In the past decades, climate concern has been growing and electric energy has been identified as one of the key players in the energy transition. Meanwhile, the status of multilevel converters has evolved from ‘new’ to ‘established’ concept with industrial products in each power decade from 1W to 10MW. 

Choosing the best semiconductor has thus become more and more difficult because the choice is not guided by the voltage and current to handle: devices with a reduced voltage rating or with a reduced current rating can be respectively connected in series or parallel to build multilevel converters with improved performances: better efficiency, higher power density and/or reduced cost. 

To help designers make the right decision, specific methods can be used and design tools have been developed. This includes unified formalism for series-, parallel- and series-parallel- multilevel converters to allow analytic models for pre-design, unified simulation models for easy comparison of simulated waveforms, and modular approach of prototype construction. 

In this talk, design tools of different levels of complexity and accuracy that can be used to aid the designer at different stages of the design process will be presented. 

With the advancement of AI (Artificial Intelligence), DL (Deep Learning) methods have achieved outstanding performance and great success in a variety of real-world applications. Thanks to the versatility of DL, an increasing number of researchers are attracted to adopting it as a solution approach in diverse fields ranging from renewable energy production to bioinformatics. However, the performance of DL heavily depends on many factors, such as the learning model architecture and the hyper-parameters’ values. Moreover, the possible interactions between these different factors make the design of an effective DL model a complex task for the human practitioner. Recent research suggests EAs (Evolutionary Algorithms), such as GAs (Genetic Algorithms) and SI (Swarm Intelligence) algorithms, as effective candidate methods for the induction and tuning of highly effective DL models. This has been not only motivated but also justified by the ability of EAs to intelligently sample very large and complex search spaces of candidate DL models. Thanks to the prominent results observed and reported in various application domains, a new research field named EDL (Evolutionary Deep Learning) appeared in the machine learning area. The goal of this talk is to introduce the fundamentals of the EDL field by detailing its different components and merits. Eventually, several avenues for future research will be exposed, such as ETL (Evolutionary Transfer Learning) and HW-NAS (Hardware-Aware Neural Architecture Search).

Finally, this talk will be concluded with a set of promising applications of EDL in the green energy sector.

The production of electricity from photovoltaic (PV) energy is a fast-growing area of research in academia and industry. This trend has been boosted by the massive deployment of large-scale grid-connected photovoltaic power plants, which has stimulated research efforts regarding reliability and cost-effectiveness. However, photovoltaic systems are subject to various types of malfunction, whether temporary or permanent, and these malfunctions can have a significant impact on system performance and availability. In this context, meticulous monitoring and fault diagnosis of photovoltaic (PV) systems is essential to ensure the long-term reliability and sustainable operation of the entire PV system. Up to now, PV system fault detection and diagnosis can be classified into three main groups: process history based methods, quantitative model based methods and signal and image processing based methods. This talk will cover current practices and new trends in fault detection and diagnosis in grid connected PV systems, including the emergence of deep and machine learning approaches.