Computational Intelligence Techniques for Complex Systems with Applications to Smart Grids

The current century, just started, is living deep transformations never seen before within the humankind technological evolution. Somebody likes to call it the “digital era”. The deep changing has been possible thanks to the Information Communication Technologies revolution that has spread out pervasive and ubiquitous computing technologies with low-cost hardware, suitable network solutions and powerful Machine Learning tools. Nowadays, these tools are pushing for great changes to the actual power grid within the Smart Grid (SG) concept. Artificial Intelligence techniques in many of its facets, such as Machine Learning, Computational Intelligence, Approximate Reasoning, Evolutionary Computation, Pattern Recognition and so on, can play a decisive role for the technological challenges that the future has in store for one of the greatest technological masterpiece that mankind has ever made: the worldwide power grid. The current thesis is an effort in the direction of improving the present power grid importing methodologies and adopting solutions coming from the Machine Learning field. Within the SG concept, the power grid is seen as a decentralized structure where autonomous portions, known as micro-grids, plays an active role in producing and exchanging green energy with the remainder of the grid. In this context, it is presented an application of the Fuzzy-Genetic Algorithms paradigm to perform approximate reasoning and decision-making tasks within a micro-grid controller able to send commands for dispatching energy towards the micro-grid energy storage capacity or for selling to or to purchase from the main-grid available energy, given the market prices. Fuzzy Logic together with evolutionary computation techniques allows to “inject distributed intelligence” in the power grid. Moreover, Computational Intelligence and Pattern recognition are the perfect frameworks to provide the power grid with adaptation, sense-making and decisionmaking capabilities. The availability of Smart Sensors and powerful communication networks together with high-performance computational architectures makes possible to collect and analyze a wide amount of data related to the SG state and the surrounding environment. Within this context a novel Decision Support System (DSS) able to learn, recognize and characterize faults occurring in the real-world power grid of Rome, Italy, is presented. The DSS takes advantage of data-driven methods together with metric learning procedures to synthesize the faults model and to discriminate faults state from a standard functioning state of the SG in real-time. The methodology presented is based on clustering techniques that allow to constructing a white-box model of faults during the data-driven process. Thereby, after a learning phase performed by means of an evolutionary algorithm, the retrieved model can be analyzed with Data Science and Data visualization methods in order to increase the semantic content of collected raw data. In other words, a new Data Knowledge Discovery paradigm is adopted to confer high value to information that is further used in Condition-Based Maintenance (CBM) applications or more in general for building successful business strategies among the corporate decision-making procedures.