Article Preview
Top1. Introduction
The smart power grid ought to involve smart consuming appliances and intelligent energy producers that strategically place bids in a short-term energy market. It is widely accepted that such smartness cannot be easily incorporated solely as hardware in the consuming and producing devices during the manufacturing time. Therefore, the recent research efforts that focus on software solutions to add intelligence to the smart grid are both scientifically interesting and technically well justified. Such task is surely not trivial as in the stochasticity traditional involved in the energy consumption, next generation power grids added stochasticity from the producers site through the integration of the distributed producers (renewable energy sources). This double stochasticity together with emerging market mechanisms and other more traditional issues make the task of adding intelligence to the grid in general a challenging issue and adding strategies on producing and consuming energy even more difficult.
We contribute to the efforts of smartening the smart grid by developing micro-learning (see section 2.2 for details) procedures that utilize weather data to train producing and/or consuming devices to strategically place bits. Our methodology, although simple, has the potential to be rather effective, in terms of the economic benefits for the energy market participants. Our learning algorithms reside in every device consuming or producing power and turn it in a strategic consumer and/or producer. This leads to what is known as distributed artificial intelligence in the smart grid.
Before we put any trust in our machine-learning procedures and before even testing them in a real environment, we must first study them in a (as realistic as possible) simulation environment. For this we have developed algorithms that lead to a machine learning layer, which we have integrated in a very comprehensive, detailed and widely used simulation power grid distribution network framework. This enables us to perform a series of experiments that involve very detailed energy producing and consuming configurations on a distribution network with more than 600 houses and several local energy renewable sources.
The rest of this paper can be divided in two parts. In the first we briefly give the basic concepts and issues on energy markets and learning methods that consist our background for the material presented in the sections 3 to 5 that will follow and describe our approach and present the experimental results. In the second part we study the use of game theory in energy grids. In section 6 we briefly present the basic concepts of game theory and in section 7 we give an overview of previous works that use game theory to implement strategic consumers and producers. In sections 8 and 9 we describe our approach in integrating game theory in our implementation and the experimental results. Our synopsis and future prospects can be found in Section 10.