Optimization with Reinforcement Learning: a case study with ENI

by Roberto Cornali (Istituto Nazionale di Fisica Nucleare)

Thursday 22 May 2025, 14:30 → 15:30 Europe/Rome

Sala Venturi

Many industrial processes are extremely complex, and our ability to
effectively model them is limited. Traditional optimization methods, such as
nonlinear programming or heuristic algorithms, often struggle to adapt to
real-time operational uncertainties. Reinforcement Learning (RL) can exploit
huge streams of data from modern IoT devices to optimize dynamically
industrial processes that we aren't capable to fully represent in a model. In
our case study, We train various Deep Reinforcement Learning on real data
provided by ENI to optimize a Natural Gas Liquefaction process. This work
highlights RL’s potential to bridge the gap between static-design optimization
and real-world dynamic optimization in LNG plants, offering a pathway toward
autonomous, adaptive liquefaction systems.

Link Teams: https://teams.microsoft.com/meet/3208744803117?p=sfhWTeiwsiBsgyg9o2