If you are interested in collaborating on challenging Optimization and Artificial Intelligence challenges, please, contact us
2022-25, CESI: Renewable Energy Sources
The activities of management and planning of resources on a national scale of an electrical network involve solving several linear and nonlinear optimization problems. One of these problems is the Optimal Power Flow (OPF) problem.
We aim to develop models and algorithms for the OPF problem in the stochastic setting that arises from using Renewable Energy Sources (RES). Integrating RES in a generation portfolio conveys several benefits, such as a reduction in greenhouse gas emissions and the country’s dependency on imported energy, and it decreases spot prices. However, the generation from RES can be variable and uncertain, in contrast to conventional generation.
Numerical methods will be validated using industrial benchmarks provided by CESI, the project's industrial partner.
2021-24, FEDEGARI: Interpretable Machine Learning
The aim of this research project is to study models and algorithms of Artificial Intelligence to optimize processes and services related to the machinery of industrial production sold by Fedegari Autoclavi SpA. Such models will support troubleshooting activities on machinery, enhancing their efficiency. Moreover, algorithms of Artificial Intelligence will improve current solutions of predictive maintenance, developing a system of condition-based monitoring and early warning.
2019-22, SEA VISION: Deep Learning Models
Together with SEA Vision and ARGO Vision, we have developed deep learning models for artificial vision in the field of Change Detection and surveillance.
Artificial Intelligence is increasingly integrating into industries today. A challenge that should not be underestimated is bringing models that represent the state of the art of scientific literature into industrial production. In fact, these models often need to consider the needs of the industrial world, such as deployment on edge devices with real-time inference.
For this reason, we have studied the main characteristics of Change Detection and Deep Learning. We have built models with performances comparable or superior to the state-of-the-art, with a saving in parameters ranging from 12 to 180 times and at least one-third of the computational complexity.
2017-19, COMDATA: Call Centers Optimization
The goal of this project was to design and implement an OPTIMIZATION SOFTWARE SYSTEM that includes (1) a constrained job scheduler, (2) an execution monitor algorithm, and (3) a real-time job re-scheduler.
The core scheduling problem, which must be solved before the cut-off time, consists of assigning resources to jobs in such a way that the compatibility (hard) constraints are satisfied, all the jobs are completed in time respecting the Service Level Agreement (SLA) constraint, and the overall assignment is balanced among all the used resources. The compatibility constraints are mainly due to the skill job requirements.
The dynamic reallocation problem is solved while reacting to plan delays and disruptions and minimizing the number of reallocations of jobs to resources: if a resource was already assigned to a job should not be reassigned to a different job.
Machine Learning-based simulator and optimization solver for predictive and prescriptive analytics (code in C++11)
2019-20, BINARY SYSTEMS: Crew and Vehicle Scheduling
In this project, we developed a mathematical model to solve an integrated crew scheduling and rostering problem for rail freight transportation designed to manage unplanned disruptions. First, we introduced the models which are solved to perform nominal planning operations, and second, we discussed how an integrated approach could be used to re-plan the activities in the face of disruptions.
We have implemented a Column Generation framework that includes all the operational constraints, and which can consider partial assignments of the solution which cannot be changed during the optimization process.
Crew scheduler developed in C++, with Python wrapper