggregation, and network
simulations in monitoring and predicting the state of the electricity
distribution network in order to identify impending problems
(bottlenecks, downtime, etc). The rationale is that at rapidly growing
number of power distribution points, various sensors are being
installed in order to gauge the physical characteristics of the local
network. In some cases, the value of this sensor information is
obvious, e.g. an increase in the temperature of a transformer might be
indicative of overloading.  However, it is anticipated that the
usefulness of these sensor networks will extend far beyond such
straightforward inferences. The challenge therefore amounts to
extracting the proper features that will allow autonomous intelligent
agents to closely monitor the actual state of the network, and predict
its future behaviour. These predictions can then be used to optimize
network conditions in view of expected utilization, as well as
anticipate (and act upon) potential problems.

Members
3 PhD vacancies, Eric Pauwels, Han La Poutré

Partners
KEMA www.kema.com/nl