Artificial Intelligence Techniques for the Smart Grid

In recent years, there is a rush in Artificial Intelligence (AI) research to produce practical solutions for the Smart Grid, the anticipated new generation of energy (primarily electricity) networks that will be able to make efficient use of renewable energy sources, support real time and efficient demand response, as well as the large-scale deployment of electric vehicles (EVs). AI techniques and methodologies can be instrumental in addressing sustainability problems, for example, to increase the efficiency and effectiveness of the way we manage and allocate our natural and societal resources. The drive to use AI for the Smart Grid has in turn led to novel questions and challenges for AI research, and to the realization that only the cross-fertilization of ideas and mixing of various techniques originating in different (sub-)fields can lead to the Holy Grail of an electricity Grid that takes full advantage of AI technologies to deliver power that is at the same time ‘green’, stable, affordable and accessible to all. This Special Issue brings together research questions and approaches originating in different (sub-)fields, such as multiagent systems, machine learning, optimization and statistics. As such, it provides an overview of a broad spectrum of ongoing Smart Grid research. To begin, going beyond traditional constant electricity pricing to more complex time of use (ToU) or real-time pricing methods to increase efficiency in energy use, poses serious technical and socio-political questions. With multiple pricing methods available, it is important for Grid regulators, utility companies and consumers to have the means to compare and even combine such policies. Against this background, Chrysopoulos and Mitkas (2017) introduce an optimization methodology for individualized ToU pricing policies. Using a multi-objective particle swarm optimization mechanism, the appropriate rates for each implemented pricing policy are identified, leading to higher consumer acceptance rates, along with costs and peak load reduction. At the same time, as readers of this journal know very well, a central question in Smart Grid research is modelling and analysing the energy consumption patterns of buildings. There are two articles in this Special Issue on the topic. On the one hand, Karatzas and Katsifarakis (2017) employ neural networks, decision trees and regression techniques to analyse and improve the electricity consumption of household appliances. On the other hand, Alam and colleagues (2017) provide a methodical development of thermal models using Extended Kalman Filters for parameter estimation, along with a case study using data from an actual family house. Moving from individual house settings, Akasiadis and Georgogiannis (2017) strive to improve the effectiveness of schemes promoting the large-scale, coordinated shifting of power consumption. In particular, they provide machine learning techniques – evaluated on real world datasets encompassing thousands of users – to monitor, assess and predict the trustworthiness of individual entities participating in such demand-side management activities. Then, Kofinas and colleagues (2017) propose the use of reinforcement learning (RL), one of the three main machine learning paradigms, to optimize the operation of a solar microgrid. To this end, they present an RL automated agent that employs a “fuzzy” reward function to learn a policy towards optimizing system performance. On a distinct issue of extremely high importance, the introduction of EVs in the Smart Grid – if accompanied with mature Vehicle-to-Grid (V2G) and Grid-to-Vehicle (G2V) technologies – is