Integrating large-scale micro-generation in distribution grids is challenging for distribution grid operators, particularly when renewable energy sources (RES) and micro-cogeneration are involved. In this paper we contend that recent developments in multiagent decision making under uncertainty can positively contribute to safe, e cient and coste ective operation of future distribution grids.
Electric power systems have been undergoing momentous changes over the last
decade. In the past, power was supplied predominantly by a limited number
of large power plants, mainly nuclear powered or fossil fueled, and then
transmitted to the consumers [
The inclusion of micro-generation enhances the overall system in terms of sustainability. However, we advocate that additional improvement may be achieved through intelligent agent-based decision making. Customers, or agents acting on their behalf, should play an active role in managing the energy produced by the controllable micro-generation units. Additionally, the balance between consumption and supply is required for a proper and stable operation. Thus, agents that control energy consumption can also support the distribution network by matching the timing of their demand to the dynamic availability of the energy supply. As a result, a more e cient operation can be obtained by reducing the peak load while maintaining the power balance.
The massive introduction of small-scale RES-based production and active
consumption management introduces signi cant uncertainty in the normal
operation of the distribution grid. Some of the sources behind such uncertainty
are [
With the introduction of decentralized generation, several important changes regarding the planning, operation and control of power systems have taken place, particularly because of the following di erences: decentralized generation units are connected to the distribution network and not to the transmission network; several types of decentralized generation units are connected to the grid by means of power electronic interfaces (whereas large generation plants are coupled to the electricity grid directly); the power generated by micro-generation units is considerably less than power generated from traditional power plants (several orders of magnitude); and renewable energy generators depend on natural and uncontrollable sources, which adds a high level of uncertainty to the system. Given that distribution generation will play an important role at distribution level, power systems are forced to adapt in order to perform control actions at this operating level as well.
In the literature, a large number of the planning and control architectures
designed for distribution network and micro-grid applications have two-level
hierarchical con gurations and only take into account electrical parameters and
electrical interactions, even though heat outputs from micro-cogeneration units
are also available. For example, in [
In contrast to these prior works, we propose addressing the electrical ows,
but also the usable heat produced by the micro-generation units, as considered
in [
Hence, we envision a distribution grid that is able to self-regulate with little human supervision. We defend the use of new decentralized planning and control techniques for the distribution grid that take into account the dynamics and the topology of the grid and also handle the uncertainty inherent in the production and consumption of electricity. These techniques should allow the grid to preserve its properties as it scales in size and should also accommodate the possibility of massive micro-generation from renewable energy sources and from microcogeneration units. Finally, it should facilitate the inclusion of new technologies such as smart heating, ventilation, and air conditioning equipment. 2
Consider the distribution grid as a complex system composed of interconnected components, many of which need to be controlled in order to optimize system objectives. Decentralized planning accomplishes this optimization by distributing the control among a team of intelligent agents, each of which operates an individual component. For instance, an agent controlling a particular power substation decides where and when to route power. In planning its decisions, each agent should account for uncertainty in the consequences of its actions, reasoning over, for instance, the likelihoods of di erent volumes of future energy consumption. Agents may only be able to base their decisions on incomplete and local information, depending on sensory capabilities and on infrastructure supporting information exchange throughout the system. Nevertheless, because the actuation of one component may a ect the state of another, the agents should work together to formulate coordinated plans that ful ll quanti able global objectives.
In the literature, these characteristics serve as the basis for a formal model of
multiagent decision-making called a Decentralized Partially Observable Markov
Decision Process (Dec-POMDP) [
Framing the control problem as one of decentralized planning, one can address
the problems of keeping the network under stable operation, performing balance
control, and economically optimizing the system, in a single integrated solution,
all while accounting for uncertainty. An appealing aspect of this application is
the structure in the distribution grid control problem that we expect can be
leveraged to improve the e ciency and scalability of decentralized planning [
Although there is yet little application of MSDM techniques to real problems, control of smart distribution grids provides a well-motivated application domain, and with it a golden opportunity to break free of the status quo and to develop and validate MSDM research on realistic problems. It not only allows the testing of conventional assumptions of existing models and algorithms that have long been taken for granted, but it can also inspire the development of more useful models and methods whose assumptions are more realistic. This would constitute an important step forward in grounding recent MSDM work, and one that is essential for maturing the eld.
Acknowledgements This work was partially supported by national funds through Fundac~ao para a Ci^encia e a Tecnologia under project PEst-OE/EEI/LA0021/2011. M.S. is funded by the FP7 Marie Curie Actions Individual Fellowship #275217 (FP7-PEOPLE-2010- IEF).