The main aim of this research project consists of the modeling and synthesis of advanced control strategies of energy flows in new-generation power grids, known as Smart Grids. This project is in line with the guidelines of the European Green Deal, specifically in terms of energy consumption efficiency and low carbon impact. The main research topics and applications within this context are the intelligent control of energy flows, demand-side programs and scheduling through evolutionary fuzzy modeling techniques and artificial neural networks. The project involves also the topic of time series forecasting related to energy production from renewable sources and consumption from smart users.

Li-ion cells will always age, due to innate chemical reactions between the negative electrode and electrolyte. This will even happen while the vehicle is not being used, a phenomenon called 'calendar ageing'. The rate at which ageing occurs is a complex relationship between cell design and cell usage. The aim of this project consists of synthesizing innovative hybrid modeling techniques of high-performance Li-ion cells grounded on physical knowledge of electrochemical phenomena and machine learning. This specific data-driven methodology is thought in line with the Explainable AI paradigm, so it is able to give insight on the underlying battery aging process starting from acquired data.

In collaboration with FERRARI S.p.A.

Predictive maintenance is a widely adopted technique across sectors and during the last period its adoption is starting to spread across new markets such as the Energy Sector. Nowadays, predictive maintenance is commonly adopted in the energy sector to manage infrastructures, networks, and electrical grids. The reasons that are driving the adoption of predictive maintenance techniques by Transmission System Operators (TSO) and hardware manufacturers across the world are:

• The increasing availability of IoT devices connected to the network through cloud or fog technologies;

• The rising offer of predictive maintenance software in the market equipped with artificial intelligence that allows an easier and faster real-time data analysis for plants and devices’ maintenance;

• The accelerated transformation of the electricity market is moving the energy production from centralized models to distributed models.

The main aim of this project is to adopt Artificial Intelligence and data-driven techniques for predictive maintenance in cyber-physical systems such as, Smart Grids, railways train, industrial processes, etc.

Natural Language Processing (NLP) is a subfield of linguistics, computer science, and artificial intelligence concerned with the interactions between computers and human language, in particular how to program computers to process and analyze large amounts of natural language data. Today with the widespread usage of electronic text such as on the web or social media it is possible to gain powerful insight into text data through advanced machine learning and data-driven technique. The main aims of this project are twofold. From one and it is important to synthesize a new way of analyzing text from a scientific point of view conceiving human or machine-produced text as a complex system. On the other hand the great availability of electronic sources allows applications such as Social Network Analysis (e.g., hate speech detection, text summarization, etc.) through text mining and classification techniques.

Pattern recognition is the automated recognition of patterns and regularities in data. It has applications in statistical data analysis, signal processing, image analysis, information retrieval, bioinformatics, data compression, computer graphics and machine learning. An interesting research area of Pattern Recognition with interesting data-driven applications is the one related to non-conventional domains, that is objects (graphs, sequences, images, etc.) lie in unstructured often non-metric spaces where the similarity between patterns has to be computed through customized dissimilarity measures. For example, the starting objects' domain can be formed by graphs or networks and special dissimilarity measures need to be adopted to measure the dissimilarity between different size graphs. Moreover, it is really interesting investigating new representation methodologies to improve classification performances in machine learning-related applications.  

Evolutive Agent-Based Clustering Classifier (E-ABC), is an agent-based algorithm developed both for solving clustering and classification problems. In E-ABC each agent runs a very simple clustering procedure on a small sub-sample of the whole dataset. A genetic algorithm orchestrates the evolution of such agents in order to return a set of well-formed clusters, thus discovering possible regularities and recurrences in the data set at hand. Many clustering algorithms deal with a global metric. This means that the dissimilarity measure between patterns weights, in a suitable fashion, each feature leading to a global feature selection procedure. In the proposed approach weights are valid locally. The property of ``locality'' is ensured by a weighted dissimilarity measure alongside the not straightforward clustering procedure adopted and, practically, it is valid in the region around the cluster representatives. In other words, the evolutionary clustering process selects, through the agents, the subspace where these clusters are well-formed. Three main concepts are well addressed with E-ABC: i) the possibility of conceiving an evolutionary learning algorithm capable of dealing with local metrics, instead of global metrics, hence learning which features are important in characterizing input space regions, possibly belonging to different classes for the classification problem at hand.